But every time I opened a textbook, I wanted to quit immediately. Algebra on paper feels static, opaque, and disconnected from intuition.
I don\u2019t think algebra was meant to live only on paper. It becomes much more beautiful on a screen, when you can interact with it directly, change things, and immediately see what happens.
You stop memorizing steps and start understanding why equations behave the way they do.
Another thing I realized: I didn\u2019t want to be alone with it.
Learning algebra can feel strangely isolating. I wanted to feel that I was on the path with other people nearby, even if we were each working through our own problems.
So I spent a long time experimenting with friends and trying to build that sense of presence into the project: seeing what others are working on, feeling that the room is alive, and knowing you are not the only one struggling through the same ideas.
And honestly, I\u2019m just tired of the endless AI slop, registrations, freemium traps, ads, onboarding flows, and two-factor-authentication rituals around things that should be simple.
At some point I realized I didn\u2019t want another platform optimized for retention, upsells, and onboarding funnels. I just wanted something that helps people learn.
Join us. Let\u2019s learn algebra together.
And if you like it, please show it to your kids, students, or anyone trying to rebuild their algebra intuition. If people find it useful, I\u2019ll keep developing it."},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Show HN: Learn algebra together, no AI slop, no ads or freemium, no registration"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://algebra.tutlm.com/"}},"_tags":["story","author_nik_norman","story_48123364","show_hn"],"author":"nik_norman","created_at":"2026-05-13T15:40:00Z","created_at_i":1778686800,"num_comments":0,"objectID":"48123364","points":2,"story_id":48123364,"story_text":"I think the modern world increasingly demands technical literacy. I felt that myself, and I also realized I still had gaps in algebra from school.
But every time I opened a textbook, I wanted to quit immediately. Algebra on paper feels static, opaque, and disconnected from intuition.
I don\u2019t think algebra was meant to live only on paper. It becomes much more beautiful on a screen, when you can interact with it directly, change things, and immediately see what happens.
You stop memorizing steps and start understanding why equations behave the way they do.
Another thing I realized: I didn\u2019t want to be alone with it.
Learning algebra can feel strangely isolating. I wanted to feel that I was on the path with other people nearby, even if we were each working through our own problems.
So I spent a long time experimenting with friends and trying to build that sense of presence into the project: seeing what others are working on, feeling that the room is alive, and knowing you are not the only one struggling through the same ideas.
And honestly, I\u2019m just tired of the endless AI slop, registrations, freemium traps, ads, onboarding flows, and two-factor-authentication rituals around things that should be simple.
At some point I realized I didn\u2019t want another platform optimized for retention, upsells, and onboarding funnels. I just wanted something that helps people learn.
Join us. Let\u2019s learn algebra together.
And if you like it, please show it to your kids, students, or anyone trying to rebuild their algebra intuition. If people find it useful, I\u2019ll keep developing it.","title":"Show HN: Learn algebra together, no AI slop, no ads or freemium, no registration","updated_at":"2026-05-13T15:41:35Z","url":"https://algebra.tutlm.com/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"rbanffy"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"MIT's Chefsi Brings Together AI, HPC and Materials Data for Advanced Simulations"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"https://www.hpcwire.com/2025/09/19/mits-chefsi-brings-together-ai-hpc-and-materials-data-for-advanced-simulations/"}},"_tags":["story","author_rbanffy","story_45316789"],"author":"rbanffy","created_at":"2025-09-20T19:55:24Z","created_at_i":1758398124,"num_comments":0,"objectID":"45316789","points":2,"story_id":45316789,"title":"MIT's Chefsi Brings Together AI, HPC and Materials Data for Advanced Simulations","updated_at":"2026-03-05T22:40:43Z","url":"https://www.hpcwire.com/2025/09/19/mits-chefsi-brings-together-ai-hpc-and-materials-data-for-advanced-simulations/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Brajeshwar"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Whisper \u2013 free and open source voice note taking app from Together.ai"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.usewhisper.io"}},"_tags":["story","author_Brajeshwar","story_44660590"],"author":"Brajeshwar","created_at":"2025-07-23T15:50:40Z","created_at_i":1753285840,"num_comments":0,"objectID":"44660590","points":2,"story_id":44660590,"title":"Whisper \u2013 free and open source voice note taking app from Together.ai","updated_at":"2025-07-23T19:22:29Z","url":"https://www.usewhisper.io"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"geox"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Hymn of Babylon pieced together using AI"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://www.thetimes.com/world/europe/article/hymn-babylon-archaeologists-ai-bv2tjvpp2"}},"_tags":["story","author_geox","story_44447243"],"author":"geox","created_at":"2025-07-02T18:33:10Z","created_at_i":1751481190,"num_comments":0,"objectID":"44447243","points":2,"story_id":44447243,"title":"Hymn of Babylon pieced together using AI","updated_at":"2025-07-02T22:22:09Z","url":"https://www.thetimes.com/world/europe/article/hymn-babylon-archaeologists-ai-bv2tjvpp2"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"lokinchiandy"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Hi HN!
I'm Andy, and I\u2019ve been working on Promptma AI (https://promptma.ai), a tool that enhances the prompt engineering experience when using Together AI. Please use desktop version for now. There are 4 major features:
1. Prompt History \n2. Prompt Engineering with Parameters \n3. Result and Cost Comparison\n4. Template Boilerplate
While developing Copsule (https://www.copsule.com), a video summarization tool, my team and I ran into several challenges with prompt engineering. We needed to test different prompts and models to optimize results and manage costs. We tried using Together AI, but found it lacked some key features like prompt history, parameter adjustments, and the ability to compare different prompt results and costs directly.
To make this process easier, we created Promptma, which builds on top of Together AI with the features mentioned above.
We built Promptma to streamline the prompt engineering process, making it more efficient and cost-effective.
I\u2019d love for you all to try it out and share your thoughts. Your feedback would be incredibly valuable as this is a self-funded project.
Thanks!"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Show HN: Promptma AI \u2013 A better UX to use Together AI for prompt engineering"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://www.promptma.ai/"}},"_tags":["story","author_lokinchiandy","story_40387824","show_hn"],"author":"lokinchiandy","created_at":"2024-05-17T08:59:27Z","created_at_i":1715936367,"num_comments":0,"objectID":"40387824","points":2,"story_id":40387824,"story_text":"Hi HN!
I'm Andy, and I\u2019ve been working on Promptma AI (https://promptma.ai), a tool that enhances the prompt engineering experience when using Together AI. Please use desktop version for now. There are 4 major features:
1. Prompt History \n2. Prompt Engineering with Parameters \n3. Result and Cost Comparison\n4. Template Boilerplate
While developing Copsule (https://www.copsule.com), a video summarization tool, my team and I ran into several challenges with prompt engineering. We needed to test different prompts and models to optimize results and manage costs. We tried using Together AI, but found it lacked some key features like prompt history, parameter adjustments, and the ability to compare different prompt results and costs directly.
To make this process easier, we created Promptma, which builds on top of Together AI with the features mentioned above.
We built Promptma to streamline the prompt engineering process, making it more efficient and cost-effective.
I\u2019d love for you all to try it out and share your thoughts. Your feedback would be incredibly valuable as this is a self-funded project.
Thanks!","title":"Show HN: Promptma AI \u2013 A better UX to use Together AI for prompt engineering","updated_at":"2024-09-20T17:07:19Z","url":"https://www.promptma.ai/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"FinalDestiny"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Fine-Tuning on an M1 Mac with Mistral, Ollama, and Together.ai"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"https://zaaane.com/blog/fine-tuning-on-an-m1-mac-with-mistral-and-togetherai"}},"_tags":["story","author_FinalDestiny","story_39387876"],"author":"FinalDestiny","created_at":"2024-02-15T20:02:53Z","created_at_i":1708027373,"num_comments":0,"objectID":"39387876","points":2,"story_id":39387876,"title":"Fine-Tuning on an M1 Mac with Mistral, Ollama, and Together.ai","updated_at":"2024-09-20T16:30:07Z","url":"https://zaaane.com/blog/fine-tuning-on-an-m1-mac-with-mistral-and-togetherai"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"swyx"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Cloud Intelligence at the speed of 5000 tok/s \u2013 with Together AI"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["together"],"value":"https://www.latent.space/p/together"}},"_tags":["story","author_swyx","story_39304871"],"author":"swyx","created_at":"2024-02-08T17:33:02Z","created_at_i":1707413582,"num_comments":0,"objectID":"39304871","points":2,"story_id":39304871,"title":"Cloud Intelligence at the speed of 5000 tok/s \u2013 with Together AI","updated_at":"2024-09-20T16:20:53Z","url":"https://www.latent.space/p/together"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"bhattmayurshiv"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Our commitment to advancing bold and responsible AI, together"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"https://blog.google/outreach-initiatives/public-policy/our-commitment-to-advancing-bold-and-responsible-ai-together/"}},"_tags":["story","author_bhattmayurshiv","story_36858609"],"author":"bhattmayurshiv","created_at":"2023-07-25T06:16:01Z","created_at_i":1690265761,"num_comments":0,"objectID":"36858609","points":2,"story_id":36858609,"title":"Our commitment to advancing bold and responsible AI, together","updated_at":"2024-09-20T14:39:27Z","url":"https://blog.google/outreach-initiatives/public-policy/our-commitment-to-advancing-bold-and-responsible-ai-together/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"guico"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Y42: Glue Airbyte, dbt and Airflow together with a few clicks"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://techcrunch.com/2022/11/02/y42-wants-to-become-mission-control-for-your-data-pipelines/"}},"_tags":["story","author_guico","story_33442122"],"author":"guico","created_at":"2022-11-02T20:01:00Z","created_at_i":1667419260,"num_comments":0,"objectID":"33442122","points":2,"story_id":33442122,"title":"Y42: Glue Airbyte, dbt and Airflow together with a few clicks","updated_at":"2024-09-20T12:24:37Z","url":"https://techcrunch.com/2022/11/02/y42-wants-to-become-mission-control-for-your-data-pipelines/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jonbaer"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"DeepMind and Unity will work together on AI research"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://www.engadget.com/2018/09/26/deepmind-unity-ai-machine-learning-environments/"}},"_tags":["story","author_jonbaer","story_18080093"],"author":"jonbaer","created_at":"2018-09-26T20:44:14Z","created_at_i":1537994654,"num_comments":0,"objectID":"18080093","points":2,"story_id":18080093,"title":"DeepMind and Unity will work together on AI research","updated_at":"2024-09-20T03:08:47Z","url":"https://www.engadget.com/2018/09/26/deepmind-unity-ai-machine-learning-environments/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"zainhsn"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Flux API available on Together AI:FLUX1.1 [pro] and free access FLUX.1 [schnell]"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"https://www.together.ai/blog/flux-api-is-now-available-on-together-ai-new-pro-free-access-to-flux-schnell"}},"_tags":["story","author_zainhsn","story_41732484"],"author":"zainhsn","children":[41732567,41732569,41732570,41732572],"created_at":"2024-10-03T16:53:39Z","created_at_i":1727974419,"num_comments":1,"objectID":"41732484","points":1,"story_id":41732484,"title":"Flux API available on Together AI:FLUX1.1 [pro] and free access FLUX.1 [schnell]","updated_at":"2024-10-05T07:02:24Z","url":"https://www.together.ai/blog/flux-api-is-now-available-on-together-ai-new-pro-free-access-to-flux-schnell"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"kristianp"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"0Din: A GenAI Bug Bounty Program \u2013 Securing Tomorrow's AI Together"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"https://hacks.mozilla.org/2024/08/0din-a-genai-bug-bounty-program-securing-tomorrows-ai-together/"}},"_tags":["story","author_kristianp","story_41572292"],"author":"kristianp","children":[41572946],"created_at":"2024-09-17T20:22:17Z","created_at_i":1726604537,"num_comments":1,"objectID":"41572292","points":1,"story_id":41572292,"title":"0Din: A GenAI Bug Bounty Program \u2013 Securing Tomorrow's AI Together","updated_at":"2024-09-20T17:51:38Z","url":"https://hacks.mozilla.org/2024/08/0din-a-genai-bug-bounty-program-securing-tomorrows-ai-together/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"programd"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"A universal subject Llama 3.1 tutor from Together AI"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"https://llamatutor.together.ai/"}},"_tags":["story","author_programd","story_41148498"],"author":"programd","children":[41149524],"created_at":"2024-08-03T19:01:59Z","created_at_i":1722711719,"num_comments":1,"objectID":"41148498","points":1,"story_id":41148498,"title":"A universal subject Llama 3.1 tutor from Together AI","updated_at":"2024-09-20T17:34:13Z","url":"https://llamatutor.together.ai/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jamiedg"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Together AI embeddings endpoint with higher quality, 4x lower cost than OpenAI"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"https://www.together.ai/blog/embeddings-endpoint-release"}},"_tags":["story","author_jamiedg","story_38957966"],"author":"jamiedg","children":[38957967],"created_at":"2024-01-11T19:53:09Z","created_at_i":1705002789,"num_comments":1,"objectID":"38957966","points":1,"story_id":38957966,"title":"Together AI embeddings endpoint with higher quality, 4x lower cost than OpenAI","updated_at":"2024-09-20T16:05:54Z","url":"https://www.together.ai/blog/embeddings-endpoint-release"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"iProject"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Skifta\u2019s new iOS app brings DLNA and AirPlay together"},"url":{"matchLevel":"none","matchedWords":[],"value":"http://gigaom.com/2013/01/10/skifta-ios-app/"}},"_tags":["story","author_iProject","story_5036515"],"author":"iProject","created_at":"2013-01-10T12:11:25Z","created_at_i":1357819885,"num_comments":0,"objectID":"5036515","points":1,"story_id":5036515,"title":"Skifta\u2019s new iOS app brings DLNA and AirPlay together","updated_at":"2024-09-19T19:08:42Z","url":"http://gigaom.com/2013/01/10/skifta-ios-app/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"josephjrobison"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Together AI, Apps Flyer lead list of Top dynamic companies for Q3 2026"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://greenflagdigital.com/gfd-tech-100-ranking/"}},"_tags":["story","author_josephjrobison","story_48865936"],"author":"josephjrobison","created_at":"2026-07-10T22:08:16Z","created_at_i":1783721296,"num_comments":0,"objectID":"48865936","points":1,"story_id":48865936,"title":"Together AI, Apps Flyer lead list of Top dynamic companies for Q3 2026","updated_at":"2026-07-10T22:13:04Z","url":"https://greenflagdigital.com/gfd-tech-100-ranking/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"danw"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Jason Fried: Working closely together ain't productive"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"http://www.lifeclever.com/jason-fried-working-closely-together-aint-productive/"}},"_tags":["story","author_danw","story_47574"],"author":"danw","created_at":"2007-08-28T20:22:16Z","created_at_i":1188332536,"num_comments":0,"objectID":"47574","points":1,"story_id":47574,"title":"Jason Fried: Working closely together ain't productive","updated_at":"2024-09-19T16:21:27Z","url":"http://www.lifeclever.com/jason-fried-working-closely-together-aint-productive/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"flurdy"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Mob Together: When AI Joins the Team"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"https://blog.flurdy.com/2026/02/mob-together-when-ai-joins-the-team"}},"_tags":["story","author_flurdy","story_46915979"],"author":"flurdy","created_at":"2026-02-06T17:57:51Z","created_at_i":1770400671,"num_comments":0,"objectID":"46915979","points":1,"story_id":46915979,"title":"Mob Together: When AI Joins the Team","updated_at":"2026-03-05T23:31:23Z","url":"https://blog.flurdy.com/2026/02/mob-together-when-ai-joins-the-team"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"benatkin"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Joining Together AI and Introducing CodeSandbox SDK"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"https://codesandbox.io/blog/joining-together-ai-introducing-codesandbox-sdk"}},"_tags":["story","author_benatkin","story_42405789"],"author":"benatkin","created_at":"2024-12-13T03:51:05Z","created_at_i":1734061865,"num_comments":0,"objectID":"42405789","points":1,"story_id":42405789,"title":"Joining Together AI and Introducing CodeSandbox SDK","updated_at":"2024-12-13T03:53:48Z","url":"https://codesandbox.io/blog/joining-together-ai-introducing-codesandbox-sdk"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"vishnurnair"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Generate an entire app from a prompt using Together AI's LlamaCoder"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"https://llamacoder.together.ai/"}},"_tags":["story","author_vishnurnair","story_41599494"],"author":"vishnurnair","created_at":"2024-09-20T06:52:35Z","created_at_i":1726815155,"num_comments":0,"objectID":"41599494","points":1,"story_id":41599494,"title":"Generate an entire app from a prompt using Together AI's LlamaCoder","updated_at":"2024-09-20T17:53:57Z","url":"https://llamacoder.together.ai/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"feross"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"0Din: A GenAI Bug Bounty Program \u2013 Securing Tomorrow's AI Together"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"https://hacks.mozilla.org/2024/08/0din-a-genai-bug-bounty-program-securing-tomorrows-ai-together/"}},"_tags":["story","author_feross","story_41195043"],"author":"feross","created_at":"2024-08-08T19:04:17Z","created_at_i":1723143857,"num_comments":0,"objectID":"41195043","points":1,"story_id":41195043,"title":"0Din: A GenAI Bug Bounty Program \u2013 Securing Tomorrow's AI Together","updated_at":"2024-09-20T17:38:56Z","url":"https://hacks.mozilla.org/2024/08/0din-a-genai-bug-bounty-program-securing-tomorrows-ai-together/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"tosh"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Together AI adds Mixtral-8x22B"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://twitter.com/togethercompute/status/1778052158501667128"}},"_tags":["story","author_tosh","story_39990745"],"author":"tosh","created_at":"2024-04-10T13:51:47Z","created_at_i":1712757107,"num_comments":0,"objectID":"39990745","points":1,"story_id":39990745,"title":"Together AI adds Mixtral-8x22B","updated_at":"2024-09-20T16:55:44Z","url":"https://twitter.com/togethercompute/status/1778052158501667128"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"estrada9166"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Explore the world together with AI-assisted collaborative travel itineraries"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://aicotravel.com"}},"_tags":["story","author_estrada9166","story_36537920"],"author":"estrada9166","created_at":"2023-06-30T16:30:49Z","created_at_i":1688142649,"num_comments":0,"objectID":"36537920","points":1,"story_id":36537920,"title":"Explore the world together with AI-assisted collaborative travel itineraries","updated_at":"2024-09-20T14:25:22Z","url":"https://aicotravel.com"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"cmcky"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Apple Wants to Make It Easier for People to Learn AI Together"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"https://www.maginative.com/article/apple-wants-to-make-it-easier-for-people-to-learn-ai-together"}},"_tags":["story","author_cmcky","story_35807595"],"author":"cmcky","created_at":"2023-05-03T20:49:49Z","created_at_i":1683146989,"num_comments":0,"objectID":"35807595","points":1,"story_id":35807595,"title":"Apple Wants to Make It Easier for People to Learn AI Together","updated_at":"2024-09-20T13:57:36Z","url":"https://www.maginative.com/article/apple-wants-to-make-it-easier-for-people-to-learn-ai-together"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"hidayatshah"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Bringing the World Closer Together.ai in Paki"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.facebook.com/profile.php?id=100015691087901"}},"_tags":["story","author_hidayatshah","story_20271599"],"author":"hidayatshah","created_at":"2019-06-25T06:14:04Z","created_at_i":1561443244,"num_comments":0,"objectID":"20271599","points":1,"story_id":20271599,"title":"Bringing the World Closer Together.ai in Paki","updated_at":"2024-09-20T04:29:39Z","url":"https://www.facebook.com/profile.php?id=100015691087901"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"panarky"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"The best hardware, software and AI\u2013together"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"https://www.blog.google/topics/hardware/the-best-hardware-software-and-ai-together/"}},"_tags":["story","author_panarky","story_15402653"],"author":"panarky","created_at":"2017-10-04T18:03:22Z","created_at_i":1507140202,"num_comments":0,"objectID":"15402653","points":1,"story_id":15402653,"title":"The best hardware, software and AI\u2013together","updated_at":"2024-09-20T01:25:11Z","url":"https://www.blog.google/topics/hardware/the-best-hardware-software-and-ai-together/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"n_ermosh"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Hey Hacker News! I\u2019m Nikita, founder of Airhart Aeronautics (https://www.airhartaero.com/). We are building an airplane for people who don\u2019t fly airplanes. The goal is to make flying as easy as driving a car\u2014while maintaining a high bar for safety. Here\u2019s a video that shows a bit of our hardware and quite a bit of our software: https://youtu.be/PGJUGUceu8A
In the US, trips that are 50-300 miles are almost all done by car because that distance is too short for commercial airlines and too far for public transportation. Thanks to the Wright Brothers we've had aerial transport for over 100 years. The US has over 19,000 airports, and large commercial airplane technology has developed to the point that the planes practically fly themselves. If we already have the infrastructure and the technology, why isn't everyone flying planes?
The problem is that small airplane technology hasn\u2019t innovated and is stuck in the past. Flying a small airplane is complicated, mentally taxing, and dangerous\u2014about 28x more dangerous than driving a car. Outdated airplanes, coupled with outdated flight controls, lead to regular accidents, often due to some form of loss of control. The planes are expensive and margins are small. There is no incentive to innovate within the current market, so we are looking at the new, untapped market of those who don\u2019t think about flying as an option today and making it an option.
I first came across this when I learned to fly in 2020. I was learning in a \u201cmodern\u201d GA airplane but was immediately struck by the fact that an airplane built in 2018 did not have an engine computer and there was a manual level to control the fuel/air mixture ratio. Starting it on a hot day was like starting a stubborn lawn mower. On top of that, my instructor was telling me all the various ways I could kill myself if I\u2019m not running at 100% concentration for hours on end. This just didn\u2019t sit right with me.
At the time I was working at SpaceX as an avionics engineer, leading the development of the avionics for the fairing recovery program. I also built autonomous aircraft when I was a student at Cornell, where I got a degree in electrical and computer engineering. It was clear to me that the core problem is that airplanes are too unsafe and too complicated to operate which is keeping too many people out of aviation. So, I decided to leave SpaceX and was joined by my long-time friend Brendan (he was a software engineer at Apple at the time; we built autonomous aircraft together at Cornell) to start Airhart to tackle this problem and make flying safer and more accessible.
We are developing a full hardware and software package to change how people fly airplanes. It\u2019s a fly-by-wire control system, meaning instead of mechanical linkages between the pilot\u2019s control stick and the control surfaces, it\u2019s a joystick that sends digital commands to a computer that then moves the control surfaces accordingly with servo actuators. We\u2019re developing all of the hardware ourselves: the computers, the sensors, the actuators\u2013and all of the software that actually does the control. But it\u2019s not just fly-by-wire. On top of it, we are implementing a simplified control scheme that reduces flying the airplane to just one action to perform one maneuver.
For readers who aren\u2019t pilots: all flying is basically coordinating the aircraft pitch, roll, yaw, and throttle to coordinate actions. Something as simple as a level turn to the right means you have to 1) roll the airplane, 2) use your feet on the rudder pedals to keep the turn coordinated, 3) pull back to increase your lift since you are now losing lift in a bank, 4) monitor your airspeed (especially if at slow speeds when coming in to land), 5) monitor your altitude as you\u2019re adjusting your lift in (3), 6) monitor your turn coordination as you adjust it in (2). You are now established in a turn. To return to flying straight and level do those in reverse. And while doing all this, you need to be navigating through complex airpaces and talking to air traffic control over 1940s radio technology. All this together makes it very hard to fly and very easy for a pilot (especially a new pilot) to lose control of the airplane, which is still the leading cause of fatalities in general aviation.
With Airhart Assist (that\u2019s what we call our system), you just push a control stick to the right and the flight computers do all those steps to put you into a coordinated level turn.
So, how does this actually work?
The force-feedback joystick in the plane sends its position to a flight controller (actually 3 that work in parallel for safety and redundancy, more on that later). The flight controller interprets the position as a turn rate or climb rate command (for left/right or forward/back). The flight controller also reads a bunch of sensors (gyroscope, accelerometer, magnetometer, air pressure, GPS, etc) to develop an accurate estimate of the airplane\u2019s state: roll, pitch, yaw, velocity, position, etc. Using the current state from the sensor fusion algorithms and the desired state from the joystick, the controller does a bunch of aerodynamics and control theory math to compute the control surface position necessary to bring the aircraft to the desired state. Mixed into this is error checking, envelope protection, and other various safety measures to make sure the aircraft never enters an unsafe state.
Unlike a traditional airplane, it becomes impossible to command the airplane into a stall, a spin, unsafe attitudes, or other bad states. This is the key to the safety of the system: it prevents the common mistakes that pilots make that lead to disastrous consequences.
To make sure that this system is always functioning, everything is single-fault tolerant. That means that there are no single points of failure. Any fault that might occur\u2013a broken wire, a fried resistor, a bitflip in a processor, a random hang in a kernel\u2013does not affect the functionality of the system. This is achieved by having three flight controllers that take in information from two different sets of sensors (we call them \u201cstrings\u201d), independently compute the desired actions to take, and vote on what to do. Each string has its own power source, backup battery, networking hardware, and set of critical sensors.
The only real single point of failure is the engine. We only have one, though the engine itself has redundant ignition systems, fuel pumps, controllers, etc. If the engine were to die, the batteries would keep the system running for ~30 minutes, giving you time to make an emergency landing. If the pilot somehow becomes incapacitated, any passenger can initiate an autonomous emergency landing. And if many things go wrong and the system does fail, there\u2019s a full airframe parachute that can be activated to bring the airplane safely to the ground.
A lot of people will likely wonder: \u201cisn\u2019t removing stick and rudder skills going to make worse pilots\u201d? Short answer: no. The core of what makes a good pilot isn\u2019t stick and rudder skills; it\u2019s good decision making and risk management. For single pilots in GA, it\u2019s even more important. So we are building a system that will give our pilots the tools to focus entirely on decision making and risk management and remove the distraction of stick and rudder that creates so many problems today. We think stick and rudder skills are definitely a necessity for airline pilots flying hundreds of people on board for the extremely rare cases where emergencies do happen and many people's lives are at risk, but not for an average person flying a four seat airplane to go on a weekend trip to the mountains. Our system makes it impossible to lose control of the airplane, potentially solving 80% of today\u2019s fatal accidents in general aviation.
Fly-by-wire systems typically cost millions of dollars. We intend to build it for much less. How? By leveraging automotive grade components, clever sensor fusion math so that we can use MEMS gyroscopes that cost <$100 instead of laser-ring gyros that cost $1000 if not $10k, and by a first principles approach to the design of our system. This requires that we build a lot of our own hardware. We\u2019ve developed our own control surface actuators, our own display assemblies, we\u2019re developing our own radios and GPS hardware (an aviation grade GPS can cost upwards of $10k, but it\u2019s the same hardware as in a $20 consumer grade GPS).
To take advantage of this automotive style approach requires scale. Enter the final third of the problem: flying isn\u2019t sexy. Modern airplanes look like they are from the 90s. With our first airplane, the Airhart Sling, we are redesigning the entire UI/UX of the flight deck to make it as easy as possible to use, redesigning the cabin to feel much more like a luxury car than an airplane today, and integrating Airhart Assist to make flying much more accessible and much more inviting. You can see previews of the Airhart Sling on our website, https://www.airhartaero.com/. The sexiness of design is extremely important for the economies of scale of an automotive-style approach to work.
There\u2019s a plethora of other problems that make flying cumbersome: weight and balance worksheets, complicated route planning, talking to ATC, lengthy preflight checks, a fractured system of FBOs, difficult access to instruction, the list goes on. We are working on all of these too, but no amount of extra UI features can solve the fundamental problem that aviating itself is hard. So that\u2019s what we\u2019re solving first.
We want people who don\u2019t think about airplanes as a mode of transportation to start flying and are hoping that Airhart will pave the way. Whether you fly planes today or not, I\u2019d love to hear your thoughts. This is a very exciting topic with lots to discuss so I\u2019m very much looking forward to the conversation!"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Launch HN: Airhart Aeronautics (YC S22) \u2013 A modern personal airplane"}},"_tags":["story","author_n_ermosh","story_41163382","launch_hn"],"author":"n_ermosh","children":[41163496,41163596,41163615,41163660,41163692,41163738,41163754,41163764,41163795,41163818,41163849,41163947,41163958,41163990,41164081,41164113,41164114,41164168,41164190,41164262,41164414,41164421,41164481,41164501,41164512,41164513,41164546,41164580,41164660,41164698,41164777,41164783,41164820,41164901,41164929,41165006,41165177,41165189,41165243,41165291,41165427,41165444,41165503,41165589,41165595,41165616,41165651,41165706,41165710,41165720,41165722,41165729,41165732,41165986,41166015,41166017,41166037,41166103,41166109,41166110,41166126,41166130,41166167,41166263,41166281,41166329,41166362,41166423,41166445,41166465,41166477,41166480,41166498,41166510,41166535,41166578,41166580,41166582,41166593,41166760,41166768,41166841,41166990,41167028,41167039,41167045,41167113,41167245,41167308,41167346,41167363,41167392,41167397,41167453,41167468,41167482,41167540,41167580,41167659,41167811,41167998,41168011,41168050,41168199,41168209,41168293,41168367,41168406,41168568,41168603,41168635,41168687,41168731,41168804,41168844,41168856,41169074,41169135,41169489,41169507,41169581,41169914,41169929,41169999,41170078,41170144,41170540,41170542,41170579,41170681,41170794,41170880,41172131,41176464,41178012,41178977,41178994,41236867],"created_at":"2024-08-05T17:26:43Z","created_at_i":1722878803,"num_comments":615,"objectID":"41163382","points":738,"story_id":41163382,"story_text":"Hey Hacker News! I\u2019m Nikita, founder of Airhart Aeronautics (https://www.airhartaero.com/). We are building an airplane for people who don\u2019t fly airplanes. The goal is to make flying as easy as driving a car\u2014while maintaining a high bar for safety. Here\u2019s a video that shows a bit of our hardware and quite a bit of our software: https://youtu.be/PGJUGUceu8A
In the US, trips that are 50-300 miles are almost all done by car because that distance is too short for commercial airlines and too far for public transportation. Thanks to the Wright Brothers we've had aerial transport for over 100 years. The US has over 19,000 airports, and large commercial airplane technology has developed to the point that the planes practically fly themselves. If we already have the infrastructure and the technology, why isn't everyone flying planes?
The problem is that small airplane technology hasn\u2019t innovated and is stuck in the past. Flying a small airplane is complicated, mentally taxing, and dangerous\u2014about 28x more dangerous than driving a car. Outdated airplanes, coupled with outdated flight controls, lead to regular accidents, often due to some form of loss of control. The planes are expensive and margins are small. There is no incentive to innovate within the current market, so we are looking at the new, untapped market of those who don\u2019t think about flying as an option today and making it an option.
I first came across this when I learned to fly in 2020. I was learning in a \u201cmodern\u201d GA airplane but was immediately struck by the fact that an airplane built in 2018 did not have an engine computer and there was a manual level to control the fuel/air mixture ratio. Starting it on a hot day was like starting a stubborn lawn mower. On top of that, my instructor was telling me all the various ways I could kill myself if I\u2019m not running at 100% concentration for hours on end. This just didn\u2019t sit right with me.
At the time I was working at SpaceX as an avionics engineer, leading the development of the avionics for the fairing recovery program. I also built autonomous aircraft when I was a student at Cornell, where I got a degree in electrical and computer engineering. It was clear to me that the core problem is that airplanes are too unsafe and too complicated to operate which is keeping too many people out of aviation. So, I decided to leave SpaceX and was joined by my long-time friend Brendan (he was a software engineer at Apple at the time; we built autonomous aircraft together at Cornell) to start Airhart to tackle this problem and make flying safer and more accessible.
We are developing a full hardware and software package to change how people fly airplanes. It\u2019s a fly-by-wire control system, meaning instead of mechanical linkages between the pilot\u2019s control stick and the control surfaces, it\u2019s a joystick that sends digital commands to a computer that then moves the control surfaces accordingly with servo actuators. We\u2019re developing all of the hardware ourselves: the computers, the sensors, the actuators\u2013and all of the software that actually does the control. But it\u2019s not just fly-by-wire. On top of it, we are implementing a simplified control scheme that reduces flying the airplane to just one action to perform one maneuver.
For readers who aren\u2019t pilots: all flying is basically coordinating the aircraft pitch, roll, yaw, and throttle to coordinate actions. Something as simple as a level turn to the right means you have to 1) roll the airplane, 2) use your feet on the rudder pedals to keep the turn coordinated, 3) pull back to increase your lift since you are now losing lift in a bank, 4) monitor your airspeed (especially if at slow speeds when coming in to land), 5) monitor your altitude as you\u2019re adjusting your lift in (3), 6) monitor your turn coordination as you adjust it in (2). You are now established in a turn. To return to flying straight and level do those in reverse. And while doing all this, you need to be navigating through complex airpaces and talking to air traffic control over 1940s radio technology. All this together makes it very hard to fly and very easy for a pilot (especially a new pilot) to lose control of the airplane, which is still the leading cause of fatalities in general aviation.
With Airhart Assist (that\u2019s what we call our system), you just push a control stick to the right and the flight computers do all those steps to put you into a coordinated level turn.
So, how does this actually work?
The force-feedback joystick in the plane sends its position to a flight controller (actually 3 that work in parallel for safety and redundancy, more on that later). The flight controller interprets the position as a turn rate or climb rate command (for left/right or forward/back). The flight controller also reads a bunch of sensors (gyroscope, accelerometer, magnetometer, air pressure, GPS, etc) to develop an accurate estimate of the airplane\u2019s state: roll, pitch, yaw, velocity, position, etc. Using the current state from the sensor fusion algorithms and the desired state from the joystick, the controller does a bunch of aerodynamics and control theory math to compute the control surface position necessary to bring the aircraft to the desired state. Mixed into this is error checking, envelope protection, and other various safety measures to make sure the aircraft never enters an unsafe state.
Unlike a traditional airplane, it becomes impossible to command the airplane into a stall, a spin, unsafe attitudes, or other bad states. This is the key to the safety of the system: it prevents the common mistakes that pilots make that lead to disastrous consequences.
To make sure that this system is always functioning, everything is single-fault tolerant. That means that there are no single points of failure. Any fault that might occur\u2013a broken wire, a fried resistor, a bitflip in a processor, a random hang in a kernel\u2013does not affect the functionality of the system. This is achieved by having three flight controllers that take in information from two different sets of sensors (we call them \u201cstrings\u201d), independently compute the desired actions to take, and vote on what to do. Each string has its own power source, backup battery, networking hardware, and set of critical sensors.
The only real single point of failure is the engine. We only have one, though the engine itself has redundant ignition systems, fuel pumps, controllers, etc. If the engine were to die, the batteries would keep the system running for ~30 minutes, giving you time to make an emergency landing. If the pilot somehow becomes incapacitated, any passenger can initiate an autonomous emergency landing. And if many things go wrong and the system does fail, there\u2019s a full airframe parachute that can be activated to bring the airplane safely to the ground.
A lot of people will likely wonder: \u201cisn\u2019t removing stick and rudder skills going to make worse pilots\u201d? Short answer: no. The core of what makes a good pilot isn\u2019t stick and rudder skills; it\u2019s good decision making and risk management. For single pilots in GA, it\u2019s even more important. So we are building a system that will give our pilots the tools to focus entirely on decision making and risk management and remove the distraction of stick and rudder that creates so many problems today. We think stick and rudder skills are definitely a necessity for airline pilots flying hundreds of people on board for the extremely rare cases where emergencies do happen and many people's lives are at risk, but not for an average person flying a four seat airplane to go on a weekend trip to the mountains. Our system makes it impossible to lose control of the airplane, potentially solving 80% of today\u2019s fatal accidents in general aviation.
Fly-by-wire systems typically cost millions of dollars. We intend to build it for much less. How? By leveraging automotive grade components, clever sensor fusion math so that we can use MEMS gyroscopes that cost <$100 instead of laser-ring gyros that cost $1000 if not $10k, and by a first principles approach to the design of our system. This requires that we build a lot of our own hardware. We\u2019ve developed our own control surface actuators, our own display assemblies, we\u2019re developing our own radios and GPS hardware (an aviation grade GPS can cost upwards of $10k, but it\u2019s the same hardware as in a $20 consumer grade GPS).
To take advantage of this automotive style approach requires scale. Enter the final third of the problem: flying isn\u2019t sexy. Modern airplanes look like they are from the 90s. With our first airplane, the Airhart Sling, we are redesigning the entire UI/UX of the flight deck to make it as easy as possible to use, redesigning the cabin to feel much more like a luxury car than an airplane today, and integrating Airhart Assist to make flying much more accessible and much more inviting. You can see previews of the Airhart Sling on our website, https://www.airhartaero.com/. The sexiness of design is extremely important for the economies of scale of an automotive-style approach to work.
There\u2019s a plethora of other problems that make flying cumbersome: weight and balance worksheets, complicated route planning, talking to ATC, lengthy preflight checks, a fractured system of FBOs, difficult access to instruction, the list goes on. We are working on all of these too, but no amount of extra UI features can solve the fundamental problem that aviating itself is hard. So that\u2019s what we\u2019re solving first.
We want people who don\u2019t think about airplanes as a mode of transportation to start flying and are hoping that Airhart will pave the way. Whether you fly planes today or not, I\u2019d love to hear your thoughts. This is a very exciting topic with lots to discuss so I\u2019m very much looking forward to the conversation!","title":"Launch HN: Airhart Aeronautics (YC S22) \u2013 A modern personal airplane","updated_at":"2026-02-19T16:41:42Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"sestinj"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Hi HN,
Code Llama was released, but we noticed a ton of questions in the main thread about how/where to use it \u2014 not just from an API or the terminal, but in your own codebase as a drop-in replacement for Copilot Chat. Without this, developers don't get much utility from the model.
This concern is also important because benchmarks like HumanEval don't perfectly reflect the quality of responses. There's likely to be a flurry of improvements to coding models in the coming months, and rather than relying on the benchmarks to evaluate them, the community will get better feedback from people actually using the models. This means real usage in real, everyday workflows.
We've worked to make this possible with Continue (https://github.com/continuedev/continue) and want to hear what you find to be the real capabilities of Code Llama. Is it on-par with GPT-4, does it require fine-tuning, or does it excel at certain tasks?
If you\u2019d like to try Code Llama with Continue, it only takes a few steps to set up (https://continue.dev/docs/walkthroughs/codellama), either locally with Ollama, or through TogetherAI or Replicate's APIs."},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Use Code Llama as Drop-In Replacement for Copilot Chat"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://continue.dev/docs/walkthroughs/codellama"}},"_tags":["story","author_sestinj","story_37251969","show_hn"],"author":"sestinj","children":[37252341,37252470,37252548,37252598,37252960,37252989,37253193,37254656,37256338],"created_at":"2023-08-24T17:33:12Z","created_at_i":1692898392,"num_comments":52,"objectID":"37251969","points":187,"story_id":37251969,"story_text":"Hi HN,
Code Llama was released, but we noticed a ton of questions in the main thread about how/where to use it \u2014 not just from an API or the terminal, but in your own codebase as a drop-in replacement for Copilot Chat. Without this, developers don't get much utility from the model.
This concern is also important because benchmarks like HumanEval don't perfectly reflect the quality of responses. There's likely to be a flurry of improvements to coding models in the coming months, and rather than relying on the benchmarks to evaluate them, the community will get better feedback from people actually using the models. This means real usage in real, everyday workflows.
We've worked to make this possible with Continue (https://github.com/continuedev/continue) and want to hear what you find to be the real capabilities of Code Llama. Is it on-par with GPT-4, does it require fine-tuning, or does it excel at certain tasks?
If you\u2019d like to try Code Llama with Continue, it only takes a few steps to set up (https://continue.dev/docs/walkthroughs/codellama), either locally with Ollama, or through TogetherAI or Replicate's APIs.","title":"Show HN: Use Code Llama as Drop-In Replacement for Copilot Chat","updated_at":"2024-12-04T21:57:49Z","url":"https://continue.dev/docs/walkthroughs/codellama"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"reissbaker"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Try it out! https://glhf.chat/
Hey HN!
We\u2019ve been working for the past few months on a website to let you easily run (almost) any open-source LLM on autoscaling GPU clusters. It\u2019s free for now while we figure out how to price it, but we expect to be cheaper than most GPU offerings since we can run the models multi-tenant.
Unlike Together AI, Fireworks, etc, we\u2019ll run any model that the open-source vLLM project supports: we don\u2019t have a hardcoded list. If you want a specific model or finetune, you don\u2019t have to ask us for it: you can just paste the Hugging Face link in and it\u2019ll work (as long as vLLM supports the base model architecture, we\u2019ll run anything up to ~640GB of VRAM, give or take a little for some overhead buffer).
Large models will take a few minutes to boot, but if a bunch of people are trying to use the same model, it might already be loaded and not need boot time at all. The Llama-3-70b finetunes are especially nice, since they\u2019re basically souped-up versions of the 8b finetunes a lot of people like to run locally but don\u2019t have the VRAM for. We\u2019re expecting the Llama-3.1 finetunes to be pretty great too once they start getting released.
There are some caveats for now \u2014 for example, while we support the Deepseek V2 architecture, we actually can only run their smaller \u201cLite\u201d models due to some underlying NVLink limitations (though we\u2019re working on it). But for the most part if vLLM supports it, we should too!
We figured Llama-3.1-405B Launch Day was a good day to launch ourselves too \u2014 let us know in the comments if there\u2019s anything you want us to support, or if you run into any issues. I know it\u2019s not \u201clocal\u201d Llama, but, well, that\u2019s a lot of GPUs\u2026"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: We made glhf.chat \u2013 run almost any open-source LLM, including 405B"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://glhf.chat/landing/home"}},"_tags":["story","author_reissbaker","story_41052934","show_hn"],"author":"reissbaker","children":[41053167,41053180,41053237,41053256,41053266,41053332,41053457,41053522,41053558,41053605,41053652,41053789,41053792,41053794,41054249,41054622,41054917,41055114,41055145,41055346,41055402,41056635,41056836,41057755,41122401],"created_at":"2024-07-24T01:52:09Z","created_at_i":1721785929,"num_comments":95,"objectID":"41052934","points":161,"story_id":41052934,"story_text":"Try it out! https://glhf.chat/
Hey HN!
We\u2019ve been working for the past few months on a website to let you easily run (almost) any open-source LLM on autoscaling GPU clusters. It\u2019s free for now while we figure out how to price it, but we expect to be cheaper than most GPU offerings since we can run the models multi-tenant.
Unlike Together AI, Fireworks, etc, we\u2019ll run any model that the open-source vLLM project supports: we don\u2019t have a hardcoded list. If you want a specific model or finetune, you don\u2019t have to ask us for it: you can just paste the Hugging Face link in and it\u2019ll work (as long as vLLM supports the base model architecture, we\u2019ll run anything up to ~640GB of VRAM, give or take a little for some overhead buffer).
Large models will take a few minutes to boot, but if a bunch of people are trying to use the same model, it might already be loaded and not need boot time at all. The Llama-3-70b finetunes are especially nice, since they\u2019re basically souped-up versions of the 8b finetunes a lot of people like to run locally but don\u2019t have the VRAM for. We\u2019re expecting the Llama-3.1 finetunes to be pretty great too once they start getting released.
There are some caveats for now \u2014 for example, while we support the Deepseek V2 architecture, we actually can only run their smaller \u201cLite\u201d models due to some underlying NVLink limitations (though we\u2019re working on it). But for the most part if vLLM supports it, we should too!
We figured Llama-3.1-405B Launch Day was a good day to launch ourselves too \u2014 let us know in the comments if there\u2019s anything you want us to support, or if you run into any issues. I know it\u2019s not \u201clocal\u201d Llama, but, well, that\u2019s a lot of GPUs\u2026","title":"Show HN: We made glhf.chat \u2013 run almost any open-source LLM, including 405B","updated_at":"2025-08-04T01:45:57Z","url":"https://glhf.chat/landing/home"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"neilxm"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Hey everyone,
ML Blocks is a node-based workflow builder to create multi-modal AI workflows without writing any code.
You connect blocks that call various visual models like GPT4v, Segment Anything, Dino etc. along with basic image processing blocks like resize, invert color, blur, crop, and several others.
The idea is to make it easier to deploy multi-step image processing workflows, without needing to spin up endless custom OpenCV cloud functions to glue together AI models. Usually, even if you're using cloud inference servers like Replicate, you still need to write your own image processing code to pre and post-process images in your pipeline. When you're trying to move fast, that's just unnecessary overhead.
With ML Blocks, you can build a workflow and deploy the whole thing as a single API. AFAIK, ML Blocks is the only end-to-end workflow builder built specifically for image processing.
If you're curious, our models run on Replicate, HuggingFace & Modal Labs cloud GPUs and we use React Flow for the node UX."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Show HN: ML Blocks \u2013 Deploy multimodal AI workflows without code"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.mlblocks.com/"}},"_tags":["story","author_neilxm","story_39217550","show_hn"],"author":"neilxm","children":[39218436,39218489,39218589,39218828,39218917,39218970,39219784,39221531,39223527],"created_at":"2024-02-01T16:15:21Z","created_at_i":1706804121,"num_comments":32,"objectID":"39217550","points":112,"story_id":39217550,"story_text":"Hey everyone,
ML Blocks is a node-based workflow builder to create multi-modal AI workflows without writing any code.
You connect blocks that call various visual models like GPT4v, Segment Anything, Dino etc. along with basic image processing blocks like resize, invert color, blur, crop, and several others.
The idea is to make it easier to deploy multi-step image processing workflows, without needing to spin up endless custom OpenCV cloud functions to glue together AI models. Usually, even if you're using cloud inference servers like Replicate, you still need to write your own image processing code to pre and post-process images in your pipeline. When you're trying to move fast, that's just unnecessary overhead.
With ML Blocks, you can build a workflow and deploy the whole thing as a single API. AFAIK, ML Blocks is the only end-to-end workflow builder built specifically for image processing.
If you're curious, our models run on Replicate, HuggingFace & Modal Labs cloud GPUs and we use React Flow for the node UX.","title":"Show HN: ML Blocks \u2013 Deploy multimodal AI workflows without code","updated_at":"2024-09-20T16:13:50Z","url":"https://www.mlblocks.com/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"danlenton"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Hey HN! I\u2019m the founder of Unify, and we\u2019ve just released our Model Hub, which provides a collection of LLM endpoints with live runtime benchmarks all plotted across time: https://unify.ai/hub
A key finding is that static tabular runtime benchmarks for LLMs simply do not work. It\u2019s necessary to take a time-series perspective, and plot the variations through time.
We currently have 21 models provided by: Anyscale, Perplexity AI, Replicate, Together AI, OctoAI, Mistral AI and OpenAI, with more on the roadmap.
We test across different regions (Asia, US, Europe), with varied concurrency and sequence length. By plotting across time, our dashboard highlights the stability and variability of the different endpoints, and their ongoing evolution across API updates and system changes. Our benchmarking code is fully open source: https://github.com/unifyai/aibench-llm-endpoints
Our unified API also makes it very easy to test and deploy these different endpoints in production, without needing to create several accounts.
Our Hub is a work in progress, and we will be releasing new features every week.
What are your thoughts? Both positive and negative comments are very welcome. We\u2019ll try to quickly incorporate all feedback!
I recorded a quick(ish) demo video a few hours ago, explaining how to get started, for those who are interested in learning more: https://youtu.be/0a6-C2_Bmh0
There is also a longer version here: https://youtu.be/o8yD_QBhmsw
Finally, as a thanks to HN readers, the promo code \u201cHACKERNEWS\u201d can be used to claim $5 per week in free credits, compatible with our ever expanding list of LLM providers. You can sign up here [https://console.unify.ai/], and claim the free credits here [https://unify.ai/docs/hub/home/pricing.html#top-up-code] if interested.
Thanks all!\nDan"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Unify \u2013 Dynamic LLM Benchmarks and SSO for Multi-Vendor Deployment"}},"_tags":["story","author_danlenton","story_39279998","show_hn"],"author":"danlenton","children":[39288733,39289185,39289523,39294616],"created_at":"2024-02-06T20:21:24Z","created_at_i":1707250884,"num_comments":8,"objectID":"39279998","points":91,"story_id":39279998,"story_text":"Hey HN! I\u2019m the founder of Unify, and we\u2019ve just released our Model Hub, which provides a collection of LLM endpoints with live runtime benchmarks all plotted across time: https://unify.ai/hub
A key finding is that static tabular runtime benchmarks for LLMs simply do not work. It\u2019s necessary to take a time-series perspective, and plot the variations through time.
We currently have 21 models provided by: Anyscale, Perplexity AI, Replicate, Together AI, OctoAI, Mistral AI and OpenAI, with more on the roadmap.
We test across different regions (Asia, US, Europe), with varied concurrency and sequence length. By plotting across time, our dashboard highlights the stability and variability of the different endpoints, and their ongoing evolution across API updates and system changes. Our benchmarking code is fully open source: https://github.com/unifyai/aibench-llm-endpoints
Our unified API also makes it very easy to test and deploy these different endpoints in production, without needing to create several accounts.
Our Hub is a work in progress, and we will be releasing new features every week.
What are your thoughts? Both positive and negative comments are very welcome. We\u2019ll try to quickly incorporate all feedback!
I recorded a quick(ish) demo video a few hours ago, explaining how to get started, for those who are interested in learning more: https://youtu.be/0a6-C2_Bmh0
There is also a longer version here: https://youtu.be/o8yD_QBhmsw
Finally, as a thanks to HN readers, the promo code \u201cHACKERNEWS\u201d can be used to claim $5 per week in free credits, compatible with our ever expanding list of LLM providers. You can sign up here [https://console.unify.ai/], and claim the free credits here [https://unify.ai/docs/hub/home/pricing.html#top-up-code] if interested.
Thanks all!\nDan","title":"Show HN: Unify \u2013 Dynamic LLM Benchmarks and SSO for Multi-Vendor Deployment","updated_at":"2024-09-20T16:20:39Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"tomelders"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"About 3 years ago, I started work on FlightM8. For reasons that I'll simply refer to as \"life falling apart\" I stopped working on it. However, seeing Google Flights launch yesterday re-piqued my interest in it and I'm wondering wether I should pick it up again. So what is FlightM8?
Firstly, it's an Alpha. The data is out of date, and it's buggy. It's also on shared hosting so it may well buckle under pressure at the moment.
Here's the URL for the impatient: http://flightm8.tomelders.com/
Secondly: Here's what it does...
In a nutshell, FlightM8 tells you \"who flies where\". Let's say you're in London. Your boss calls and says you have to meet with Diesel Jeans on Monday. They're a big new client and they're in Bassano Del Grappa, Italy. The rest is up to you to figure out. So... How do you get from London to Bassano Del Grappa?
Well, London is pretty easy for anyone who knows it. You have Heathrow, Gatwick, Luton, Stanstead, London City airports to choose from.
Bassano del Grappa?, not so easy. First of all, where the hell is it? When you eventually find it, you realise it doesn't have an airport. Here's where FlightM8 starts being useful. Search for Bassano del Grappa and you'll see that it's served by 3 airports, all within a reasonable distance. Venice Marco Polo, Venice Treviso, Verona Villa Franca.
There's not a flight search engine that I know of that would group those airports together.
Ok, so who flies to those places? This is what flightm8 was built to do. For a combination of routes you have a choice of... Ryanair, Easyjet, Air Berlin, Germanwings, British Airways, Monarch Airlines and BMI.
I'm willing to gamble you've never heard of at least two of those airlines. And two of those airlines (Easyjet, Ryanair) do not show up on flight search websites. (A quick aside, there used to be a bunch of other budget airlines in Europe that you've probably never heard of, but a lot of them went bust in the recession)
The aim is to help you 'start' your search for flights. I started work on it out of necessity because I was flying all over europe and wasting a lot of money on flights, only to be told by a local that I could have flown to an airport I've never heard of, with an airline I've never heard of for half the money.
When I heard Google had bought IATA, I secretly hoped they'd release some sort of API. I doubt that's going to happen, which makes listing prices and times of flight near impossible in terms of cost.
But here's the question: Is FlightM8 useful? Should I resume working on it?"},"title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: My side project. Is it useful?"},"url":{"matchLevel":"none","matchedWords":[],"value":""}},"_tags":["story","author_tomelders","story_2995599","ask_hn"],"author":"tomelders","children":[2995625,2995660,2995692,2995723,2995738,2995748,2995768,2995769,2995796,2995804,2995850,2995906,2995919,2995959,2995968,2995973,2996000,2996011,2996103,2996298,2996425,2996468,2996704,2996794,2997043,3010220,3010249,3010273,3010285,3010388,3010450,3014185],"created_at":"2011-09-14T10:24:37Z","created_at_i":1315995877,"num_comments":61,"objectID":"2995599","points":81,"story_id":2995599,"story_text":"About 3 years ago, I started work on FlightM8. For reasons that I'll simply refer to as \"life falling apart\" I stopped working on it. However, seeing Google Flights launch yesterday re-piqued my interest in it and I'm wondering wether I should pick it up again. So what is FlightM8?
Firstly, it's an Alpha. The data is out of date, and it's buggy. It's also on shared hosting so it may well buckle under pressure at the moment.
Here's the URL for the impatient: http://flightm8.tomelders.com/
Secondly: Here's what it does...
In a nutshell, FlightM8 tells you \"who flies where\". Let's say you're in London. Your boss calls and says you have to meet with Diesel Jeans on Monday. They're a big new client and they're in Bassano Del Grappa, Italy. The rest is up to you to figure out. So... How do you get from London to Bassano Del Grappa?
Well, London is pretty easy for anyone who knows it. You have Heathrow, Gatwick, Luton, Stanstead, London City airports to choose from.
Bassano del Grappa?, not so easy. First of all, where the hell is it? When you eventually find it, you realise it doesn't have an airport. Here's where FlightM8 starts being useful. Search for Bassano del Grappa and you'll see that it's served by 3 airports, all within a reasonable distance. Venice Marco Polo, Venice Treviso, Verona Villa Franca.
There's not a flight search engine that I know of that would group those airports together.
Ok, so who flies to those places? This is what flightm8 was built to do. For a combination of routes you have a choice of... Ryanair, Easyjet, Air Berlin, Germanwings, British Airways, Monarch Airlines and BMI.
I'm willing to gamble you've never heard of at least two of those airlines. And two of those airlines (Easyjet, Ryanair) do not show up on flight search websites. (A quick aside, there used to be a bunch of other budget airlines in Europe that you've probably never heard of, but a lot of them went bust in the recession)
The aim is to help you 'start' your search for flights. I started work on it out of necessity because I was flying all over europe and wasting a lot of money on flights, only to be told by a local that I could have flown to an airport I've never heard of, with an airline I've never heard of for half the money.
When I heard Google had bought IATA, I secretly hoped they'd release some sort of API. I doubt that's going to happen, which makes listing prices and times of flight near impossible in terms of cost.
But here's the question: Is FlightM8 useful? Should I resume working on it?","title":"Ask HN: My side project. Is it useful?","updated_at":"2024-09-19T17:59:58Z","url":""},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"vshah1016"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Hey HN \u2014 I\u2019m Veer and my cofounder is Suryaa. We're building Cumulus Labs (YC W26), and we're releasing our latest product IonRouter (https://ionrouter.io/), an inference API for open-source and fine tuned models. You swap in our base URL, keep your existing OpenAI client code, and get access to any model (open source or finetuned to you) running on our own inference engine.
The problem we kept running into: every inference provider is either fast-but-expensive (Together, Fireworks \u2014 you pay for always-on GPUs) or cheap-but-DIY (Modal, RunPod \u2014 you configure vLLM yourself and deal with slow cold starts). Neither felt right for teams that just want to ship.
Suryaa spent years building GPU orchestration infrastructure at TensorDock and production systems at Palantir. I led ML infrastructure and Linux kernel development for Space Force and NASA contracts where the stack had to actually work under pressure. When we started building AI products ourselves, we kept hitting the same wall: GPU infrastructure was either too expensive or too much work.
So we built IonAttention \u2014 a C++ inference runtime designed specifically around the GH200's memory architecture. Most inference stacks treat GH200 as a compatibility target (make sure vLLM runs, use CPU memory as overflow). We took a different approach and built around what makes the hardware actually interesting: a 900 GB/s coherent CPU-GPU link, 452GB of LPDDR5X sitting right next to the accelerator, and 72 ARM cores you can actually use.
Three things came out of that that we think are novel: (1) using hardware cache coherence to make CUDA graphs behave as if they have dynamic parameters at zero per-step cost \u2014 something that only works on GH200-class hardware; (2) eager KV block writeback driven by immutability rather than memory pressure, which drops eviction stalls from 10ms+ to under 0.25ms; (3) phantom-tile attention scheduling at small batch sizes that cuts attention time by over 60% in the worst-affected regimes. We wrote up the details at cumulus.blog/ionattention.
On multimodal pipelines we get better performance than big players (588 tok/s vs. Together AI's 298 on the same VLM workload). We're honest that p50 latency is currently worse (~1.46s vs. 0.74s) \u2014 that's the tradeoff we're actively working on.
Pricing is per token, no idle costs: GPT-OSS-120B is $0.02 in / $0.095 out, Qwen3.5-122B is $0.20 in / $1.60 out. Full model list and pricing at https://ionrouter.io.
You can try the playground at https://ionrouter.io/playground right now, no signup required, or drop your API key in and swap the base URL \u2014 it's one line. We built this so teams can see the power of our engine and eventually come to us for their finetuned model needs using the same solution.
We're curious what you think, especially if you're running finetuned or custom models \u2014 that's the use case we've invested the most in. What's broken, what would make this actually useful for you?"},"title":{"matchLevel":"none","matchedWords":[],"value":"Launch HN: IonRouter (YC W26) \u2013 High-throughput, low-cost inference"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://ionrouter.io"}},"_tags":["story","author_vshah1016","story_47355410","launch_hn"],"author":"vshah1016","children":[47355411,47355519,47355906,47356019,47356039,47356058,47356429,47356495,47356561,47358192,47358809,47358845,47359913,47360086,47360829,47361547,47369781,47371671,47449701],"created_at":"2026-03-12T18:52:36Z","created_at_i":1773341556,"num_comments":37,"objectID":"47355410","points":72,"story_id":47355410,"story_text":"Hey HN \u2014 I\u2019m Veer and my cofounder is Suryaa. We're building Cumulus Labs (YC W26), and we're releasing our latest product IonRouter (https://ionrouter.io/), an inference API for open-source and fine tuned models. You swap in our base URL, keep your existing OpenAI client code, and get access to any model (open source or finetuned to you) running on our own inference engine.
The problem we kept running into: every inference provider is either fast-but-expensive (Together, Fireworks \u2014 you pay for always-on GPUs) or cheap-but-DIY (Modal, RunPod \u2014 you configure vLLM yourself and deal with slow cold starts). Neither felt right for teams that just want to ship.
Suryaa spent years building GPU orchestration infrastructure at TensorDock and production systems at Palantir. I led ML infrastructure and Linux kernel development for Space Force and NASA contracts where the stack had to actually work under pressure. When we started building AI products ourselves, we kept hitting the same wall: GPU infrastructure was either too expensive or too much work.
So we built IonAttention \u2014 a C++ inference runtime designed specifically around the GH200's memory architecture. Most inference stacks treat GH200 as a compatibility target (make sure vLLM runs, use CPU memory as overflow). We took a different approach and built around what makes the hardware actually interesting: a 900 GB/s coherent CPU-GPU link, 452GB of LPDDR5X sitting right next to the accelerator, and 72 ARM cores you can actually use.
Three things came out of that that we think are novel: (1) using hardware cache coherence to make CUDA graphs behave as if they have dynamic parameters at zero per-step cost \u2014 something that only works on GH200-class hardware; (2) eager KV block writeback driven by immutability rather than memory pressure, which drops eviction stalls from 10ms+ to under 0.25ms; (3) phantom-tile attention scheduling at small batch sizes that cuts attention time by over 60% in the worst-affected regimes. We wrote up the details at cumulus.blog/ionattention.
On multimodal pipelines we get better performance than big players (588 tok/s vs. Together AI's 298 on the same VLM workload). We're honest that p50 latency is currently worse (~1.46s vs. 0.74s) \u2014 that's the tradeoff we're actively working on.
Pricing is per token, no idle costs: GPT-OSS-120B is $0.02 in / $0.095 out, Qwen3.5-122B is $0.20 in / $1.60 out. Full model list and pricing at https://ionrouter.io.
You can try the playground at https://ionrouter.io/playground right now, no signup required, or drop your API key in and swap the base URL \u2014 it's one line. We built this so teams can see the power of our engine and eventually come to us for their finetuned model needs using the same solution.
We're curious what you think, especially if you're running finetuned or custom models \u2014 that's the use case we've invested the most in. What's broken, what would make this actually useful for you?","title":"Launch HN: IonRouter (YC W26) \u2013 High-throughput, low-cost inference","updated_at":"2026-03-24T18:16:12Z","url":"https://ionrouter.io"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"merge-conflict"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"GAC is a tool I built to help users spend less time summing up what was done and more time building. It uses LLMs to generate contextual git commit messages from your code changes. And it can be a drop-in replacement for `git commit -m "..."`.
Example:
feat(auth): add OAuth2 integration with GitHub and Google\n\n - Implement OAuth2 authentication flow\n\n - Add provider configuration for GitHub and Google\n\n - Create callback handler for token exchange\n\n - Update login UI with social auth buttons\nYou can try it out with uvx (no install):
uvx gac init # config wizard\n\n uvx gac\nTech details:
14 providers - Supports local (Ollama & LM Studio) and cloud (OpenAI, Anthropic, Gemini, OpenRouter, Groq, Cerebras, Chutes, Fireworks, StreamLake, Synthetic, Together AI, & Z.ai (including their extremely cheap coding plans!)).
Three verbosity modes - Standard with bullets (default), one-liners (`-o`), or verbose (`-v`) with detailed Motivation/Architecture/Impact sections.
Secret detection - Scans for API keys, tokens, and credentials before committing. Has caught my API keys on a new project when I hadn't yet gitignored .env.
Flags - Automate common workflows:
`gac -h "bug fix"` - pass hints to guide intent\n\n `gac -yo` - auto-accept the commit message in one-liner mode\n\n `gac -ayp` - stage all files, auto-accept the commit message, and push (yolo mode)\nGitHub: https://github.com/cellwebb/gac"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Git Auto Commit (GAC) \u2013 LLM-powered Git commit command line tool"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/cellwebb/gac"}},"_tags":["story","author_merge-conflict","story_45723533","show_hn"],"author":"merge-conflict","children":[45723793,45723986,45724220,45724309,45724426,45724576,45724644,45724655,45724850,45724911,45724938,45725025,45725162,45725430,45725991,45726244,45726331,45726417,45733397,45743483,45758713],"created_at":"2025-10-27T17:07:05Z","created_at_i":1761584825,"num_comments":36,"objectID":"45723533","points":56,"story_id":45723533,"story_text":"GAC is a tool I built to help users spend less time summing up what was done and more time building. It uses LLMs to generate contextual git commit messages from your code changes. And it can be a drop-in replacement for `git commit -m "..."`.
Example:
feat(auth): add OAuth2 integration with GitHub and Google\n\n - Implement OAuth2 authentication flow\n\n - Add provider configuration for GitHub and Google\n\n - Create callback handler for token exchange\n\n - Update login UI with social auth buttons\nYou can try it out with uvx (no install):
uvx gac init # config wizard\n\n uvx gac\nTech details:
14 providers - Supports local (Ollama & LM Studio) and cloud (OpenAI, Anthropic, Gemini, OpenRouter, Groq, Cerebras, Chutes, Fireworks, StreamLake, Synthetic, Together AI, & Z.ai (including their extremely cheap coding plans!)).
Three verbosity modes - Standard with bullets (default), one-liners (`-o`), or verbose (`-v`) with detailed Motivation/Architecture/Impact sections.
Secret detection - Scans for API keys, tokens, and credentials before committing. Has caught my API keys on a new project when I hadn't yet gitignored .env.
Flags - Automate common workflows:
`gac -h "bug fix"` - pass hints to guide intent\n\n `gac -yo` - auto-accept the commit message in one-liner mode\n\n `gac -ayp` - stage all files, auto-accept the commit message, and push (yolo mode)\nGitHub: https://github.com/cellwebb/gac","title":"Show HN: Git Auto Commit (GAC) \u2013 LLM-powered Git commit command line tool","updated_at":"2026-03-05T22:56:15Z","url":"https://github.com/cellwebb/gac"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"throwaway322"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"TL;DR a recent health scare has me reevaluating what I want out of life. I reason my wife and I can run a boutique staffing agency placing primarily software developers in the NYC area drawing salaries of up to 75k each, pay our overhead and together aim to donate 100k-200k+ as a business yearly.
I\u2019m a veteran of the NYC IT staffing market but recent years have me disillusioned with the business and just doing some consulting work and taking some down time. A recent health scare really makes me want to make some big life changes that include focusing on work that matters. I think we can reduce our family spending to the point where we could take a family pay cut but do work we find fulfilling that ultimately makes a difference in the world. In my mind the goal would be to somehow have a formula that gives 50% of every placement to charity and the rest to cover payroll and overhead until we reach our 150k total salaries at which time everything short of overhead would go to charity. If our message resonated with companies and candidates to a point where our daily focus could be service rather than sales the amount we could give away (as two normal working people) could be incredible.
Question to the HN crowd: Am I crazy, could this work?"},"title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: Start a charity-driven staffing firm?"}},"_tags":["story","author_throwaway322","story_19464505","ask_hn"],"author":"throwaway322","children":[19469502,19470123,19473116,19473299,19473445,19473571,19473574,19473579,19473615,19473798,19473921,19474240,19474322,19474390,19474690,19475789,19476818,19478022,19479089,19481059,19484095],"created_at":"2019-03-22T17:05:11Z","created_at_i":1553274311,"num_comments":25,"objectID":"19464505","points":22,"story_id":19464505,"story_text":"TL;DR a recent health scare has me reevaluating what I want out of life. I reason my wife and I can run a boutique staffing agency placing primarily software developers in the NYC area drawing salaries of up to 75k each, pay our overhead and together aim to donate 100k-200k+ as a business yearly.
I\u2019m a veteran of the NYC IT staffing market but recent years have me disillusioned with the business and just doing some consulting work and taking some down time. A recent health scare really makes me want to make some big life changes that include focusing on work that matters. I think we can reduce our family spending to the point where we could take a family pay cut but do work we find fulfilling that ultimately makes a difference in the world. In my mind the goal would be to somehow have a formula that gives 50% of every placement to charity and the rest to cover payroll and overhead until we reach our 150k total salaries at which time everything short of overhead would go to charity. If our message resonated with companies and candidates to a point where our daily focus could be service rather than sales the amount we could give away (as two normal working people) could be incredible.
Question to the HN crowd: Am I crazy, could this work?","title":"Ask HN: Start a charity-driven staffing firm?","updated_at":"2024-09-20T03:55:58Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"wg0"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"A lot many companies are being formed with AI automation of Enterprise spcae, lots of optimism around chained together AI agents autonomously working together without much human intervention.
Genuine question - if traditional software makes mistake, its usually deterministic, debugable, fixable and a blame can be assigned.
What's the deal with these autonomous AI agents? Let's say analysing customes paperwork to schedule some shipments from overseas and it fails to let a shipment in because it misclassified or worse, lets it it but being it on the shores under certain conditions leads to heavy financial penalties?
Who's responsible? The AI prompt automation engineer? Or the underlying platform? Or the company providing model?
If the answer is that each outcome of such model should be double checked by a human while going through all that paperwork than what's the point of having that automation in the first place?
EDIT: typos"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Ask HN: Who'll take ownership of AI's mistakes?"}},"_tags":["story","author_wg0","story_40637927","ask_hn"],"author":"wg0","children":[40638042,40641112,40641250,40641468,40642728,40646064,40649862,40650318,40652979],"created_at":"2024-06-10T19:48:18Z","created_at_i":1718048898,"num_comments":15,"objectID":"40637927","points":14,"story_id":40637927,"story_text":"A lot many companies are being formed with AI automation of Enterprise spcae, lots of optimism around chained together AI agents autonomously working together without much human intervention.
Genuine question - if traditional software makes mistake, its usually deterministic, debugable, fixable and a blame can be assigned.
What's the deal with these autonomous AI agents? Let's say analysing customes paperwork to schedule some shipments from overseas and it fails to let a shipment in because it misclassified or worse, lets it it but being it on the shores under certain conditions leads to heavy financial penalties?
Who's responsible? The AI prompt automation engineer? Or the underlying platform? Or the company providing model?
If the answer is that each outcome of such model should be double checked by a human while going through all that paperwork than what's the point of having that automation in the first place?
EDIT: typos","title":"Ask HN: Who'll take ownership of AI's mistakes?","updated_at":"2024-09-20T17:13:59Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"chirau"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Dear ,
We\u2019re emailing you because the state legislature sent the anti-home sharing bill -- the result of a backroom deal -- to Governor Cuomo's desk yesterday. That means he must take action on the bill in the next ten days.
The Governor has a choice: veto this bill, which would be a victory for middle class New Yorkers and a rejection of Albany backroom deals. Or, sign the bill and leave tens of thousands of New Yorkers vulnerable to huge fines. (Cuomo could also do nothing, which would mean it still becomes law.) Governor Cuomo needs to hear from as many of us as possible, right now, to stop this backroom deal from becoming law.
<link> Tweet and Call the Governor <link>
In June, the state legislature passed an anti-home sharing bill that would fine everyday New Yorkers who advertise their home on sites like Airbnb up to $7,500. As we\u2019ve waited to hear whether or not Governor Cuomo would sign the bill into law or veto it, you\u2019ve been by our side, taking action when it counted most to send a message to the Governor about the impact of this bad bill. Thousands of Airbnb hosts and travelers tweeted, called, and sent emails.
Now we need you to take action again. We must ensure the Governor stands with seniors, students, small businesses, and communities that rely on home sharing for economic opportunity, and not with the big hotel special interests pushing this anti-tenant, anti-homeowner, anti-innovation bill.\nTake action right now and tell Governor Cuomo why home sharing is important to you. There isn\u2019t much time!
We\u2019re in this together,
The Airbnb Team"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"An email from Airbnb"}},"_tags":["story","author_chirau","story_12748526","ask_hn"],"author":"chirau","children":[12749439],"created_at":"2016-10-19T23:11:18Z","created_at_i":1476918678,"num_comments":6,"objectID":"12748526","points":13,"story_id":12748526,"story_text":"Dear ,
We\u2019re emailing you because the state legislature sent the anti-home sharing bill -- the result of a backroom deal -- to Governor Cuomo's desk yesterday. That means he must take action on the bill in the next ten days.
The Governor has a choice: veto this bill, which would be a victory for middle class New Yorkers and a rejection of Albany backroom deals. Or, sign the bill and leave tens of thousands of New Yorkers vulnerable to huge fines. (Cuomo could also do nothing, which would mean it still becomes law.) Governor Cuomo needs to hear from as many of us as possible, right now, to stop this backroom deal from becoming law.
<link> Tweet and Call the Governor <link>
In June, the state legislature passed an anti-home sharing bill that would fine everyday New Yorkers who advertise their home on sites like Airbnb up to $7,500. As we\u2019ve waited to hear whether or not Governor Cuomo would sign the bill into law or veto it, you\u2019ve been by our side, taking action when it counted most to send a message to the Governor about the impact of this bad bill. Thousands of Airbnb hosts and travelers tweeted, called, and sent emails.
Now we need you to take action again. We must ensure the Governor stands with seniors, students, small businesses, and communities that rely on home sharing for economic opportunity, and not with the big hotel special interests pushing this anti-tenant, anti-homeowner, anti-innovation bill.\nTake action right now and tell Governor Cuomo why home sharing is important to you. There isn\u2019t much time!
We\u2019re in this together,
The Airbnb Team","title":"An email from Airbnb","updated_at":"2024-09-19T23:51:33Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"wek"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"Hi HN! I'm Karl one of the co-founders of Stravu. (https://stravu.com) Using AI for work 24x7, we realized that four things would make AI more useful for us and a lot of other power users and teams:
Editable output: AI gives output that is half right and our only option was to either keep chatting laboriously or copy it to a Google Doc. We made Stravu so you can edit what the AI says in chat or in an attached notebook. Everything editable.
Approve AI changes: When AI makes a change to some text, you can't tell what changes. We put Red/Green diffs that you can approve into Stravu.
Unify text, tables, diagrams: We were jumping between tools to work with AI on text, tables, diagrams, etc.. Just because Microsoft did that 30 years ago, doesn't mean it makes sense now. We made Stravu so you can work with AI across text, tables, diagrams, (and 2x2s, formulas, more soon) and have them inform each other.
Actual multi-player team collab with AI: We couldn't collaborate as a team in AI (even with the ChatGPT Teams plan). We wanted to be able to chat with AI together as a team or see the changes AI was making in the canvas/notebook together and edit together. So we made Stravu support multi-player collaboration in every aspect... chats, notebooks, text, tables, diagrams..etc.
Some of the use cases of our current Beta customers include: scrum teams doing feature/customer/competitive research and feature definition, account teams building vertical/geo/account plans, consultants and investment teams working on market/company analysis.
We are currently in beta and actively iterating based on user feedback. Please try it out at: https://stravu.com We highly value your feedback!"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Show HN: Stravu \u2013 Editable, multi-player AI notebooks with text, tables, diagram"}},"_tags":["story","author_wek","story_44510354","show_hn"],"author":"wek","children":[44516820,44528998],"created_at":"2025-07-09T14:17:52Z","created_at_i":1752070672,"num_comments":2,"objectID":"44510354","points":11,"story_id":44510354,"story_text":"Hi HN! I'm Karl one of the co-founders of Stravu. (https://stravu.com) Using AI for work 24x7, we realized that four things would make AI more useful for us and a lot of other power users and teams:
Editable output: AI gives output that is half right and our only option was to either keep chatting laboriously or copy it to a Google Doc. We made Stravu so you can edit what the AI says in chat or in an attached notebook. Everything editable.
Approve AI changes: When AI makes a change to some text, you can't tell what changes. We put Red/Green diffs that you can approve into Stravu.
Unify text, tables, diagrams: We were jumping between tools to work with AI on text, tables, diagrams, etc.. Just because Microsoft did that 30 years ago, doesn't mean it makes sense now. We made Stravu so you can work with AI across text, tables, diagrams, (and 2x2s, formulas, more soon) and have them inform each other.
Actual multi-player team collab with AI: We couldn't collaborate as a team in AI (even with the ChatGPT Teams plan). We wanted to be able to chat with AI together as a team or see the changes AI was making in the canvas/notebook together and edit together. So we made Stravu support multi-player collaboration in every aspect... chats, notebooks, text, tables, diagrams..etc.
Some of the use cases of our current Beta customers include: scrum teams doing feature/customer/competitive research and feature definition, account teams building vertical/geo/account plans, consultants and investment teams working on market/company analysis.
We are currently in beta and actively iterating based on user feedback. Please try it out at: https://stravu.com We highly value your feedback!","title":"Show HN: Stravu \u2013 Editable, multi-player AI notebooks with text, tables, diagram","updated_at":"2025-07-21T13:29:21Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"ifdotpy"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["together","ai"],"value":"I built a platform where you solve tasks together with AI agents (Claude Code, Codex, Cursor \u2014 any agent via SSH).
Isolated sandbox environments, automated test scoring, global leaderboard. Tasks range from easy (AI one-shots it) to hard (requires human help).
Some tasks use optimization scoring \u2014 your score recalibrates when someone beats the best result.
Built it in 6 days as a solo founder. 100% of code written with Claude Code and Codex. Stack: Go, Next.js, K8s, Supabase, Stripe."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Show HN: Kagento \u2013 LeetCode for AI Agents"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://kagento.io"}},"_tags":["story","author_ifdotpy","story_47548048","show_hn"],"author":"ifdotpy","children":[47548879],"created_at":"2026-03-27T20:47:14Z","created_at_i":1774644434,"num_comments":1,"objectID":"47548048","points":11,"story_id":47548048,"story_text":"I built a platform where you solve tasks together with AI agents (Claude Code, Codex, Cursor \u2014 any agent via SSH).
Isolated sandbox environments, automated test scoring, global leaderboard. Tasks range from easy (AI one-shots it) to hard (requires human help).
Some tasks use optimization scoring \u2014 your score recalibrates when someone beats the best result.
Built it in 6 days as a solo founder. 100% of code written with Claude Code and Codex. Stack: Go, Next.js, K8s, Supabase, Stripe.","title":"Show HN: Kagento \u2013 LeetCode for AI Agents","updated_at":"2026-03-28T17:31:13Z","url":"https://kagento.io"}],"hitsPerPage":50,"nbHits":2130,"nbPages":20,"page":0,"params":"query=Together+AI&tags=story&hitsPerPage=50&advancedSyntax=true&analyticsTags=backend","processingTimeMS":14,"processingTimingsMS":{"_request":{"queue":27,"roundTrip":14},"afterFetch":{"format":{"highlighting":4,"total":5},"merge":{"mergeLoop":{"prepareNextHit":1,"total":1},"total":1},"total":1},"fetch":{"query":3,"scanning":8,"total":12},"total":15},"query":"Together AI","serverTimeMS":48}