Our ability to measure AI has been outpaced by our ability to develop it, and we believe this evaluation gap is one of the most important problems in AI. Open benchmarks are one of the most important levers for advancing AI safely and responsibly\u2014but the academic and open-source teams driving them often hit resource constraints, especially in the face of the exponentially expanding complexity of what tomorrow\u2019s benchmarks need to cover.
We think the next wave of benchmarks needs to push on three axes:\n- Environment complexity - How realistic is the operating environment?\n- Autonomy horizon - How far can an agent operate independently? We need to measure \n- Output complexity - How sophisticated is the work product?
Happy to answer questions about the grants, the framework, and would love to hear more about what you\u2019re building!"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Open Benchmarks Grants\u2013 a $3M commitment to close the AI eval gap"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://benchmarks.snorkel.ai/closing-the-evaluation-gap-in-agentic-ai/"}},"_tags":["story","author_vincentschen","story_46979278","show_hn"],"author":"vincentschen","created_at":"2026-02-11T19:06:18Z","created_at_i":1770836778,"num_comments":0,"objectID":"46979278","points":6,"story_id":46979278,"story_text":"Today, we're launching the Open Benchmarks Grants: a $3M commitment to fund open-source and academic teams building benchmarks for AI agents. In partnership with HuggingFace, PrimeIntellect, FactoryHQ, Together, Harbor, and PyTorch, the grants provide funding, data development support, and research collaboration.
Our ability to measure AI has been outpaced by our ability to develop it, and we believe this evaluation gap is one of the most important problems in AI. Open benchmarks are one of the most important levers for advancing AI safely and responsibly\u2014but the academic and open-source teams driving them often hit resource constraints, especially in the face of the exponentially expanding complexity of what tomorrow\u2019s benchmarks need to cover.
We think the next wave of benchmarks needs to push on three axes:\n- Environment complexity - How realistic is the operating environment?\n- Autonomy horizon - How far can an agent operate independently? We need to measure \n- Output complexity - How sophisticated is the work product?
Happy to answer questions about the grants, the framework, and would love to hear more about what you\u2019re building!","title":"Show HN: Open Benchmarks Grants\u2013 a $3M commitment to close the AI eval gap","updated_at":"2026-03-05T23:34:57Z","url":"https://benchmarks.snorkel.ai/closing-the-evaluation-gap-in-agentic-ai/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"tinymagician"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"DNA models have been chalked up to be pretty useless but I think that's not true. I sketch out reasons why I think they are useful. Feedback and comments welcome. High-level takeaways:
1) a Prime Intellect like RL environment for biology can unlock a lot of value for DNA models, but first needs to be built\n2) like most deep learning models, data is king. Curating high quality low sample size task specific data to fine tune a general dna model is the clearest path to realizing value here\n3) Specific applications I find interesting are CAR T promoter design, AAV gene therapy regulatory element design, reassigning variants of uncertain significance, bridging the variant to function gap, and biosecurity (details for each one in the post)\n4) This is an early technology and the potential exists for two main reasons: a) we have a lot more DNA sequencing data than protein data, b) DNA contains more information than proteins. So if protein models like AlphaFold and ESM can unlock new medicines and discoveries in biology, the possibility for DNA models are orders of magnitude higher"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Applications of DNA Foundation Models"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.aditharun.com/p/dna-foundation-models"}},"_tags":["story","author_tinymagician","story_45140225","show_hn"],"author":"tinymagician","created_at":"2025-09-05T16:13:14Z","created_at_i":1757088794,"num_comments":0,"objectID":"45140225","points":3,"story_id":45140225,"story_text":"DNA models have been chalked up to be pretty useless but I think that's not true. I sketch out reasons why I think they are useful. Feedback and comments welcome. High-level takeaways:
1) a Prime Intellect like RL environment for biology can unlock a lot of value for DNA models, but first needs to be built\n2) like most deep learning models, data is king. Curating high quality low sample size task specific data to fine tune a general dna model is the clearest path to realizing value here\n3) Specific applications I find interesting are CAR T promoter design, AAV gene therapy regulatory element design, reassigning variants of uncertain significance, bridging the variant to function gap, and biosecurity (details for each one in the post)\n4) This is an early technology and the potential exists for two main reasons: a) we have a lot more DNA sequencing data than protein data, b) DNA contains more information than proteins. So if protein models like AlphaFold and ESM can unlock new medicines and discoveries in biology, the possibility for DNA models are orders of magnitude higher","title":"Show HN: Applications of DNA Foundation Models","updated_at":"2026-03-05T22:39:23Z","url":"https://www.aditharun.com/p/dna-foundation-models"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Philpax"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["intellect"],"value":"Intellect-2 Release: The First 32B Model Trained Through Globally Distributed RL"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/intellect-2-release"}},"_tags":["story","author_Philpax","story_43958898"],"author":"Philpax","children":[43959166,43959182,43959188,43959246,43959254,43959261,43959266,43959432,43959453,43959759,43959771,43959788,43960661,43961760,43963805,43964179,43964406],"created_at":"2025-05-12T01:46:57Z","created_at_i":1747014417,"num_comments":65,"objectID":"43958898","points":201,"story_id":43958898,"title":"Intellect-2 Release: The First 32B Model Trained Through Globally Distributed RL","updated_at":"2026-02-07T12:43:38Z","url":"https://www.primeintellect.ai/blog/intellect-2-release"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jasondavies"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["intellect"],"value":"INTELLECT\u20131: Launching the First Decentralized Training of a 10B Parameter Model"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/intellect-1"}},"_tags":["story","author_jasondavies","story_41812562"],"author":"jasondavies","children":[41814380,41815871,41816103,41816137,41816626,41817150,41817170,41817524,41817744,41818076,41818288,41818299,41818390,41818680,41819209,41819403],"created_at":"2024-10-11T19:14:19Z","created_at_i":1728674059,"num_comments":36,"objectID":"41812562","points":111,"story_id":41812562,"title":"INTELLECT\u20131: Launching the First Decentralized Training of a 10B Parameter Model","updated_at":"2025-08-14T22:30:58Z","url":"https://www.primeintellect.ai/blog/intellect-1"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"algoman"},"title":{"matchLevel":"none","matchedWords":[],"value":"The First Globally Trained 10B Parameter Model"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/intellect-1-release"}},"_tags":["story","author_algoman","story_42277713"],"author":"algoman","children":[42277820,42278020],"created_at":"2024-11-29T22:32:17Z","created_at_i":1732919537,"num_comments":2,"objectID":"42277713","points":8,"story_id":42277713,"title":"The First Globally Trained 10B Parameter Model","updated_at":"2024-12-02T22:31:10Z","url":"https://www.primeintellect.ai/blog/intellect-1-release"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"skhameneh"},"title":{"matchLevel":"none","matchedWords":[],"value":"Recursive Language Models: the paradigm of 2026"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/rlm"}},"_tags":["story","author_skhameneh","story_46458385"],"author":"skhameneh","children":[46460231],"created_at":"2026-01-01T21:45:34Z","created_at_i":1767303934,"num_comments":1,"objectID":"46458385","points":5,"story_id":46458385,"title":"Recursive Language Models: the paradigm of 2026","updated_at":"2026-03-05T23:17:32Z","url":"https://www.primeintellect.ai/blog/rlm"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"vincentweisser"},"title":{"matchLevel":"none","matchedWords":[],"value":"Peer-to-Peer Decentralized Inference"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/inference"}},"_tags":["story","author_vincentweisser","story_43827796"],"author":"vincentweisser","children":[43827797],"created_at":"2025-04-29T01:11:43Z","created_at_i":1745889103,"num_comments":1,"objectID":"43827796","points":4,"story_id":43827796,"title":"Peer-to-Peer Decentralized Inference","updated_at":"2025-04-29T16:16:10Z","url":"https://www.primeintellect.ai/blog/inference"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"anacleto"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["intellect"],"value":"Intellect-3: A 100B+ MoE trained with large-scale RL"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/intellect-3"}},"_tags":["story","author_anacleto","story_46074863"],"author":"anacleto","children":[46075127],"created_at":"2025-11-28T01:53:19Z","created_at_i":1764294799,"num_comments":1,"objectID":"46074863","points":3,"story_id":46074863,"title":"Intellect-3: A 100B+ MoE trained with large-scale RL","updated_at":"2026-03-05T23:09:19Z","url":"https://www.primeintellect.ai/blog/intellect-3"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"nateb2022"},"title":{"matchLevel":"none","matchedWords":[],"value":"Lab: The Full-Stack Platform for Training Your Own Models"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/lab"}},"_tags":["story","author_nateb2022","story_48845777"],"author":"nateb2022","created_at":"2026-07-09T13:47:20Z","created_at_i":1783604840,"num_comments":0,"objectID":"48845777","points":3,"story_id":48845777,"title":"Lab: The Full-Stack Platform for Training Your Own Models","updated_at":"2026-07-14T21:27:49Z","url":"https://www.primeintellect.ai/blog/lab"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"ronfriedhaber"},"title":{"matchLevel":"none","matchedWords":[],"value":"$130M Series A to Build the Open Superintelligence Stack"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/series-a"}},"_tags":["story","author_ronfriedhaber","story_48842314"],"author":"ronfriedhaber","created_at":"2026-07-09T07:48:34Z","created_at_i":1783583314,"num_comments":0,"objectID":"48842314","points":3,"story_id":48842314,"title":"$130M Series A to Build the Open Superintelligence Stack","updated_at":"2026-07-09T16:18:08Z","url":"https://www.primeintellect.ai/blog/series-a"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"thomasm6m6"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["prime"],"value":"Systematic Reward Hacking and Prime Sprints"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/reward-hacking"}},"_tags":["story","author_thomasm6m6","story_48219165"],"author":"thomasm6m6","created_at":"2026-05-21T07:40:15Z","created_at_i":1779349215,"num_comments":0,"objectID":"48219165","points":3,"story_id":48219165,"title":"Systematic Reward Hacking and Prime Sprints","updated_at":"2026-05-21T11:07:35Z","url":"https://www.primeintellect.ai/blog/reward-hacking"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"dominik-space"},"title":{"matchLevel":"none","matchedWords":[],"value":"Lab: The Full-Stack Platform for Training Your Own Models"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/lab"}},"_tags":["story","author_dominik-space","story_46971767"],"author":"dominik-space","created_at":"2026-02-11T06:57:52Z","created_at_i":1770793072,"num_comments":0,"objectID":"46971767","points":3,"story_id":46971767,"title":"Lab: The Full-Stack Platform for Training Your Own Models","updated_at":"2026-03-05T23:34:32Z","url":"https://www.primeintellect.ai/blog/lab"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"hexhowells"},"title":{"matchLevel":"none","matchedWords":[],"value":"Environments Hub: A Community Hub to Scale RL to Open AGI"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/environments"}},"_tags":["story","author_hexhowells","story_45049312"],"author":"hexhowells","created_at":"2025-08-28T07:06:06Z","created_at_i":1756364766,"num_comments":0,"objectID":"45049312","points":3,"story_id":45049312,"title":"Environments Hub: A Community Hub to Scale RL to Open AGI","updated_at":"2026-03-05T22:32:54Z","url":"https://www.primeintellect.ai/blog/environments"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"emersonmacro"},"title":{"matchLevel":"none","matchedWords":[],"value":"Recursive Language Models: the paradigm of 2026"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/rlm"}},"_tags":["story","author_emersonmacro","story_46485905"],"author":"emersonmacro","children":[46486097],"created_at":"2026-01-04T07:52:09Z","created_at_i":1767513129,"num_comments":1,"objectID":"46485905","points":2,"story_id":46485905,"title":"Recursive Language Models: the paradigm of 2026","updated_at":"2026-03-05T23:19:07Z","url":"https://www.primeintellect.ai/blog/rlm"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"meetpateltech"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["intellect"],"value":"Intellect-3: A 100B+ MoE trained with large-scale RL"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/intellect-3"}},"_tags":["story","author_meetpateltech","story_46066466"],"author":"meetpateltech","children":[46067125],"created_at":"2025-11-27T06:51:13Z","created_at_i":1764226273,"num_comments":1,"objectID":"46066466","points":2,"story_id":46066466,"title":"Intellect-3: A 100B+ MoE trained with large-scale RL","updated_at":"2026-03-05T23:08:45Z","url":"https://www.primeintellect.ai/blog/intellect-3"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"dominik-space"},"title":{"matchLevel":"none","matchedWords":[],"value":"Verifiers: Environments for LLM Reinforcement Learning"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://github.com/PrimeIntellect-ai/verifiers"}},"_tags":["story","author_dominik-space","story_45382483"],"author":"dominik-space","created_at":"2025-09-26T04:00:01Z","created_at_i":1758859201,"num_comments":0,"objectID":"45382483","points":2,"story_id":45382483,"title":"Verifiers: Environments for LLM Reinforcement Learning","updated_at":"2026-03-05T22:44:44Z","url":"https://github.com/PrimeIntellect-ai/verifiers"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"anacleto"},"title":{"matchLevel":"none","matchedWords":[],"value":"PCCL: Low-level, fault-tolerant communication library for decentralized training"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/pccl"}},"_tags":["story","author_anacleto","story_44073312"],"author":"anacleto","created_at":"2025-05-23T14:37:18Z","created_at_i":1748011038,"num_comments":0,"objectID":"44073312","points":2,"story_id":44073312,"title":"PCCL: Low-level, fault-tolerant communication library for decentralized training","updated_at":"2025-05-23T20:49:20Z","url":"https://www.primeintellect.ai/blog/pccl"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"anacleto"},"title":{"matchLevel":"none","matchedWords":[],"value":"Decentralized Training in the Inference-Time-Compute Paradigm"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/intellect-math"}},"_tags":["story","author_anacleto","story_42788074"],"author":"anacleto","created_at":"2025-01-22T02:17:13Z","created_at_i":1737512233,"num_comments":0,"objectID":"42788074","points":2,"story_id":42788074,"title":"Decentralized Training in the Inference-Time-Compute Paradigm","updated_at":"2025-01-22T08:35:54Z","url":"https://www.primeintellect.ai/blog/intellect-math"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"SchwKatze"},"title":{"matchLevel":"none","matchedWords":[],"value":"INTELLLECT_1 Technical Report [pdf]"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://github.com/PrimeIntellect-ai/prime/blob/main/INTELLECT_1_Technical_Report.pdf"}},"_tags":["story","author_SchwKatze","story_42278242"],"author":"SchwKatze","created_at":"2024-11-29T23:59:25Z","created_at_i":1732924765,"num_comments":0,"objectID":"42278242","points":2,"story_id":42278242,"title":"INTELLLECT_1 Technical Report [pdf]","updated_at":"2024-11-30T02:43:45Z","url":"https://github.com/PrimeIntellect-ai/prime/blob/main/INTELLECT_1_Technical_Report.pdf"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"liamdgray"},"title":{"matchLevel":"none","matchedWords":[],"value":"Synthetic-1: Scaling Distributed Synthetic Data Generation for VerifiedReasoning"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/synthetic-1"}},"_tags":["story","author_liamdgray","story_43028943"],"author":"liamdgray","children":[43028944],"created_at":"2025-02-12T19:39:51Z","created_at_i":1739389191,"num_comments":1,"objectID":"43028943","points":1,"story_id":43028943,"title":"Synthetic-1: Scaling Distributed Synthetic Data Generation for VerifiedReasoning","updated_at":"2025-02-12T19:45:54Z","url":"https://www.primeintellect.ai/blog/synthetic-1"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"pseudolus"},"title":{"matchLevel":"none","matchedWords":[],"value":"Recursive Language Models: the paradigm of 2026"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/rlm"}},"_tags":["story","author_pseudolus","story_46760660"],"author":"pseudolus","created_at":"2026-01-26T01:28:09Z","created_at_i":1769390889,"num_comments":0,"objectID":"46760660","points":1,"story_id":46760660,"title":"Recursive Language Models: the paradigm of 2026","updated_at":"2026-03-05T23:25:40Z","url":"https://www.primeintellect.ai/blog/rlm"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"noch"},"title":{"matchLevel":"none","matchedWords":[],"value":"Toploc: A Locality Sensitive Hashing Scheme for Trustless Verifiable Inference"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/toploc"}},"_tags":["story","author_noch","story_42862476"],"author":"noch","created_at":"2025-01-29T07:22:14Z","created_at_i":1738135334,"num_comments":0,"objectID":"42862476","points":1,"story_id":42862476,"title":"Toploc: A Locality Sensitive Hashing Scheme for Trustless Verifiable Inference","updated_at":"2025-01-29T07:26:02Z","url":"https://www.primeintellect.ai/blog/toploc"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jasondavies"},"title":{"matchLevel":"none","matchedWords":[],"value":"OpenDiLoCo, globally distributed low-communication AI model training"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"https://www.primeintellect.ai/blog/opendiloco"}},"_tags":["story","author_jasondavies","story_40943858"],"author":"jasondavies","created_at":"2024-07-12T08:58:56Z","created_at_i":1720774736,"num_comments":0,"objectID":"40943858","points":1,"story_id":40943858,"title":"OpenDiLoCo, globally distributed low-communication AI model training","updated_at":"2024-09-20T17:26:09Z","url":"https://www.primeintellect.ai/blog/opendiloco"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"lapatride"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"This post introduces GEO\u2011Prime ELITE, a modular AI infrastructure designed to deliver citation-ready content across generative search engines (SGE, GPT\u2011Browse, Perplexity, etc.), using classic source texts.
Purpose:\nGEO\u2011Prime ELITE is a modular pipeline designed to generate citation-ready content for generative AI engines (SGE, GPT\u2011Browse, Perplexity, etc.) from classical source texts.
Pipeline Components:\n-RAG (Retrieval-Augmented Generation) with local vector database\n-Logical fallback: hierarchy GPT\u20114o > Claude > local model
Post-processing:\n-segmentation \u2264 80 words\n-citation-readiness scoring\n-JSON\u2011LD formatting\n-LoRA fine-tuning based on AI citation scoring\n-Cost governance (CAPEX/OPEX) + escalation model toward 70B-scale
Business Goals:\n-Reducing no-click SEO\n-Increasing citation visibility in AI engines\n-Optimizing content for generative search\n-Monetization via licensing, MVP rollout, or strategic sale
Content:\n-pitch-deck/: PDF presentation of the pipeline\n-architecture/: modular system structure (coming soon)\n-scoring/: citation-readiness criteria (demo or simulated JSON)\n-uplift-study/: upcoming evaluation protocol
Intellectual Property:\nThis repository constitutes a public proof of prior art for the concept, architecture, and logic behind GEO\u2011Prime ELITE.
Original author: Fr\u00e9d\u00e9ric Cl\u00e9ment-Tribouilloy (Lapatride)\nInitial publication date: July 10, 2025
View on GitHub:\n https://github.com/Lapatride/geo-prime-ELITE/blob/main/README%20initial%20GEO-Prime%20ELITE.md"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["prime"],"value":"Geo\u2011Prime ELITE: Modular AI pipeline for citation-focused content generation"}},"_tags":["story","author_lapatride","story_44567744","ask_hn"],"author":"lapatride","created_at":"2025-07-15T04:08:55Z","created_at_i":1752552535,"num_comments":0,"objectID":"44567744","points":2,"story_id":44567744,"story_text":"This post introduces GEO\u2011Prime ELITE, a modular AI infrastructure designed to deliver citation-ready content across generative search engines (SGE, GPT\u2011Browse, Perplexity, etc.), using classic source texts.
Purpose:\nGEO\u2011Prime ELITE is a modular pipeline designed to generate citation-ready content for generative AI engines (SGE, GPT\u2011Browse, Perplexity, etc.) from classical source texts.
Pipeline Components:\n-RAG (Retrieval-Augmented Generation) with local vector database\n-Logical fallback: hierarchy GPT\u20114o > Claude > local model
Post-processing:\n-segmentation \u2264 80 words\n-citation-readiness scoring\n-JSON\u2011LD formatting\n-LoRA fine-tuning based on AI citation scoring\n-Cost governance (CAPEX/OPEX) + escalation model toward 70B-scale
Business Goals:\n-Reducing no-click SEO\n-Increasing citation visibility in AI engines\n-Optimizing content for generative search\n-Monetization via licensing, MVP rollout, or strategic sale
Content:\n-pitch-deck/: PDF presentation of the pipeline\n-architecture/: modular system structure (coming soon)\n-scoring/: citation-readiness criteria (demo or simulated JSON)\n-uplift-study/: upcoming evaluation protocol
Intellectual Property:\nThis repository constitutes a public proof of prior art for the concept, architecture, and logic behind GEO\u2011Prime ELITE.
Original author: Fr\u00e9d\u00e9ric Cl\u00e9ment-Tribouilloy (Lapatride)\nInitial publication date: July 10, 2025
View on GitHub:\n https://github.com/Lapatride/geo-prime-ELITE/blob/main/README%20initial%20GEO-Prime%20ELITE.md","title":"Geo\u2011Prime ELITE: Modular AI pipeline for citation-focused content generation","updated_at":"2025-07-15T05:59:54Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"foxbarrington"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"This started as a reaction to a conversational trope. Despite being a tranquil place, even conversations at my yoga studio often start with, "Can you believe what's going on right now?" with that angry/scared undertone.
I'm a news avoider, so I usually feel some smug self-satisfaction in those instances, but I wondered if there was a way to satisfy the urge to doomscroll without the anxiety.
My hypothesis: Apply a 40-year latency buffer. You get the intellectual stimulation of "Big Events" without the fog of war, because you know the world didn't end.
40 years creates a mirror between the Reagan Era and today. The parallels include celebrity populism, Cold War tensions (Soviets vs. Russia), and inflation economics.
The system ingests raw newspaper scans and uses a multi-step LLM pipeline to generate the daily edition:
OCR & Ingestion: Converts raw pixels to text.
Scoring: Grades events on metrics like Dramatic Irony and Name Recognition to surface stories that are interesting with hindsight. For example, a dry business blurb about Steve Jobs leaving Apple scores highly because the future context creates a narrative arc.
Objective Fact Extraction: Extracts a list of discrete, verifiable facts from the raw text.
Generation: Uses those extracted facts as the ground truth to write new headlines and story summaries.
I expected a zen experience. Instead, I got an entertaining docudrama. Historical events are surprisingly compelling when serialized over weeks.
For example, on Oct 7, 1985, Palestinian hijackers took over the cruise ship Achille Lauro. Reading this on a delay in 2025, the story unfolded over weeks: first they threw an American in a wheelchair overboard, then US fighter jets forced the escape plane to land, leading to a military standoff between US Navy SEALs and the Italian Air Force. Unbelievably, the US backed down, but the later diplomatic fallout led the Italian Prime Minister to resign.
It hits the dopamine receptors of the news cycle, but with the comfort of a known outcome.
Stack: React, Node.js (Caskada for the LLM pipeline orchestration), Gemini for OCR/Scoring.
Link: https://forty.news (No signup required, it's only if you want the stories emailed to you daily/weekly)"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Forty.News \u2013 Daily news, but on a 40-year delay"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://forty.news"}},"_tags":["story","author_foxbarrington","story_46017175","show_hn"],"author":"foxbarrington","children":[46018047,46018048,46018074,46018136,46018200,46018203,46018225,46018239,46018244,46018245,46018259,46018276,46018291,46018300,46018323,46018326,46018376,46018409,46018412,46018496,46018552,46018654,46018833,46018854,46018863,46018904,46018943,46018983,46019204,46019397,46019487,46019552,46019578,46019655,46019664,46019697,46019763,46019937,46019984,46020216,46020735,46020764,46020909,46021233,46021263,46021415,46021590,46021709,46021755,46021857,46022071,46022237,46022596,46022698,46022717,46022995,46023107,46023141,46024001,46024736,46024750,46026263,46027067,46027285,46029435,46031730,46034751,46050251,46064089],"created_at":"2025-11-22T18:47:08Z","created_at_i":1763837228,"num_comments":180,"objectID":"46017175","points":443,"story_id":46017175,"story_text":"This started as a reaction to a conversational trope. Despite being a tranquil place, even conversations at my yoga studio often start with, "Can you believe what's going on right now?" with that angry/scared undertone.
I'm a news avoider, so I usually feel some smug self-satisfaction in those instances, but I wondered if there was a way to satisfy the urge to doomscroll without the anxiety.
My hypothesis: Apply a 40-year latency buffer. You get the intellectual stimulation of "Big Events" without the fog of war, because you know the world didn't end.
40 years creates a mirror between the Reagan Era and today. The parallels include celebrity populism, Cold War tensions (Soviets vs. Russia), and inflation economics.
The system ingests raw newspaper scans and uses a multi-step LLM pipeline to generate the daily edition:
OCR & Ingestion: Converts raw pixels to text.
Scoring: Grades events on metrics like Dramatic Irony and Name Recognition to surface stories that are interesting with hindsight. For example, a dry business blurb about Steve Jobs leaving Apple scores highly because the future context creates a narrative arc.
Objective Fact Extraction: Extracts a list of discrete, verifiable facts from the raw text.
Generation: Uses those extracted facts as the ground truth to write new headlines and story summaries.
I expected a zen experience. Instead, I got an entertaining docudrama. Historical events are surprisingly compelling when serialized over weeks.
For example, on Oct 7, 1985, Palestinian hijackers took over the cruise ship Achille Lauro. Reading this on a delay in 2025, the story unfolded over weeks: first they threw an American in a wheelchair overboard, then US fighter jets forced the escape plane to land, leading to a military standoff between US Navy SEALs and the Italian Air Force. Unbelievably, the US backed down, but the later diplomatic fallout led the Italian Prime Minister to resign.
It hits the dopamine receptors of the news cycle, but with the comfort of a known outcome.
Stack: React, Node.js (Caskada for the LLM pipeline orchestration), Gemini for OCR/Scoring.
Link: https://forty.news (No signup required, it's only if you want the stories emailed to you daily/weekly)","title":"Show HN: Forty.News \u2013 Daily news, but on a 40-year delay","updated_at":"2026-03-05T23:05:21Z","url":"https://forty.news"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"XiphiasX"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"Thinking of best long term investments for the next decades, I came to the following conclusions.
With the advancements in AI technology in the next 10-20 years, the cost of intelligence, manpower, and construction is expected to decrease significantly. As a result, prime real estate ownership, farmland, and locations with precious resources will likely become valuable investments.*
Reasoning:
- The emergence of AGI (Artificial General Intelligence) will lead to rapid developments in other technologies, such as fusion power, which will greatly decrease energy costs.
- With companies like Boston Dynamics making strides in creating realistic robots with superior intelligence, we can expect to see a decrease in the need for human labor and access to intellect.
- This will lead to a reduction in the costs associated with construction, manpower, and other creative and intelligent pursuits.
- Despite the decrease in costs, land will still be limited, particularly in prime real estate locations. While construction may be cheap, the scarcity of land in desirable locations will make these properties highly sought after.
- The increasing availability of open-source AI development will lead to increased competition, making AI technology more accessible to individuals and small businesses.
- Even with advancements in agriculture and food production, precious resources will continue to be highly valued, particularly those used in the production of computer chips.
Thoughts? Comments? Criticisms?"},"title":{"matchLevel":"none","matchedWords":[],"value":"Why real estate is the only investment that matters in AI dominated future"}},"_tags":["story","author_XiphiasX","story_34452168","ask_hn"],"author":"XiphiasX","children":[34452365],"created_at":"2023-01-20T12:28:39Z","created_at_i":1674217719,"num_comments":0,"objectID":"34452168","points":1,"story_id":34452168,"story_text":"Thinking of best long term investments for the next decades, I came to the following conclusions.
With the advancements in AI technology in the next 10-20 years, the cost of intelligence, manpower, and construction is expected to decrease significantly. As a result, prime real estate ownership, farmland, and locations with precious resources will likely become valuable investments.*
Reasoning:
- The emergence of AGI (Artificial General Intelligence) will lead to rapid developments in other technologies, such as fusion power, which will greatly decrease energy costs.
- With companies like Boston Dynamics making strides in creating realistic robots with superior intelligence, we can expect to see a decrease in the need for human labor and access to intellect.
- This will lead to a reduction in the costs associated with construction, manpower, and other creative and intelligent pursuits.
- Despite the decrease in costs, land will still be limited, particularly in prime real estate locations. While construction may be cheap, the scarcity of land in desirable locations will make these properties highly sought after.
- The increasing availability of open-source AI development will lead to increased competition, making AI technology more accessible to individuals and small businesses.
- Even with advancements in agriculture and food production, precious resources will continue to be highly valued, particularly those used in the production of computer chips.
Thoughts? Comments? Criticisms?","title":"Why real estate is the only investment that matters in AI dominated future","updated_at":"2024-09-20T13:03:45Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"BTBurke"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"I recently landed my first freelance job based off a little side project that I had on Github. I'm not a professional developer, I just do this as a hobby in my free time. It's serendipitous that I have a few months off between jobs and I'd like to make sure I don't make a rookie mistake in pricing this job and setting up the contract.
The work is machine learning algorithm development and then integrating the solution into the client's existing infrastructure.
A few questions:
* What are the pros/cons of an hourly rate vs. fixed price?
* Anything to watch out for in how the contract is structured? e.g., provisions about intellectual property, dealing with changes in requirements, or spelling out what deliverables are required?
* What kind of padding factor do you apply to your estimates of hours to complete a feature?
Based on some other things that I've read, I'm leaning towards pitching an hourly rate with an upper limit. For example, quoting a high-end price of $X and charging by the hour, giving them the savings if I can implement it in fewer hours or (hopefully not) having to renegotiate if it's significantly more complicated than I envision.
Any advice would be appreciated. If someone has a boilerplate contract that has worked well for them in the past, I'd love to see it. Thanks HN!"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["prime"],"value":"Ask HN: Landed my first freelance job, how should I price the project?"},"url":{"matchLevel":"none","matchedWords":[],"value":""}},"_tags":["story","author_BTBurke","story_6505170","ask_hn"],"author":"BTBurke","children":[6505347,6505440,6506218,6511727],"created_at":"2013-10-06T19:41:59Z","created_at_i":1381088519,"num_comments":8,"objectID":"6505170","points":11,"story_id":6505170,"story_text":"I recently landed my first freelance job based off a little side project that I had on Github. I'm not a professional developer, I just do this as a hobby in my free time. It's serendipitous that I have a few months off between jobs and I'd like to make sure I don't make a rookie mistake in pricing this job and setting up the contract.
The work is machine learning algorithm development and then integrating the solution into the client's existing infrastructure.
A few questions:
* What are the pros/cons of an hourly rate vs. fixed price?
* Anything to watch out for in how the contract is structured? e.g., provisions about intellectual property, dealing with changes in requirements, or spelling out what deliverables are required?
* What kind of padding factor do you apply to your estimates of hours to complete a feature?
Based on some other things that I've read, I'm leaning towards pitching an hourly rate with an upper limit. For example, quoting a high-end price of $X and charging by the hour, giving them the savings if I can implement it in fewer hours or (hopefully not) having to renegotiate if it's significantly more complicated than I envision.
Any advice would be appreciated. If someone has a boilerplate contract that has worked well for them in the past, I'd love to see it. Thanks HN!","title":"Ask HN: Landed my first freelance job, how should I price the project?","updated_at":"2024-09-19T19:58:33Z","url":""},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"octosphere"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["prime"],"value":"Huawei Nova 5i Price in India, Specifications, and Features"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["intellect"],"value":"https://www.intellectdigest.in/huawei-nova-5i-42182/"}},"_tags":["story","author_octosphere","story_20086860"],"author":"octosphere","created_at":"2019-06-03T18:57:43Z","created_at_i":1559588263,"num_comments":0,"objectID":"20086860","points":1,"story_id":20086860,"title":"Huawei Nova 5i Price in India, Specifications, and Features","updated_at":"2024-09-20T04:20:46Z","url":"https://www.intellectdigest.in/huawei-nova-5i-42182/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jacquesm"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"There are a bunch of steps to dealing with VCs that you should\nbe familiar with before stepping in to the fray.
My background in this is that I did several pitches to VCs and have worked\nfor several VCs in the last couple of years, mostly on the technical\nside but you get to see a good part of the process as perceived by\nthe companies pitching. I also helped one company get seed\ncapital, because I thought they showed great promise.
So, the following is from personal experience,\nwhich is limited but it may be useful.
Some VC's are more forgiving than others when it comes to following\nproper form, but even if they're forgiving they'll appreciate it if you\nknow your stuff before you apply. Their agendas are almost\nalways quite full and the amount of time they have to spend on you\nis limited, so you should use it wisely to maximize your chances of\nsuccess.
== before you start pitching ==
- First off, let's dispel a popular mistaken belief, getting an investment\n is not 'success'. It's a step towards a possible success, and it may\n give you a better chance, but it basically comes down to another\n party estimating that you have a chance of success and that they\n want a piece of that success in return for an investment.
- You have to know your stuff. This of course, sounds completely\n obvious but it really is surprising how many people will pitch\n to a VC (in itself something of an opportunity) and completely\n blow it by not having done basic homework. A fairly recent\n example, someone wrote a businessplan around a certain\n type of person and was pitching for capital. Halfway through\n the presentation one of the partners of the company being\n pitched interrupted and asked 'And how many of these people\n do you know ?'. A sensible question. The answer '0', more or\n less ended the interview, even though it went on for a while\n beyond that point, for politeness' sake.
Not knowing your target market, not having done any basic\n research in to the demographic that you intend to sell your\n product to is lethal.
As is not knowing your competive arena. If you come up with a\n brilliant plan that looks like some successful competitor is already executing and\n you don't know they exist that's probably the end of the ride for you. You\n really need to spend solid time on mapping out the competition. Know their weak points, know your strong points.
In short, know your stuff, expect to be challenged.
- Make sure you inform your partners about all your moves and get\n them on board before approaching a new party. To find out that\n someone isn't on board after you approach a VC is a real problem.
If there is any problem between the founders get it ironed out\n before you start making pitches, and make sure problems\n are resolved to all parties satisfaction.
- It isn't a must to be incorporated before approaching VCs, but it\n can be a problem if you picked the wrong form.
- if there are pending lawsuits it is usually a good idea to get\n those resolved before pitching as well. This can royally screw\n up your timing, a window of opportunity can easily close while\n you attend to this.
- you have to know what it is that you want that investment for,\n no matter how sure you are that you'll be able to spend it\n wisely, you need to pretty much lay out how you intend to\n use the money an investment will bring. This is where a\n businessplan comes in. These are no longer the 90's, so please\n no columns with more than 9 0's in it. Keep it realistic and\n make sure that it contains realistic estimates for the costs\n of the various components of your business. Factor in\n market rates for salaries, office space and so on. Get a feel\n for what it costs a business of a similar size to operate.
- scouting for capital takes time. Sometimes LOTS of time. Make\n sure you have that time, and make sure that your business\n does not suffer from this. Farm out as much of the work to your\n co-founders as you can, spread that load. It will give them more\n insight in the process and it will get them more involved.
== approaching target VCs ==
- when approaching a VC try to find out the names of their 'spotters',\n and pitch the spotters first. That way, you essentially get two\n chances, the spotters might be able to tell you how to shape\n up your presentation in areas where it is perceived as weak.
- VCs are busy. Most of them are very busy, you have exactly\n one shot at making a first impression. One good way to get\n their attention is to send a single sheet executive summary\n to one of the junior partners if there are multiple partners,\n otherwise to a senior partner. If they're interested they'll\n hand it off to one of their underlings who will contact you,\n or they might contact you directly. Don't harrass them, but\n do mail two weeks or so later if you haven't had a response\n to inquire what they thought of it.
== a word of warning ==
- There are 'middle-men' in the VC world that sell their\n services to unsuspecting young companies in order to\n get them capital - so they say. The trick is that once they\n have your signature on a piece of paper that gives them\n exclusivity they no longer have to do anything. If you find\n your own capital they will claim their pound of flesh.
Selling 'access' is meaningless. For the price of a google\n search you can find more VCs than you could possibly hope\n to pitch, if your stuff is good and you can present it well\n you will most likely succeed in finding funding, even if\n you do not have an inside track. Work your network, get\n on that phone. It's a lot cheaper than giving some loser\n equity for doing nothing.
If you can't find anybody to refer you then maybe your\n product isn't that good, or maybe your team has a red\n flag. Ask why if you won't get a reference.
== you've been spotted ==
- If a VC approaches you because one of their spotters has\n alerted them to your existence then don't panic. They are\n simply interested, you've materialized on their radar and\n they would probably like to establish some kind of contact\n to be kept in the loop. Give them a bit of backstory about\n your company, don't gush out anything that you would not\n want a competitor to know.
If at some point you feel that the time for approaching investors\n is ripe then inform them. Until then simply keep them up to\n date of major public developments, if you get to know them\n a bit better you can ask for advice on business issues.
- make sure the VC that is asking you for information has not\n invested in a competitor! A bit of googling goes a long way\n before you start giving out confidential data. This is\n obviously not the normal case, but it does occasionally happen.
== pitching ==
- get an NDA signed by everybody that you are going to give\n confidential information if you think that there is something\n non-obvious about what you are going to present. Most people\n are over protective in this respect, but every now and then there\n is a bit of data that is really crucial. Think about if you\n even need to reveal it at this stage.
- Don't assume anything. You are pitching to people that are\n probably whip smart, but they don't have your background in\n your field. They will know business, but they may not know\n a thing about what it is that you are doing. So when you use\n words that are 'obvious' to the incrowd keep in mind that\n you are not talking to the incrowd. Get out of your techie\n mindset (unless the VC you're pitching to is extremely\n technical) and present your company as though building up\n from the ground.
- if you're the CEO of your fledgeling company keep in mind that\n you are speaking for everyone, not just for yourself, and make\n sure that you do not let any conflict of interest arise between\n you and the other founders (you really should have at least one\n co-founder). One of you should speak for all of you, but that one\n person should have the unconditional backing of the others.
- You pay your way (and they pay theirs). A VC is not under any\n obligation to refund you air-fare, hotel costs, legal, presentational\n or any other costs associated with the pitch. Conversely, you\n are not obliged to pay for any of their costs, such as legal\n and technical due dilligence, transportation and so on.
If you're short on cash and you want to pitch to a VC that is\n in an out of the way location for you, then the reality is that\n you may not be able to afford to pitch to them.
A recent weirdness is VCs charging an 'entrance fee', this is\n something to stay very far away from, anybody that wants an\n entrance fee is making money OF you, not WITH you and that\n is why they shouldn't be able to call themselves Venture\n Capitalists. Maybe Vulture Capitalists is a better term for\n such characters.
- know the terminology. If you don't know an NDA from a MOU\n then you will have to spend some time on that. Having your\n eyes glaze over halfway an interview or agreeing to something\n because you do not know what it means and you don't want to\n admit your ignorance is simply stupid. It does not mean that\n you have to know everything, it simply means that it is a lot\n easier to have a conversation with people if everybody is aware\n of the meaning of all the terms. It saves time, and makes you\n come across more professional, and hence will increase your\n chances of success (both to find capital as well as in succeeding\n with your venture).
- Other than NDA's nobody expects anything to be signed\n when it is delivered. So, do not sign stuff that you haven't\n had the time to go over, with your partners and your lawyer.
- When pitching time comes around: Sleep! Again, dead obvious,\n for sure. But the best way to get around being nervous is to\n be well rested. If you have to pitch several parties then\n try to schedule a break between them. I know that when I\n was done with a pitch I would literally be exhausted, unable\n to drive back to the office. It takes every bit of concentration\n and energy from you in a few hours time.
- never go alone. Bring someone along that you can trust and that\n will give you a no-holds barred evaluation of how you performed.
- if it doesn't work out, don't despair. No angry letters to a VC that\n rejected you, instead, thank them for the opportunity and ask them\n if it is ok to keep them informed of your further development.
Ask them why they rejected you, in as much detail as possible.
Not as a way to get the door to open again, but simply because that\n is the best you can take away from this pitch, a lesson on what went\n wrong or why you did not make the grade.
Try again, and do it better next time.
== due dilligence ==
- Due Dilligence usually consists of several parallel jobs. There\n are legal, technical and financial stages.
Legal is to make sure that you own what you're selling, that all\n the proper procedures and contracts are in place and that\n there are no hidden liabilities. Usually this will also look at\n intellectual property issues and patents if applicable.
Technical is to make sure that what you've built is solid and that\n it will not open up the investor to a potential liability because of\n technical weaknesses.
Financial is to make sure that your books are in order and up to\n date and that there are no skeletons in the closet.
Due Dilligence is a VERY invasive process, depending on the\n quality of the people that the VC hires. My own specialty,\n technical due dilligence usually takes the form of a several hour\n long grilling of the CTO of a company with anybody they wish\n to call on, subsequently they get a long list of follow up questions\n via email. I will want to see your code, meet your developers,\n look at your documentation, inspect your physical security if\n you store private information and so on.
The questions range from simple ones to very complicated\n ones and I've seen at least one CTO flee the room to fix a SPOF\n that became apparent only during the interview. (what do you mean\n you run a single database server and you've never tried to restore\n a backup ? What if that drive crashes and it turns out that none of\n your backups are restorable ?)
== getting to a deal ==
- You have to get your own legal representation. Remember, in\n this phase of the process you are on opposite sides of the table,\n and if you are lax and let the VCs handle your legal bill you are\n effectively using a lawyer who is not working in your interest.
This will cost you dearly.
Pay your own lawyer, and pay him out-of-pocket, not out of\n a deal that hasn't been done yet, the situation should be the\n same if you walk away from it or if you take it. That's important\n because otherwise you might have to do a bad deal just to pay\n the legal fees.
- Nothing is binding until it is signed. Even a LOI isn't as strong\n as a real contract, and money in the bank. That goes both\n ways, but it is considered very bad form to back out once a\n LOI is signed. Still, a letter of intent is not a contract and VCs\n have been known to bow out in spite of signing and if anything\n major happens to your company between the LOI and a real\n deal you will probably be able to back out. But forget about\n pitching that VC ever again.
- Stay in constant touch with your co-founders during the whole\n process, if possible have them there when you're pitching and\n discussing the deal. People kept in the dark are usually not\n going to be happy with a fait-accompli that is not in their\n best interests. By bringing them in on the negotiations you\n stand a much better chance of not messing up your internal\n affairs.
- Feel free to request a better deal! Remember, the VCs will\n negotiate what's best for them. You have to be in control of\n your side, you have to know what it is that you want and how\n much you are willing to give up for it. There is no 'bad' deal\n that was not done, the only deals people regret are the ones\n that they did do, for too little money or too large a stake in\n the company.
This phase can take quite a while, don't feel rushed.
- be careful, there doesn't seem to be much difference from a\n funding perspective between a convertible loan and giving\n out equity, but in practice the difference is huge, especially\n if there are survival clauses and the company goes bust.
== the data room ==
- when preparing a larger deal there will usually be a data room\n set up, a centralized spot at your lawyers, or their lawyers office\n where all the documentation that both parties provide gets\n integrated in to a seamless whole
- you have the fiduciary obligation to inform the other party of\n anything material that you think may influence the deal. If\n someone has threatened to sue you recently then state it,\n make it part of the record. If you don't and the suit does\n happen you are going to be in big trouble.
- the VC has the obligation to do their research as thorough as\n they can, time permitting.
== doing the deal ==
- Once all the details are ironed out, there will be a concept investment\n contract. Usually this will involve changes to the articles of incorporation\n or the shareholder agreements of the company that is being invested in.
Unless you are a legal eagle I'd suggest you spend a lot of quality\n time ( ;) ) with your lawyer during this phase.
- have all your co-founders go over the contract, if they're unsure about\n the language get them to bring their lawyers at their expense. Make\n sure everybody knows exactly what will happen.
Best of luck!
thanks to Mahmud for the critique."},"title":{"matchLevel":"none","matchedWords":[],"value":"Tell HN: Dealing with VCs, my experience"}},"_tags":["story","author_jacquesm","story_866299","ask_hn"],"author":"jacquesm","children":[866330,866349,866358,866366,866390,866475,866552,866682,866888,866986],"created_at":"2009-10-07T08:01:00Z","created_at_i":1254902460,"num_comments":38,"objectID":"866299","points":183,"story_id":866299,"story_text":"There are a bunch of steps to dealing with VCs that you should\nbe familiar with before stepping in to the fray.
My background in this is that I did several pitches to VCs and have worked\nfor several VCs in the last couple of years, mostly on the technical\nside but you get to see a good part of the process as perceived by\nthe companies pitching. I also helped one company get seed\ncapital, because I thought they showed great promise.
So, the following is from personal experience,\nwhich is limited but it may be useful.
Some VC's are more forgiving than others when it comes to following\nproper form, but even if they're forgiving they'll appreciate it if you\nknow your stuff before you apply. Their agendas are almost\nalways quite full and the amount of time they have to spend on you\nis limited, so you should use it wisely to maximize your chances of\nsuccess.
== before you start pitching ==
- First off, let's dispel a popular mistaken belief, getting an investment\n is not 'success'. It's a step towards a possible success, and it may\n give you a better chance, but it basically comes down to another\n party estimating that you have a chance of success and that they\n want a piece of that success in return for an investment.
- You have to know your stuff. This of course, sounds completely\n obvious but it really is surprising how many people will pitch\n to a VC (in itself something of an opportunity) and completely\n blow it by not having done basic homework. A fairly recent\n example, someone wrote a businessplan around a certain\n type of person and was pitching for capital. Halfway through\n the presentation one of the partners of the company being\n pitched interrupted and asked 'And how many of these people\n do you know ?'. A sensible question. The answer '0', more or\n less ended the interview, even though it went on for a while\n beyond that point, for politeness' sake.
Not knowing your target market, not having done any basic\n research in to the demographic that you intend to sell your\n product to is lethal.
As is not knowing your competive arena. If you come up with a\n brilliant plan that looks like some successful competitor is already executing and\n you don't know they exist that's probably the end of the ride for you. You\n really need to spend solid time on mapping out the competition. Know their weak points, know your strong points.
In short, know your stuff, expect to be challenged.
- Make sure you inform your partners about all your moves and get\n them on board before approaching a new party. To find out that\n someone isn't on board after you approach a VC is a real problem.
If there is any problem between the founders get it ironed out\n before you start making pitches, and make sure problems\n are resolved to all parties satisfaction.
- It isn't a must to be incorporated before approaching VCs, but it\n can be a problem if you picked the wrong form.
- if there are pending lawsuits it is usually a good idea to get\n those resolved before pitching as well. This can royally screw\n up your timing, a window of opportunity can easily close while\n you attend to this.
- you have to know what it is that you want that investment for,\n no matter how sure you are that you'll be able to spend it\n wisely, you need to pretty much lay out how you intend to\n use the money an investment will bring. This is where a\n businessplan comes in. These are no longer the 90's, so please\n no columns with more than 9 0's in it. Keep it realistic and\n make sure that it contains realistic estimates for the costs\n of the various components of your business. Factor in\n market rates for salaries, office space and so on. Get a feel\n for what it costs a business of a similar size to operate.
- scouting for capital takes time. Sometimes LOTS of time. Make\n sure you have that time, and make sure that your business\n does not suffer from this. Farm out as much of the work to your\n co-founders as you can, spread that load. It will give them more\n insight in the process and it will get them more involved.
== approaching target VCs ==
- when approaching a VC try to find out the names of their 'spotters',\n and pitch the spotters first. That way, you essentially get two\n chances, the spotters might be able to tell you how to shape\n up your presentation in areas where it is perceived as weak.
- VCs are busy. Most of them are very busy, you have exactly\n one shot at making a first impression. One good way to get\n their attention is to send a single sheet executive summary\n to one of the junior partners if there are multiple partners,\n otherwise to a senior partner. If they're interested they'll\n hand it off to one of their underlings who will contact you,\n or they might contact you directly. Don't harrass them, but\n do mail two weeks or so later if you haven't had a response\n to inquire what they thought of it.
== a word of warning ==
- There are 'middle-men' in the VC world that sell their\n services to unsuspecting young companies in order to\n get them capital - so they say. The trick is that once they\n have your signature on a piece of paper that gives them\n exclusivity they no longer have to do anything. If you find\n your own capital they will claim their pound of flesh.
Selling 'access' is meaningless. For the price of a google\n search you can find more VCs than you could possibly hope\n to pitch, if your stuff is good and you can present it well\n you will most likely succeed in finding funding, even if\n you do not have an inside track. Work your network, get\n on that phone. It's a lot cheaper than giving some loser\n equity for doing nothing.
If you can't find anybody to refer you then maybe your\n product isn't that good, or maybe your team has a red\n flag. Ask why if you won't get a reference.
== you've been spotted ==
- If a VC approaches you because one of their spotters has\n alerted them to your existence then don't panic. They are\n simply interested, you've materialized on their radar and\n they would probably like to establish some kind of contact\n to be kept in the loop. Give them a bit of backstory about\n your company, don't gush out anything that you would not\n want a competitor to know.
If at some point you feel that the time for approaching investors\n is ripe then inform them. Until then simply keep them up to\n date of major public developments, if you get to know them\n a bit better you can ask for advice on business issues.
- make sure the VC that is asking you for information has not\n invested in a competitor! A bit of googling goes a long way\n before you start giving out confidential data. This is\n obviously not the normal case, but it does occasionally happen.
== pitching ==
- get an NDA signed by everybody that you are going to give\n confidential information if you think that there is something\n non-obvious about what you are going to present. Most people\n are over protective in this respect, but every now and then there\n is a bit of data that is really crucial. Think about if you\n even need to reveal it at this stage.
- Don't assume anything. You are pitching to people that are\n probably whip smart, but they don't have your background in\n your field. They will know business, but they may not know\n a thing about what it is that you are doing. So when you use\n words that are 'obvious' to the incrowd keep in mind that\n you are not talking to the incrowd. Get out of your techie\n mindset (unless the VC you're pitching to is extremely\n technical) and present your company as though building up\n from the ground.
- if you're the CEO of your fledgeling company keep in mind that\n you are speaking for everyone, not just for yourself, and make\n sure that you do not let any conflict of interest arise between\n you and the other founders (you really should have at least one\n co-founder). One of you should speak for all of you, but that one\n person should have the unconditional backing of the others.
- You pay your way (and they pay theirs). A VC is not under any\n obligation to refund you air-fare, hotel costs, legal, presentational\n or any other costs associated with the pitch. Conversely, you\n are not obliged to pay for any of their costs, such as legal\n and technical due dilligence, transportation and so on.
If you're short on cash and you want to pitch to a VC that is\n in an out of the way location for you, then the reality is that\n you may not be able to afford to pitch to them.
A recent weirdness is VCs charging an 'entrance fee', this is\n something to stay very far away from, anybody that wants an\n entrance fee is making money OF you, not WITH you and that\n is why they shouldn't be able to call themselves Venture\n Capitalists. Maybe Vulture Capitalists is a better term for\n such characters.
- know the terminology. If you don't know an NDA from a MOU\n then you will have to spend some time on that. Having your\n eyes glaze over halfway an interview or agreeing to something\n because you do not know what it means and you don't want to\n admit your ignorance is simply stupid. It does not mean that\n you have to know everything, it simply means that it is a lot\n easier to have a conversation with people if everybody is aware\n of the meaning of all the terms. It saves time, and makes you\n come across more professional, and hence will increase your\n chances of success (both to find capital as well as in succeeding\n with your venture).
- Other than NDA's nobody expects anything to be signed\n when it is delivered. So, do not sign stuff that you haven't\n had the time to go over, with your partners and your lawyer.
- When pitching time comes around: Sleep! Again, dead obvious,\n for sure. But the best way to get around being nervous is to\n be well rested. If you have to pitch several parties then\n try to schedule a break between them. I know that when I\n was done with a pitch I would literally be exhausted, unable\n to drive back to the office. It takes every bit of concentration\n and energy from you in a few hours time.
- never go alone. Bring someone along that you can trust and that\n will give you a no-holds barred evaluation of how you performed.
- if it doesn't work out, don't despair. No angry letters to a VC that\n rejected you, instead, thank them for the opportunity and ask them\n if it is ok to keep them informed of your further development.
Ask them why they rejected you, in as much detail as possible.
Not as a way to get the door to open again, but simply because that\n is the best you can take away from this pitch, a lesson on what went\n wrong or why you did not make the grade.
Try again, and do it better next time.
== due dilligence ==
- Due Dilligence usually consists of several parallel jobs. There\n are legal, technical and financial stages.
Legal is to make sure that you own what you're selling, that all\n the proper procedures and contracts are in place and that\n there are no hidden liabilities. Usually this will also look at\n intellectual property issues and patents if applicable.
Technical is to make sure that what you've built is solid and that\n it will not open up the investor to a potential liability because of\n technical weaknesses.
Financial is to make sure that your books are in order and up to\n date and that there are no skeletons in the closet.
Due Dilligence is a VERY invasive process, depending on the\n quality of the people that the VC hires. My own specialty,\n technical due dilligence usually takes the form of a several hour\n long grilling of the CTO of a company with anybody they wish\n to call on, subsequently they get a long list of follow up questions\n via email. I will want to see your code, meet your developers,\n look at your documentation, inspect your physical security if\n you store private information and so on.
The questions range from simple ones to very complicated\n ones and I've seen at least one CTO flee the room to fix a SPOF\n that became apparent only during the interview. (what do you mean\n you run a single database server and you've never tried to restore\n a backup ? What if that drive crashes and it turns out that none of\n your backups are restorable ?)
== getting to a deal ==
- You have to get your own legal representation. Remember, in\n this phase of the process you are on opposite sides of the table,\n and if you are lax and let the VCs handle your legal bill you are\n effectively using a lawyer who is not working in your interest.
This will cost you dearly.
Pay your own lawyer, and pay him out-of-pocket, not out of\n a deal that hasn't been done yet, the situation should be the\n same if you walk away from it or if you take it. That's important\n because otherwise you might have to do a bad deal just to pay\n the legal fees.
- Nothing is binding until it is signed. Even a LOI isn't as strong\n as a real contract, and money in the bank. That goes both\n ways, but it is considered very bad form to back out once a\n LOI is signed. Still, a letter of intent is not a contract and VCs\n have been known to bow out in spite of signing and if anything\n major happens to your company between the LOI and a real\n deal you will probably be able to back out. But forget about\n pitching that VC ever again.
- Stay in constant touch with your co-founders during the whole\n process, if possible have them there when you're pitching and\n discussing the deal. People kept in the dark are usually not\n going to be happy with a fait-accompli that is not in their\n best interests. By bringing them in on the negotiations you\n stand a much better chance of not messing up your internal\n affairs.
- Feel free to request a better deal! Remember, the VCs will\n negotiate what's best for them. You have to be in control of\n your side, you have to know what it is that you want and how\n much you are willing to give up for it. There is no 'bad' deal\n that was not done, the only deals people regret are the ones\n that they did do, for too little money or too large a stake in\n the company.
This phase can take quite a while, don't feel rushed.
- be careful, there doesn't seem to be much difference from a\n funding perspective between a convertible loan and giving\n out equity, but in practice the difference is huge, especially\n if there are survival clauses and the company goes bust.
== the data room ==
- when preparing a larger deal there will usually be a data room\n set up, a centralized spot at your lawyers, or their lawyers office\n where all the documentation that both parties provide gets\n integrated in to a seamless whole
- you have the fiduciary obligation to inform the other party of\n anything material that you think may influence the deal. If\n someone has threatened to sue you recently then state it,\n make it part of the record. If you don't and the suit does\n happen you are going to be in big trouble.
- the VC has the obligation to do their research as thorough as\n they can, time permitting.
== doing the deal ==
- Once all the details are ironed out, there will be a concept investment\n contract. Usually this will involve changes to the articles of incorporation\n or the shareholder agreements of the company that is being invested in.
Unless you are a legal eagle I'd suggest you spend a lot of quality\n time ( ;) ) with your lawyer during this phase.
- have all your co-founders go over the contract, if they're unsure about\n the language get them to bring their lawyers at their expense. Make\n sure everybody knows exactly what will happen.
Best of luck!
thanks to Mahmud for the critique.","title":"Tell HN: Dealing with VCs, my experience","updated_at":"2024-09-19T16:44:37Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"SaratogaEnergy"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"Hi HN,
I\u2019m Drew, founder of Saratoga Energy (https://www.saratoga-energy.com). We make better carbon nanotubes at one-fifth the price. Carbon nanotubes are a form of nano-scale carbon fibers (5,000X thinner than human hair) with remarkably high strength, electrical conductivity, and thermal conductivity properties. This makes them useful in a variety of commercial applications.
Lithium-ion batteries designed for electric cars already use carbon nanotubes to reduce heat generation during charging and to improve electrical conductivity. This results in faster charging and improved battery life. However, the cost is so high ($300/kg) that battery manufacturers are forced to use the minimal amount, rather the optimal amount.
Our breakthrough in manufacturing cost enables battery manufacturers to use the optimal amount, allowing electric cars to safely recharge in 10 minutes or less. This could give electric cars the boost they need to replace gasoline engines.
The idea to start the company grew from an idea posed to me by my Dad back in 2012 - is there a way to transform carbon dioxide, a greenhouse gas, into something valuable? I thought this would be an exciting challenge because most carbon dioxide technologies at the time were focused on storing it in underground caverns or converting it into commodity chemicals where it would be difficult to profit without substantial government subsidies.
After reading an article about Tesla\u2019s intentions to only source raw materials that were sustainably produced in North America, we settled on developing a low-cost electrochemical process to convert carbon dioxide into graphite. Graphite is an essential energy storage material used in lithium-ion batteries. After deciding on the product, my Dad and I put together a small team of chemical engineers to help with the patents and applications for grants from the Department of Energy and the National Science Foundation.
When we received our funding, we began work with Lawrence Berkeley National Laboratory to construct small batteries to test the material we had made. We noticed that some cells would charge three, or sometimes five times faster than commercial reference materials. So we pulled the cells apart to have a look at the graphite with a high-powered microscope. All of the fast-charging graphite was matted with tiny hairs (I later learned that it sounds cooler if you call them \u201ccarbon nanotubes\u201d).
We thought about it for a few days and did some research. We eventually figured out that certain metals we had tested in our production process were likely responsible for the growth of carbon nanotubes. So we isolated those metals to test the theory. It worked!
Now we had a process that could either grow carbon nanotubes or graphite - both at an estimated cost under $5/kg. The difference being that the market price for battery-grade graphite is $10/kg and the market price for battery-grade carbon nanotubes is $300/kg.
If the price of graphite is reduced from $10/kg to $5/kg, electric cars get a bit less expensive and the market expands a bit more. If the price of carbon nanotubes is reduced from $300/kg to $5/kg, electric cars cold potentially charge in about the same amount of time it takes to refill a tank of gas, which could create exponential growth in the electric car market. We discussed this with our grant manager and agreed that it made sense to pivot and scale up the carbon nanotube process - which leaves us here today in W19.
What\u2019s different about our technology is that we produce carbon nanotubes through the electrolysis of molten carbonate salts. The electrochemical reaction produces carbon (nanotubes), oxygen gas, and metal oxides, which are further reacted with carbon dioxide to re-generate the carbonate salt starting material. So the net reaction is the input of energy to drive the conversion of carbon dioxide to carbon nanotubes and oxygen.
Industry has been using chemical vapor deposition to make carbon nanotubes since they were first discovered in 1991. We believe that our platform is better for a few reasons: 1) electrochemistry is tunable and this gives us control over the size and shape of the nanotubes, so they can be custom-tailored for specific battery chemistries and applications outside of energy storage as well; 2) the energy requirement for our manufacturing process is estimated to be 27 kWh/kg, five times less energy intensive than chemical vapor deposition; and 3) our technology represents a value-added use for carbon dioxide, and if powered by electricity from renewable sources, would have a negative carbon footprint.
I think the reason nobody has commercialized this production method yet, or come up with some other high-efficiency process, is because chemical vapor deposition is relatively simple to operate and relatively simple to scale. Billions of dollars worth of chemical vapor deposition infrastructure are already established, and historically, there haven\u2019t been many new market opportunities that would justify investing in new technologies to drive down cost.
Only recently have researchers demonstrated the potential for carbon nanotubes to improve the performance of new applications like advanced energy storage, high-strength carbon fibers and composites, lightweight electrical wiring, and concrete composites for roads that don\u2019t crack. If carbon nanotubes were less expensive, these new applications could be worth billions while also creating sizable reductions in greenhouse gas emissions. It is our mission to bring new markets like these to life and to develop new products that best take advantage of what our carbon nanotubes have to offer.
I believe that we are best positioned to break this cost-curve and bring this technology to market because molten carbonate electrochemistry is not a well-known science. What we\u2019ve learned since our foundation in 2012 is not commonly taught in universities. Our knowledge was acquired through hands-on experience and in-house development of intellectual property.
Hopefully some of you folks are also interested in carbon nanotubes, or at least share a mutual dislike for carbon dioxide. I\u2019m looking forward to sharing some ideas with everyone!
-Drew"},"title":{"matchLevel":"none","matchedWords":[],"value":"Launch HN: Saratoga Energy (YC W19), Better Carbon Nanotubes from Carbon Dioxide"}},"_tags":["story","author_SaratogaEnergy","story_19303387","launch_hn"],"author":"SaratogaEnergy","children":[19303741,19303840,19304205,19304286,19304379,19304419,19304847,19305162,19305418,19305453,19305817,19306420,19306923,19307357,19307450,19418540],"created_at":"2019-03-04T17:53:09Z","created_at_i":1551721989,"num_comments":51,"objectID":"19303387","points":104,"story_id":19303387,"story_text":"Hi HN,
I\u2019m Drew, founder of Saratoga Energy (https://www.saratoga-energy.com). We make better carbon nanotubes at one-fifth the price. Carbon nanotubes are a form of nano-scale carbon fibers (5,000X thinner than human hair) with remarkably high strength, electrical conductivity, and thermal conductivity properties. This makes them useful in a variety of commercial applications.
Lithium-ion batteries designed for electric cars already use carbon nanotubes to reduce heat generation during charging and to improve electrical conductivity. This results in faster charging and improved battery life. However, the cost is so high ($300/kg) that battery manufacturers are forced to use the minimal amount, rather the optimal amount.
Our breakthrough in manufacturing cost enables battery manufacturers to use the optimal amount, allowing electric cars to safely recharge in 10 minutes or less. This could give electric cars the boost they need to replace gasoline engines.
The idea to start the company grew from an idea posed to me by my Dad back in 2012 - is there a way to transform carbon dioxide, a greenhouse gas, into something valuable? I thought this would be an exciting challenge because most carbon dioxide technologies at the time were focused on storing it in underground caverns or converting it into commodity chemicals where it would be difficult to profit without substantial government subsidies.
After reading an article about Tesla\u2019s intentions to only source raw materials that were sustainably produced in North America, we settled on developing a low-cost electrochemical process to convert carbon dioxide into graphite. Graphite is an essential energy storage material used in lithium-ion batteries. After deciding on the product, my Dad and I put together a small team of chemical engineers to help with the patents and applications for grants from the Department of Energy and the National Science Foundation.
When we received our funding, we began work with Lawrence Berkeley National Laboratory to construct small batteries to test the material we had made. We noticed that some cells would charge three, or sometimes five times faster than commercial reference materials. So we pulled the cells apart to have a look at the graphite with a high-powered microscope. All of the fast-charging graphite was matted with tiny hairs (I later learned that it sounds cooler if you call them \u201ccarbon nanotubes\u201d).
We thought about it for a few days and did some research. We eventually figured out that certain metals we had tested in our production process were likely responsible for the growth of carbon nanotubes. So we isolated those metals to test the theory. It worked!
Now we had a process that could either grow carbon nanotubes or graphite - both at an estimated cost under $5/kg. The difference being that the market price for battery-grade graphite is $10/kg and the market price for battery-grade carbon nanotubes is $300/kg.
If the price of graphite is reduced from $10/kg to $5/kg, electric cars get a bit less expensive and the market expands a bit more. If the price of carbon nanotubes is reduced from $300/kg to $5/kg, electric cars cold potentially charge in about the same amount of time it takes to refill a tank of gas, which could create exponential growth in the electric car market. We discussed this with our grant manager and agreed that it made sense to pivot and scale up the carbon nanotube process - which leaves us here today in W19.
What\u2019s different about our technology is that we produce carbon nanotubes through the electrolysis of molten carbonate salts. The electrochemical reaction produces carbon (nanotubes), oxygen gas, and metal oxides, which are further reacted with carbon dioxide to re-generate the carbonate salt starting material. So the net reaction is the input of energy to drive the conversion of carbon dioxide to carbon nanotubes and oxygen.
Industry has been using chemical vapor deposition to make carbon nanotubes since they were first discovered in 1991. We believe that our platform is better for a few reasons: 1) electrochemistry is tunable and this gives us control over the size and shape of the nanotubes, so they can be custom-tailored for specific battery chemistries and applications outside of energy storage as well; 2) the energy requirement for our manufacturing process is estimated to be 27 kWh/kg, five times less energy intensive than chemical vapor deposition; and 3) our technology represents a value-added use for carbon dioxide, and if powered by electricity from renewable sources, would have a negative carbon footprint.
I think the reason nobody has commercialized this production method yet, or come up with some other high-efficiency process, is because chemical vapor deposition is relatively simple to operate and relatively simple to scale. Billions of dollars worth of chemical vapor deposition infrastructure are already established, and historically, there haven\u2019t been many new market opportunities that would justify investing in new technologies to drive down cost.
Only recently have researchers demonstrated the potential for carbon nanotubes to improve the performance of new applications like advanced energy storage, high-strength carbon fibers and composites, lightweight electrical wiring, and concrete composites for roads that don\u2019t crack. If carbon nanotubes were less expensive, these new applications could be worth billions while also creating sizable reductions in greenhouse gas emissions. It is our mission to bring new markets like these to life and to develop new products that best take advantage of what our carbon nanotubes have to offer.
I believe that we are best positioned to break this cost-curve and bring this technology to market because molten carbonate electrochemistry is not a well-known science. What we\u2019ve learned since our foundation in 2012 is not commonly taught in universities. Our knowledge was acquired through hands-on experience and in-house development of intellectual property.
Hopefully some of you folks are also interested in carbon nanotubes, or at least share a mutual dislike for carbon dioxide. I\u2019m looking forward to sharing some ideas with everyone!
-Drew","title":"Launch HN: Saratoga Energy (YC W19), Better Carbon Nanotubes from Carbon Dioxide","updated_at":"2024-09-20T03:47:56Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"JPLeRouzic"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"Dear HNers,
You may have read this article:
https://www.cnn.com/2025/02/17/health/zolgensma-sma-gene-therapy-drug-pricing-propublica/index.html
To over-simplify the article, it's about a non-profit that financed pre-clinical studies on a genetic disease that strikes infants: SMA.
https://en.wikipedia.org/wiki/Spinal_muscular_atrophy
Once the pre-clinical studies were finished, contacts with the non-profit were much less frequent, and in the end, the drug was approved by the FDA. The price is more than $2 million per dose.
This drug, Zolgensma, involves administering an AAV9 virus capsid containing an SMN1 transgene. Today, commercial companies have mastered this technology, are very efficient, and can produce it at little cost.
I wonder if there could be a collaborative effort to lower the price of such drugs. One way would be to find alternate providers but a main roadblock is the intellectual property.
The first patents on this drug will expire in a few years (they were written in the mid-2000).
I know it's possible to make three kinds of attacks on PI:
- One would be to find an alternative PI not related to the original patents, but it requires pre-clinical studies. The competition of the patent holder would be pleased to finance such pre-clinical studies.
Another option is to find prior art to invalidate the patent. This seems weird, as there should have been at least prior art research before accepting a patent, yet it's feasible. I had the honor of participating in one.
- Then it's possible to evade a PI because the system of claims has a weak point. For example, let's say the administration of the drug is part of the claims, and that several claims depend on that particular. Then it would be trivial to create another patent with the same efficacy but with another administration route. Indeed this is only an example to make this last point clear, I don't claim it has some validity.
I assume there is no need to sue the patent holder as the mere existence of such a website would lower prices.
What do you think of the idea of creating a sort of website where a challenge will be presented: Find a strategy to enable lowering the price on this or that drug, and contributors could discuss and propose solutions. It would be a sort of GitHub project.
Thanks for your comments!"},"title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: Opinion on efforts to find prior art on outrageous priced drugs"}},"_tags":["story","author_JPLeRouzic","story_43119492","ask_hn"],"author":"JPLeRouzic","children":[43119782,43122793],"created_at":"2025-02-20T20:07:53Z","created_at_i":1740082073,"num_comments":10,"objectID":"43119492","points":8,"story_id":43119492,"story_text":"Dear HNers,
You may have read this article:
https://www.cnn.com/2025/02/17/health/zolgensma-sma-gene-therapy-drug-pricing-propublica/index.html
To over-simplify the article, it's about a non-profit that financed pre-clinical studies on a genetic disease that strikes infants: SMA.
https://en.wikipedia.org/wiki/Spinal_muscular_atrophy
Once the pre-clinical studies were finished, contacts with the non-profit were much less frequent, and in the end, the drug was approved by the FDA. The price is more than $2 million per dose.
This drug, Zolgensma, involves administering an AAV9 virus capsid containing an SMN1 transgene. Today, commercial companies have mastered this technology, are very efficient, and can produce it at little cost.
I wonder if there could be a collaborative effort to lower the price of such drugs. One way would be to find alternate providers but a main roadblock is the intellectual property.
The first patents on this drug will expire in a few years (they were written in the mid-2000).
I know it's possible to make three kinds of attacks on PI:
- One would be to find an alternative PI not related to the original patents, but it requires pre-clinical studies. The competition of the patent holder would be pleased to finance such pre-clinical studies.
Another option is to find prior art to invalidate the patent. This seems weird, as there should have been at least prior art research before accepting a patent, yet it's feasible. I had the honor of participating in one.
- Then it's possible to evade a PI because the system of claims has a weak point. For example, let's say the administration of the drug is part of the claims, and that several claims depend on that particular. Then it would be trivial to create another patent with the same efficacy but with another administration route. Indeed this is only an example to make this last point clear, I don't claim it has some validity.
I assume there is no need to sue the patent holder as the mere existence of such a website would lower prices.
What do you think of the idea of creating a sort of website where a challenge will be presented: Find a strategy to enable lowering the price on this or that drug, and contributors could discuss and propose solutions. It would be a sort of GitHub project.
Thanks for your comments!","title":"Ask HN: Opinion on efforts to find prior art on outrageous priced drugs","updated_at":"2025-02-23T20:07:56Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"mrjay42"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["prime","intellect"],"value":"First of, a disclaimer: I have NOTHING against IPv6. I find it great. I find also AWESOME that all my devices behind my ISP's router/modem/box (I'll just call it a "box"), are provided with a public IPv6 address.
1.
But then...you have mobile clients who use 4G/5G.\nAnd guess what, some telecommunication operators, in 2025, still do not support IPv6 over 4G.\nThat means, that you're feel completely MAD as YOU can access ALL you services from "outside" your local network using a computer. But when you switch to a phone -> you dead. Then you enable the WiFi connection on your Phone and NOW your services are accessible again
2.
Some software DO NOT support IPv6. For instance Arma 3 server doens't support IPv6. Arma 3 is a 2013 game, which is still sold for a good price on Steam (27.99\u20ac, last time I checked https://store.steampowered.com/app/107410/Arma_3/), and the game is still played A LOT https://steamcharts.com/app/107410. So money is flowing, don't tell me this couldn't be fixed/updated.
And this is just ONE example.
3.\nIPv6 SHOULD be the default if available. That's the policy that should be implemented EVERYWHERE. I'm talking over the Internet, ok? I don't mean for our local networks at home (even if honestly, it's the same thing EXCEPT that the addresses are not easily memorizable for a human).
Windows 10 (up to date, even if, yes technically they're not updating it any longer), doesn't have IPv6 enabled by default! You have to do it manually, and THEN, sometimes, for no reason it's disabled again, or weirdly malfunctions in mysterious ways. (Now it's stable on my machine tho)
4.
All those little troubles add A LOT of time figuring out how could I solve my problem.\nThe solutions, to this day are limited in some ways:
Buy a VPS, tunnel everything there (easier said than done, but doable). The problem is: I do not want to have to subscribe to a VPS just because IPv6 isn't supported properly.
Install some kind of cloudflare tunnel. And I hate that. I hate it. To set it up, I had to run some kind of docker container, that registered something somewhere, generated a URL that I clicked on, then I had to change my nameservers on my registrar's side and I have to keep a service running for this tunnel to be working.
All of those things are NOT in my control, and I cannot accept it as a "valid solution". And GUESS WHAT? Today, Cloudflare is down ( https://news.ycombinator.com/item?id=45963949 ) -> BOOM perfect example, you delegate ONE service to the cloud -> all my services are inaccessible. And I'm not criticizing Cloudflare here, they provide a tunnel for me, for free, but the thing is, I installed that solution reluctantly.
5.\nWhat works amazing well with IPv6:
* Linux -> just works
* DNS, nameserver, registrar -> you set up the AAAA register for your domain/subdomain, boom job done. And this seems like a detail, but we all know the joke "it's always the DNS" as in "it's always the DNS' fault", but, get that, DNS (the protocol) works perfectly with IPv6.
6.\nI think about small businesses, small associations, NGOs, small clubs, etc. -> if they have the will to host their own services.
Because let's say that their data must be protected or simply, like me, they like to own and host stuff, because it is technically and intellectually satisfying to finally be able to use Internet as it was intended (thank you all the Gods of Olympus for my symmetric gigabit fiber optic connection) -> well, if their ISP is as stingy as mine and running out of IPv4 addresses, they're gonna end up having to do things in IPv6...and it'll surely be the beginning of a series of small technical nightmares for them."},"title":{"matchLevel":"none","matchedWords":[],"value":"Running a self hosted server with IPv6 is still a nightmare in 2025"}},"_tags":["story","author_mrjay42","story_45967106","ask_hn"],"author":"mrjay42","children":[45973055,46007131],"created_at":"2025-11-18T15:03:02Z","created_at_i":1763478182,"num_comments":2,"objectID":"45967106","points":5,"story_id":45967106,"story_text":"First of, a disclaimer: I have NOTHING against IPv6. I find it great. I find also AWESOME that all my devices behind my ISP's router/modem/box (I'll just call it a "box"), are provided with a public IPv6 address.
1.
But then...you have mobile clients who use 4G/5G.\nAnd guess what, some telecommunication operators, in 2025, still do not support IPv6 over 4G.\nThat means, that you're feel completely MAD as YOU can access ALL you services from "outside" your local network using a computer. But when you switch to a phone -> you dead. Then you enable the WiFi connection on your Phone and NOW your services are accessible again
2.
Some software DO NOT support IPv6. For instance Arma 3 server doens't support IPv6. Arma 3 is a 2013 game, which is still sold for a good price on Steam (27.99\u20ac, last time I checked https://store.steampowered.com/app/107410/Arma_3/), and the game is still played A LOT https://steamcharts.com/app/107410. So money is flowing, don't tell me this couldn't be fixed/updated.
And this is just ONE example.
3.\nIPv6 SHOULD be the default if available. That's the policy that should be implemented EVERYWHERE. I'm talking over the Internet, ok? I don't mean for our local networks at home (even if honestly, it's the same thing EXCEPT that the addresses are not easily memorizable for a human).
Windows 10 (up to date, even if, yes technically they're not updating it any longer), doesn't have IPv6 enabled by default! You have to do it manually, and THEN, sometimes, for no reason it's disabled again, or weirdly malfunctions in mysterious ways. (Now it's stable on my machine tho)
4.
All those little troubles add A LOT of time figuring out how could I solve my problem.\nThe solutions, to this day are limited in some ways:
Buy a VPS, tunnel everything there (easier said than done, but doable). The problem is: I do not want to have to subscribe to a VPS just because IPv6 isn't supported properly.
Install some kind of cloudflare tunnel. And I hate that. I hate it. To set it up, I had to run some kind of docker container, that registered something somewhere, generated a URL that I clicked on, then I had to change my nameservers on my registrar's side and I have to keep a service running for this tunnel to be working.
All of those things are NOT in my control, and I cannot accept it as a "valid solution". And GUESS WHAT? Today, Cloudflare is down ( https://news.ycombinator.com/item?id=45963949 ) -> BOOM perfect example, you delegate ONE service to the cloud -> all my services are inaccessible. And I'm not criticizing Cloudflare here, they provide a tunnel for me, for free, but the thing is, I installed that solution reluctantly.
5.\nWhat works amazing well with IPv6:
* Linux -> just works
* DNS, nameserver, registrar -> you set up the AAAA register for your domain/subdomain, boom job done. And this seems like a detail, but we all know the joke "it's always the DNS" as in "it's always the DNS' fault", but, get that, DNS (the protocol) works perfectly with IPv6.
6.\nI think about small businesses, small associations, NGOs, small clubs, etc. -> if they have the will to host their own services.
Because let's say that their data must be protected or simply, like me, they like to own and host stuff, because it is technically and intellectually satisfying to finally be able to use Internet as it was intended (thank you all the Gods of Olympus for my symmetric gigabit fiber optic connection) -> well, if their ISP is as stingy as mine and running out of IPv4 addresses, they're gonna end up having to do things in IPv6...and it'll surely be the beginning of a series of small technical nightmares for them.","title":"Running a self hosted server with IPv6 is still a nightmare in 2025","updated_at":"2026-03-05T23:02:10Z"}],"hitsPerPage":50,"nbHits":68,"nbPages":2,"page":0,"params":"query=Prime+Intellect&tags=story&hitsPerPage=50&advancedSyntax=true&analyticsTags=backend","processingTimeMS":16,"processingTimingsMS":{"_request":{"roundTrip":15},"afterFetch":{"format":{"highlighting":3,"total":4},"merge":{"mergeLoop":{"prepareNextHit":1,"total":1},"total":1},"total":1},"fetch":{"query":8,"scanning":4,"total":13},"total":16},"query":"Prime Intellect","serverTimeMS":21}