Hacker Newsnew | past | comments | ask | show | jobs | submit | ianand's commentslogin

In that case I humbly suggest my talk from AI Engineer World's Fair https://www.youtube.com/watch?v=ZuiJjkbX0Og

Gives you the basics on LLM internals in about 90 minutes and includes an already built model in JavaScript that you can step through in browser devtools to get as detailed as you want.


Thanks for linking to it! Just in time for my own learner's journey.


It's like "GPT is all you need for the backend" [1] on steroids

[1] https://news.ycombinator.com/item?id=34503418


The future of programming and IT tech - If you need a database, you just say to the model "you're a database". Or "you're a CRM", "you're a Doom game", "you're a Word processor", etc.

It was "network is the computer", now it will be "model is the computer", and the model will be like one large ("multi tenant" - it will know on its own how/when to separate tenants' data and when to analyze it all together) model living on tens/hundreds of millions of nodes in AWS ... the AWS itself will be just that model.


And when the datacenter staff show up for work every morning the AI will them “You are employee #5378. Today you are a janitor. You will…” going off on a long list of hyper precise instructions for them to follow, like a human prompt.


Since you mention serverless it might be worth mentioning firecracker and v8 isolates.


Or CGIs running on httpd inside HP-UX Vaults, that is how old the idea happens to be.


> how old the idea happens to be

TFA is missing a host of many a popular isolation techniques like Isolates, Code Interp / Binary Translators [0], Enclaves, Exclaves, Domains/Worlds, (RISC V) SEEs, TEEs, SEs, HSMs, pKVMs ...

[0] https://news.ycombinator.com/item?id=38950949


Thank you for the feedback. I will definitely add them as example solutions for serverless.


I'm not a fan of the database lookup analogy either.

The analogy I prefer when teaching attention is celestial mechanics. Tokens are like planets in (latent) space. The attention mechanism is like a kind of "gravity" where each token is influencing each other, pushing and pulling each other around in (latent) space to refine their meaning. But instead of "distance" and "mass", this gravity is proportional to semantic inter-relatedness and instead of physical space this is occurring in a latent space.

https://www.youtube.com/watch?v=ZuiJjkbX0Og&t=3569s


Then I think you’ll like our project which aims to find the missing link between transformers and swarm simulations:

https://github.com/danielvarga/transformer-as-swarm

Basically a boid simulation where a swarm of birds can collectively solve MNIST. The goal is not some new SOTA architecture, it is to find the right trade-off where the system already exhibits complex emergent behavior while the swarming rules are still simple.

It is currently abandoned due to a serious lack of free time (*), but I would consider collaborating with anyone willing to put in some effort.

(*) In my defense, I’m not slacking meanwhile: https://arxiv.org/abs/2510.26543 https://arxiv.org/abs/2510.16522 https://www.youtube.com/watch?v=U5p3VEOWza8


This is an excellent analogy! Thank you!


The site’s domain name is the best use of a .fail tld ever.


OT from TFA, so high jacking your thread …

I don’t recall if there was ever a difference between “abort” and “fail.” I could choose to abort the operation, or tell it … to fail? That this is a failure?

¯\_(ツ)_/¯


Take reading a file from disk.

Abort would cancel the entire file read.

Retry would attempt that sector again.

Fail would fail that sector, but the program might decide to keep trying to read the rest of the file.

In practice abort and fail were often the same.


Makes sense. Maybe I ran across a proper use a time or two back then and just don’t remember. But the two being the same was the overwhelming experience.


As the guy who did GPT2 in Excel, very cool and kudos!!

Curious why you chose WebGL over WebGPU? Just to show it can be done?

(Also see my other comment about fetching weights from huggingface)


This was a final project for a graphics class where we used WebGL a lot. Also I was just more familiar with OpenGL and haven't looked that much into webGPU


Someone needs to implement Excel using graphics shaders now.



Probably because WebGPU support is still rather iffy.


> Curious why you chose WebGL over WebGPU? Just to show it can be done?

For a WebGPU implementation, one can use transformers.js directly (or many other libraries actually), maybe WebGL is more original.

[1]: https://huggingface.co/docs/transformers.js/index


Transformers.js wraps the ONNX runtime which is rather versatile (WASM, WebGL, WebGPU, and WebNN). It's not the backend that makes it novel.


ianand, I immediately thought of you when I saw this post. Miss you friend.


Dude, been forever. Thanks. Will DM you.


Checkout https://github.com/jseeio/gpt2-tfjs fetches the weights for GPT2 from huggingface on the fly.


Sounds like an interesting masters thesis. Is your masters thesis available online somewhere?


Well, not sure about the final doc that went to the university, but this is the almost final draft.

https://docs.google.com/document/d/e/2PACX-1vSyWbtX700kYJgqe...

Since its in Cyrillic you should perhaps use a translation service. There are some screens showing results, though as I was really on a tight deadline, and its not a PHD but masters thesis, I decided to not go into in-depth evaluation of the proposed methodology against SPIDER (https://yale-lily.github.io/spider). Even though you can find the simplifed GBNF grammar, also some of the outputs. The grammar, interestingly it benefits/exploits a bug in llama.cpp which allows some sort of recursively-chained rules. Bibliography is in English, but really - there is so much written on the topic, by no means comprehensive.

Sadly no open inference engine (at time of writing) was both good enough in beam search, and grammars, so this whole things needs to perhaps be redone in pytorch.

If I find myself in a position to do this for commercial goals, I'd also explore the possibility of having human-catered SQLs against the particular schema, in order to guide the model better. And then do RAG on the DB for more context. Note: I'm already doing E/R model reduction to the minimal connected graph which includes all entities of particular interest to the present query.

And finally, since you got that far - the real real problem with restricting LLM output with grammars is the tokenization. Because all parsers work reading one char at a time, and tokens are very often few chars, so the parser in a way needs to be able to "lookahead", which it normally does not. I believe OpenAI wrote they realized this also, but I can't really find the article atm.


Thanks. Took a quick look and definitely needed to use Google Translate but seems to have worked to get the gist of it.


Fun fact: A decade ago the designer of HAML and Sass created a modern alternative to XSLT. https://en.wikipedia.org/wiki/Tritium_(programming_language)


The model architecture is the same during RL but the training algorithm is substantially different.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: