DEMO TIME SupGen is a generative coding AI... except it isn't an AI. There is no model, there is no pre-training. You just give it some examples, and it gives you a program. It runs locally, in a single-core CPU. Oh, and it can also prove theorems. Here's a... demo, including a brief TT intro. (The synth examples start at the 6 min mark.)show more

I apologize for reposting. I fixed a typo on the RH, and audio issues. This tool has been highly inspired by Edwin Brady's work on type-driven synthesis for Idris2. I always found his examples mind-blowing - it was GPT before GPT - and I think this technique is criminally slept on. HVM gives it a relevant boost. Some notes: - This technique is old. The only novelty is efficiency. - SupGen is up to 1000x faster than alternatives! - Yet, the field has given up on exhaustive search. - People shifted to LLMs, hoping they'll break the exponential. - I think faster raw speed + big scale + learning will do it. - Nobody has built a Symbolic Transformer, so I will.

AI won't replace you, but a person using AI will. Join 500,000+ readers and learn how to use AI in just 5 minutes a day (for free).

HN Thread:

Huge, glad to see all the work in the background showing off big time (again, reposting my answer on a repost 😎)

if it wasn't messy, it wouldn't be me (I'd rather have it be correct than viral)

This means you solved p=np ? (Just kidding) Freaking awesome

deleting again

One thing I didn’t understand: When you have a sample of a function, such as f(1) = 2, f(2) = 4, etc., how does it determine the doubling formula? There are infinitely many functions that pass through those points, but are different from the doubling function.

It just finds the shortest solution, which, per Ocamm's Razor, is very likely to be the most general implementation. Note that if you want to make sure that's the case, you can write precise dependent types instead of examples - which is a good idea, because it makes it faster.

I swear I watched this yesterday ?

(we are so back!)