I'm Ivan, one of the founders of Slapdash (https://slapdash.com/). Slapdash lets you work across all of your cloud apps at desktop speed, sort of like an OS for cloud apps.
We have built a uniform, low-latency data browser (kind of like Finder) as well as a unified command line-like interface (kind of like Spotlight) for the applications you use at work.
When we left our big company jobs, one of the difficult things to part with was the tooling. Companies like Facebook and Stripe build a class of tools internally that unifies all the employees and any collaboration apps, so you can find anyone or anything the company knows. Everything is just a quick search away [0]. It\u2019s quite a useful way to work. Common questions in day-to-day work are easy to answer. What\u2019s the history of this code abstraction? What are my colleagues working on? What\u2019s the story with this customer?
Building such a system today means connecting people's cloud apps, because that's where most of the work is happening today. Even for a small team like us, our work spans Drive, Dropbox, Figma, GitHub, Asana, Notion, Docusign, Slack, Quip, etc.
The first thing we built was a low-latency file system for cloud apps. You connect an application like Drive, or GitHub to Slapdash and we give you a way to search and browse the data in a uniform interface (kind of like Finder). It turns your working world into a database you can easily query.
We modeled our file system as a graph and we built our architecture to match, with a focus on performance. We built an import system, which effectively solves a graph replication problem (translating the structure of the app data to the Slapdash graph and keeping it in sync). We then built a graph database on top of Postgres, added a data access layer with graph semantics, with GraphQL API delivering the data to the client.
Of course, the data we store is encrypted on disk, in-transit and in the data store. Slapdash employees can't see the contents of what we index since everything except the reverse index is encrypted. It\u2019s not zero-access yet, but we\u2019re building in that direction [1].
What we discovered is that by applying optimizations to how we store (sharding & colocation) and retrieve data (batching & coalescing) we could achieve an almost zero-latency [2] experience when browsing application data. As a result, it's much faster to browse Google Drive in Slapdash than in the Drive interface itself.
While the low-latency file system is interesting, we learned that being able to search and navigate is not enough utility for a single individual. People don\u2019t search as much as they think they do, and most have their unique information foraging habits that work well-enough.
However, we wanted Slapdash to be useful for anyone, not just an employee at a big company, so we turned our attention to building a new experience on top of the file system. Our goal was to take a leap in speed with which people can control their computers. We thought this was possible because the difference in UX between desktop and cloud app environments was so acute: the desktop OS is principled, integrated and fast, while cloud apps are latency-laden and confined to crowded browser tabs.
To that end, we built the Command Bar (Command Line + Search Bar). The Command Bar is best experienced as a desktop app, where it\u2019s invoked with a global shortcut. You can quickly search your apps, file tasks, peek at your calendar, create zoom meetings, etc: all with a couple of keystrokes. Of course, you can also write your own commands too.
In practice, it meaningfully cuts down the time you spend controlling the computer. For example, filing a task on GitHub might take 10 seconds of just navigating to the right screen, while you can start writing the task title within 2 seconds by invoking the "Create New GitHub Issue" command with the Command Bar. Things like searching for a customer record, doing a quick spreadsheet calculation and even routine things like opening an existing document are measurably faster. [3]
For teams and companies, Slapdash provides a unified interface to a team\u2019s collective knowledge. This has traditionally been reserved for top technology companies, but we are bringing these advantages to everybody else. And for the individual, we are making the use of disparate cloud apps feel closer to the classic experience of a desktop computer OS - fast, integrated and more productive.
We are still figuring out what apps to support, what commands we should build and how we can open up the platform for others to build on as well. We would love to hear from you on any of those counts and any feedback you might have!
[0] Facebook has something called "intern" and Stripe has an internal product called "Stripe Home".
[1] Content is stored using ECIES (with Secp256k1 curve and AES256 cipher in CTR mode), public-key-encrypted with individual per-user, per-app key pairs.
[2] It's actually not zero latency, of course, but by preloading most things on the hover state we can cut ~50ms of perceived latency (as long as the server response time is under that, which we try to do, it feels instant).
[3] We use the keystroke-level GOMS model to evaluate interface speed, but the speed difference here is large enough that it can be intuited.
Example of filing a task on GitHub: Time controlling computer: open browser, command + L ( focus location bar), type partial URL of repo until it auto-completes, wait for page to load, click on Issues, wait for page to load, click on \u201cNew Issue\u201d. Expressing actual intent: typing title of task.
Filing a task with Slapdash: Time controlling computer: Type Command + J, type \u201cCre gi\u201d to fuzzy match \u201cCreate New GitHub Issue\u201d command, hit enter. Expressing actual intent: typing title of task."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["cloud"],"value":"Launch HN: Slapdash (YC W19) \u2013 A uniform, low-latency interface for cloud apps"}},"_tags":["story","author_kanevski","story_24364811","launch_hn"],"author":"kanevski","children":[24365059,24365350,24365487,24365563,24366045,24367002,24367063,24367095,24367541,24367854,24368017,24368158,24368684,24368826,24369759,24370229,24370412,24371098,24371479,24372234,24374619,24375547,24384734,24386910,24387883],"created_at":"2020-09-03T15:13:38Z","created_at_i":1599146018,"num_comments":68,"objectID":"24364811","points":227,"story_id":24364811,"story_text":"Hello HN,
I'm Ivan, one of the founders of Slapdash (https://slapdash.com/). Slapdash lets you work across all of your cloud apps at desktop speed, sort of like an OS for cloud apps.
We have built a uniform, low-latency data browser (kind of like Finder) as well as a unified command line-like interface (kind of like Spotlight) for the applications you use at work.
When we left our big company jobs, one of the difficult things to part with was the tooling. Companies like Facebook and Stripe build a class of tools internally that unifies all the employees and any collaboration apps, so you can find anyone or anything the company knows. Everything is just a quick search away [0]. It\u2019s quite a useful way to work. Common questions in day-to-day work are easy to answer. What\u2019s the history of this code abstraction? What are my colleagues working on? What\u2019s the story with this customer?
Building such a system today means connecting people's cloud apps, because that's where most of the work is happening today. Even for a small team like us, our work spans Drive, Dropbox, Figma, GitHub, Asana, Notion, Docusign, Slack, Quip, etc.
The first thing we built was a low-latency file system for cloud apps. You connect an application like Drive, or GitHub to Slapdash and we give you a way to search and browse the data in a uniform interface (kind of like Finder). It turns your working world into a database you can easily query.
We modeled our file system as a graph and we built our architecture to match, with a focus on performance. We built an import system, which effectively solves a graph replication problem (translating the structure of the app data to the Slapdash graph and keeping it in sync). We then built a graph database on top of Postgres, added a data access layer with graph semantics, with GraphQL API delivering the data to the client.
Of course, the data we store is encrypted on disk, in-transit and in the data store. Slapdash employees can't see the contents of what we index since everything except the reverse index is encrypted. It\u2019s not zero-access yet, but we\u2019re building in that direction [1].
What we discovered is that by applying optimizations to how we store (sharding & colocation) and retrieve data (batching & coalescing) we could achieve an almost zero-latency [2] experience when browsing application data. As a result, it's much faster to browse Google Drive in Slapdash than in the Drive interface itself.
While the low-latency file system is interesting, we learned that being able to search and navigate is not enough utility for a single individual. People don\u2019t search as much as they think they do, and most have their unique information foraging habits that work well-enough.
However, we wanted Slapdash to be useful for anyone, not just an employee at a big company, so we turned our attention to building a new experience on top of the file system. Our goal was to take a leap in speed with which people can control their computers. We thought this was possible because the difference in UX between desktop and cloud app environments was so acute: the desktop OS is principled, integrated and fast, while cloud apps are latency-laden and confined to crowded browser tabs.
To that end, we built the Command Bar (Command Line + Search Bar). The Command Bar is best experienced as a desktop app, where it\u2019s invoked with a global shortcut. You can quickly search your apps, file tasks, peek at your calendar, create zoom meetings, etc: all with a couple of keystrokes. Of course, you can also write your own commands too.
In practice, it meaningfully cuts down the time you spend controlling the computer. For example, filing a task on GitHub might take 10 seconds of just navigating to the right screen, while you can start writing the task title within 2 seconds by invoking the "Create New GitHub Issue" command with the Command Bar. Things like searching for a customer record, doing a quick spreadsheet calculation and even routine things like opening an existing document are measurably faster. [3]
For teams and companies, Slapdash provides a unified interface to a team\u2019s collective knowledge. This has traditionally been reserved for top technology companies, but we are bringing these advantages to everybody else. And for the individual, we are making the use of disparate cloud apps feel closer to the classic experience of a desktop computer OS - fast, integrated and more productive.
We are still figuring out what apps to support, what commands we should build and how we can open up the platform for others to build on as well. We would love to hear from you on any of those counts and any feedback you might have!
[0] Facebook has something called "intern" and Stripe has an internal product called "Stripe Home".
[1] Content is stored using ECIES (with Secp256k1 curve and AES256 cipher in CTR mode), public-key-encrypted with individual per-user, per-app key pairs.
[2] It's actually not zero latency, of course, but by preloading most things on the hover state we can cut ~50ms of perceived latency (as long as the server response time is under that, which we try to do, it feels instant).
[3] We use the keystroke-level GOMS model to evaluate interface speed, but the speed difference here is large enough that it can be intuited.
Example of filing a task on GitHub: Time controlling computer: open browser, command + L ( focus location bar), type partial URL of repo until it auto-completes, wait for page to load, click on Issues, wait for page to load, click on \u201cNew Issue\u201d. Expressing actual intent: typing title of task.
Filing a task with Slapdash: Time controlling computer: Type Command + J, type \u201cCre gi\u201d to fuzzy match \u201cCreate New GitHub Issue\u201d command, hit enter. Expressing actual intent: typing title of task.","title":"Launch HN: Slapdash (YC W19) \u2013 A uniform, low-latency interface for cloud apps","updated_at":"2024-09-20T06:56:52Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"rbanffy"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["cloud"],"value":"A Technical Analysis of the Capital One Hack \u2013 CloudSploit"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["gmi","cloud"],"value":"https://blog.cloudsploit.com/a-technical-analysis-of-the-capital-one-hack-a9b43d7c8aea?gi=fc097eefa614"}},"_tags":["story","author_rbanffy","story_20679352"],"author":"rbanffy","created_at":"2019-08-12T20:03:26Z","created_at_i":1565640206,"num_comments":0,"objectID":"20679352","points":3,"story_id":20679352,"title":"A Technical Analysis of the Capital One Hack \u2013 CloudSploit","updated_at":"2024-09-20T04:41:07Z","url":"https://blog.cloudsploit.com/a-technical-analysis-of-the-capital-one-hack-a9b43d7c8aea?gi=fc097eefa614"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"rdli"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["cloud"],"value":"Ambassador Adds Cloud UI for CNCF Projects Emissary, Telepresence, and Argo"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["gmi"],"value":"https://blog.getambassador.io/ambassador-adds-clickops-support-for-emissary-telepresence-and-argo-175c5677f7e9?gi=78fceeb9d32"}},"_tags":["story","author_rdli","story_30455490"],"author":"rdli","created_at":"2022-02-24T15:01:17Z","created_at_i":1645714877,"num_comments":0,"objectID":"30455490","points":2,"story_id":30455490,"title":"Ambassador Adds Cloud UI for CNCF Projects Emissary, Telepresence, and Argo","updated_at":"2024-09-20T10:36:32Z","url":"https://blog.getambassador.io/ambassador-adds-clickops-support-for-emissary-telepresence-and-argo-175c5677f7e9?gi=78fceeb9d32"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"ashtavakra"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["cloud"],"value":"WordPress attacks and Cloud Armor protection on GCP \u2013 True story"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["gmi","cloud"],"value":"https://blog.thecloudside.com/wordpress-attacks-cloud-armor-protection-on-gcp-true-story-59ba25ffb7f8?gi=6923e35d97a2"}},"_tags":["story","author_ashtavakra","story_30166558"],"author":"ashtavakra","created_at":"2022-02-01T18:07:20Z","created_at_i":1643738840,"num_comments":0,"objectID":"30166558","points":2,"story_id":30166558,"title":"WordPress attacks and Cloud Armor protection on GCP \u2013 True story","updated_at":"2024-09-20T10:26:42Z","url":"https://blog.thecloudside.com/wordpress-attacks-cloud-armor-protection-on-gcp-true-story-59ba25ffb7f8?gi=6923e35d97a2"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"shahargl"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["cloud"],"value":"From Cloud Run to Kubernetes: Why Internals Matter"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["gmi","cloud"],"value":"https://anecdotes.dev/from-cloud-run-to-kubernetes-how-learning-the-internals-of-your-stack-empowers-you-as-a-developer-3d0cdf7cb7ec?gi=3b548439d3ac"}},"_tags":["story","author_shahargl","story_29066122"],"author":"shahargl","created_at":"2021-11-01T12:56:03Z","created_at_i":1635771363,"num_comments":0,"objectID":"29066122","points":2,"story_id":29066122,"title":"From Cloud Run to Kubernetes: Why Internals Matter","updated_at":"2024-09-20T09:43:16Z","url":"https://anecdotes.dev/from-cloud-run-to-kubernetes-how-learning-the-internals-of-your-stack-empowers-you-as-a-developer-3d0cdf7cb7ec?gi=3b548439d3ac"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"nathan_tarbert"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["cloud"],"value":"Exploring Cloud Hexagonal Design with Wing"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["gmi"],"value":"https://itnext.io/hello-winglang-hexagon-6f2bdb550f37?gi=b7df0c3ecb24"}},"_tags":["story","author_nathan_tarbert","story_39228327"],"author":"nathan_tarbert","created_at":"2024-02-02T13:32:45Z","created_at_i":1706880765,"num_comments":0,"objectID":"39228327","points":1,"story_id":39228327,"title":"Exploring Cloud Hexagonal Design with Wing","updated_at":"2024-09-20T16:15:07Z","url":"https://itnext.io/hello-winglang-hexagon-6f2bdb550f37?gi=b7df0c3ecb24"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"anupamchugh"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["cloud"],"value":"Heyoo \u2013 An Open-Source Python Wrapper for WhatsApp Cloud API"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["gmi"],"value":"https://betterprogramming.pub/programming-whatsapp-is-now-even-easier-for-python-developers-e1a4343deed6?gi=23b1009fb82c"}},"_tags":["story","author_anupamchugh","story_31475594"],"author":"anupamchugh","created_at":"2022-05-23T05:27:06Z","created_at_i":1653283626,"num_comments":0,"objectID":"31475594","points":1,"story_id":31475594,"title":"Heyoo \u2013 An Open-Source Python Wrapper for WhatsApp Cloud API","updated_at":"2024-09-20T11:13:13Z","url":"https://betterprogramming.pub/programming-whatsapp-is-now-even-easier-for-python-developers-e1a4343deed6?gi=23b1009fb82c"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"rsecora"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["cloud"],"value":"Snaring the Bad Folks / Cloud security is a hard problem"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["gmi"],"value":"https://netflixtechblog.com/snaring-the-bad-folks-66726a1f4c80?gi=b8d807cbea97"}},"_tags":["story","author_rsecora","story_29563970"],"author":"rsecora","created_at":"2021-12-15T10:13:08Z","created_at_i":1639563188,"num_comments":0,"objectID":"29563970","points":1,"story_id":29563970,"title":"Snaring the Bad Folks / Cloud security is a hard problem","updated_at":"2024-09-20T10:05:54Z","url":"https://netflixtechblog.com/snaring-the-bad-folks-66726a1f4c80?gi=b8d807cbea97"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"coldstartops"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["gmi","cloud"],"value":"We have built KeibiDrop because me and the people close to me were tired of cloud storages, and to copy data fully across our devices in order to use it.
Now with KeibiDrop you can mount and stream folders between your computers (and web browser) access and edit data on-demand as if it was local.
You can try it instantly in the browser https://web.keibidrop.com/ (you need two devices) or by building from source/ downloading from the github releases. (there are also demo recordings in the Readme on github).
We also got a marketing website for it here: https://keibidrop.com
The desktop app, engine and CLI are FOSS AGPL-v3 and in terms of architecture and stack the following holds:
- It uses a full-duplex P2P IPv6 connection between peers with a fallback to a data forwarding relay.
- We got a dual networking stack QUIC + TCP using protocol buffers (WebRTC to be added), and we use ML-KEM-1024 + X25519 for the handshake and either aes-gcm-256 or chacha20-poly1305 for the session, and we do a latency hiding re-key inspired by Signal's SPQR.
- it relies on FUSE, macFUSE, and WinFsp as filesystem drivers and the whole gist of it is to "hide the latency" of remote filesystems and to just give the feeling of teleporting files.
- we use golang for the engine, and rust + slint.dev for the GUI
Our goal was to have a lightweight, cross platform filesystem that does not use many resources, has the RAM usage bounded and gives the feeling as if working on a local filesystem.
Currently the filesystem respects POSIX semantics, (except for symlinks and hardlinks, I have them disabled), and we tested it on intercontinental links (between Romania and Singapore), on both relayed connections and direct connections.
You can use git, pijul, Premiere Pro, DaVinci Resolve or any tools you would normally use on your local computer, or use our docker image on your NAS storage and instantly access the data you need while you are travelling."},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: KeibiDrop \u2013 Work on remote large files on-demand (no upload/download)"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/KeibiSoft/KeibiDrop/"}},"_tags":["story","author_coldstartops","story_50180280","show_hn"],"author":"coldstartops","created_at":"2026-10-21T13:06:09Z","created_at_i":1792587969,"num_comments":0,"objectID":"50180280","points":4,"story_id":50180280,"story_text":"We have built KeibiDrop because me and the people close to me were tired of cloud storages, and to copy data fully across our devices in order to use it.
Now with KeibiDrop you can mount and stream folders between your computers (and web browser) access and edit data on-demand as if it was local.
You can try it instantly in the browser https://web.keibidrop.com/ (you need two devices) or by building from source/ downloading from the github releases. (there are also demo recordings in the Readme on github).
We also got a marketing website for it here: https://keibidrop.com
The desktop app, engine and CLI are FOSS AGPL-v3 and in terms of architecture and stack the following holds:
- It uses a full-duplex P2P IPv6 connection between peers with a fallback to a data forwarding relay.
- We got a dual networking stack QUIC + TCP using protocol buffers (WebRTC to be added), and we use ML-KEM-1024 + X25519 for the handshake and either aes-gcm-256 or chacha20-poly1305 for the session, and we do a latency hiding re-key inspired by Signal's SPQR.
- it relies on FUSE, macFUSE, and WinFsp as filesystem drivers and the whole gist of it is to "hide the latency" of remote filesystems and to just give the feeling of teleporting files.
- we use golang for the engine, and rust + slint.dev for the GUI
Our goal was to have a lightweight, cross platform filesystem that does not use many resources, has the RAM usage bounded and gives the feeling as if working on a local filesystem.
Currently the filesystem respects POSIX semantics, (except for symlinks and hardlinks, I have them disabled), and we tested it on intercontinental links (between Romania and Singapore), on both relayed connections and direct connections.
You can use git, pijul, Premiere Pro, DaVinci Resolve or any tools you would normally use on your local computer, or use our docker image on your NAS storage and instantly access the data you need while you are travelling.","title":"Show HN: KeibiDrop \u2013 Work on remote large files on-demand (no upload/download)","updated_at":"2026-10-22T05:20:50Z","url":"https://github.com/KeibiSoft/KeibiDrop/"},{"_highlightResult":{"author":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["cloud"],"value":"clouddrover"},"title":{"matchLevel":"none","matchedWords":[],"value":"For your eyes only: improving Netflix video quality with neural networks (2022)"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["gmi"],"value":"https://netflixtechblog.com/for-your-eyes-only-improving-netflix-video-quality-with-neural-networks-5b8d032da09c?gi=65bdf98dcfbf"}},"_tags":["story","author_clouddrover","story_38255587"],"author":"clouddrover","created_at":"2023-11-13T21:22:31Z","created_at_i":1699910551,"num_comments":0,"objectID":"38255587","points":3,"story_id":38255587,"title":"For your eyes only: improving Netflix video quality with neural networks (2022)","updated_at":"2024-09-20T15:38:58Z","url":"https://netflixtechblog.com/for-your-eyes-only-improving-netflix-video-quality-with-neural-networks-5b8d032da09c?gi=65bdf98dcfbf"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"borissunik"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["gmi","cloud"],"value":"Just citations.
Dario Amodei (CEO of Anthropic, 2025): "People outside the field are often surprised and alarmed to learn that we do not understand how our own AI creations work. This lack of understanding is essentially unprecedented in the history of technology."
MIT Technology Review: "Despite all the stunning success, nobody really knows how \u2014 or why \u2014 it works."
Yoshua Bengio (Turing Award laureate, 2024): "If we continue as before, we are driving against the wall. Or at least we are driving into a fog, without knowing whether there is a wall."
Mathematician Roman Vershynin (UC Irvine), in a recent interview \u2014 "We advanced to the wall and hit it."
So why did AI research hit the wall?
The answer is not hard to find, and most people in the field know it, even if they prefer not to say it loudly. The point is that all training data used by AI is secondary, while the primary source is the human being. Every text, every image, every pattern a neural network absorbs \u2014 all of it was created by humans. A neural network does not discover the laws of the world; it only finds regularities in what has already been created by human intelligence.
So if we want to understand how AI forms knowledge, we must first understand how humans form knowledge. But this pretty obvious idea was never materialized because of the historical failure of all attempts to build a working, mathematically correct model of knowledge. The story started in the 19th century and continued until the end of the 20th. Tens of theories of information, hundreds of books, thousands of articles in dozens of journals \u2014 and nothing practically working came out of it.
This however does not mean no adequate theory exists. It is there since the late 2000s, but it is still completely unknown to the scientific community. This occurs for a purely ideological reason because this theory builds its formalizations on concepts borrowed from mainstream programming languages and so completely ignores mathematics.
For the major journals devoted to AI and information theory that is a taboo. I spent years trying to publish it. The answer was always the same: "Add the mathematical justification, and then we will take a look." When I contacted established researchers directly, the answer was either silence or: "Yes, you have an original idea, but we are not interested."
I stopped trying fifteen years ago. My hope is that now the theory might have a chance, because it addresses exactly the problems that the current conceptual apparatus cannot solve.
The Theory of Meaningful Information (TMI) expands the conceptual framework of the most expressive programming language, which is C++, by adding capabilities for creating standard narrative (non-executable) texts that constitute the primary form of any human language.
In TMI, information is defined as the value of a variable used in some algorithm. This works for a bit in a computer, a state of a neuron, a pressed button, a written letter. It applies to a bacterium, a program, a neural network, and a human.
The theory introduces Information Capable Entities (ICE) and traces their evolution from the simplest biological ICE \u2014 a cell \u2014 to complex technical devices and the human being, who is modeled as an object controlled by a biocomputer (the brain).
It offers a complete algorithm of human knowledge formation, from the simplest perceptions to the most complex concepts. It contains the theory of intelligence, covering both natural and artificial intelligence, and the theory of a universal representation language.
A short description of TMI and a link to the complete work:
https://doi.org/10.5281/zenodo.23403895
Boris Sunik,\nindependent researcher"},"title":{"matchLevel":"none","matchedWords":[],"value":"Why the Creators of AI Do Not Understand Their Own Creations"}},"_tags":["story","author_borissunik","story_50132747","ask_hn"],"author":"borissunik","created_at":"2026-10-17T16:54:50Z","created_at_i":1792256090,"num_comments":0,"objectID":"50132747","points":4,"story_id":50132747,"story_text":"Just citations.
Dario Amodei (CEO of Anthropic, 2025): "People outside the field are often surprised and alarmed to learn that we do not understand how our own AI creations work. This lack of understanding is essentially unprecedented in the history of technology."
MIT Technology Review: "Despite all the stunning success, nobody really knows how \u2014 or why \u2014 it works."
Yoshua Bengio (Turing Award laureate, 2024): "If we continue as before, we are driving against the wall. Or at least we are driving into a fog, without knowing whether there is a wall."
Mathematician Roman Vershynin (UC Irvine), in a recent interview \u2014 "We advanced to the wall and hit it."
So why did AI research hit the wall?
The answer is not hard to find, and most people in the field know it, even if they prefer not to say it loudly. The point is that all training data used by AI is secondary, while the primary source is the human being. Every text, every image, every pattern a neural network absorbs \u2014 all of it was created by humans. A neural network does not discover the laws of the world; it only finds regularities in what has already been created by human intelligence.
So if we want to understand how AI forms knowledge, we must first understand how humans form knowledge. But this pretty obvious idea was never materialized because of the historical failure of all attempts to build a working, mathematically correct model of knowledge. The story started in the 19th century and continued until the end of the 20th. Tens of theories of information, hundreds of books, thousands of articles in dozens of journals \u2014 and nothing practically working came out of it.
This however does not mean no adequate theory exists. It is there since the late 2000s, but it is still completely unknown to the scientific community. This occurs for a purely ideological reason because this theory builds its formalizations on concepts borrowed from mainstream programming languages and so completely ignores mathematics.
For the major journals devoted to AI and information theory that is a taboo. I spent years trying to publish it. The answer was always the same: "Add the mathematical justification, and then we will take a look." When I contacted established researchers directly, the answer was either silence or: "Yes, you have an original idea, but we are not interested."
I stopped trying fifteen years ago. My hope is that now the theory might have a chance, because it addresses exactly the problems that the current conceptual apparatus cannot solve.
The Theory of Meaningful Information (TMI) expands the conceptual framework of the most expressive programming language, which is C++, by adding capabilities for creating standard narrative (non-executable) texts that constitute the primary form of any human language.
In TMI, information is defined as the value of a variable used in some algorithm. This works for a bit in a computer, a state of a neuron, a pressed button, a written letter. It applies to a bacterium, a program, a neural network, and a human.
The theory introduces Information Capable Entities (ICE) and traces their evolution from the simplest biological ICE \u2014 a cell \u2014 to complex technical devices and the human being, who is modeled as an object controlled by a biocomputer (the brain).
It offers a complete algorithm of human knowledge formation, from the simplest perceptions to the most complex concepts. It contains the theory of intelligence, covering both natural and artificial intelligence, and the theory of a universal representation language.
A short description of TMI and a link to the complete work:
https://doi.org/10.5281/zenodo.23403895
Boris Sunik,\nindependent researcher","title":"Why the Creators of AI Do Not Understand Their Own Creations","updated_at":"2026-10-22T19:27:07Z"}],"hitsPerPage":50,"nbHits":35,"nbPages":1,"page":0,"params":"query=GMI+Cloud&tags=story&hitsPerPage=50&advancedSyntax=true&analyticsTags=backend","processingTimeMS":16,"processingTimingsMS":{"_request":{"queue":40,"roundTrip":16},"afterFetch":{"format":{"highlighting":1,"total":1},"merge":{"mergeLoop":{"prepareNextHit":2,"total":2},"total":2},"total":2},"fetch":{"query":3,"scanning":8,"total":12},"total":16},"query":"GMI Cloud","serverTimeMS":58}