Hacker News Reader

Comment by mewpmewp2 | Hacker News Reader

I'm wondering if there's going to be an initiative for a new programming language that is specifically intended for AI usage. E.g. AI has some statistical patterns built in that it would naturally guess what the method names should be or the API package names, and then we would just invent a programming language based on what is most natural for AI for any sort of use case. So even if AI "hallucinates" it would likely be the right guess since all the API and syntax is picked by AI in the first place.

View on HN

A related concept is LLM-driven API design. First write the comment of what the API function should do, then let the LLM write the function call, it will pick the most expected naming and argument order, then you write the API using this function prototype.

And maybe if the LLM sometimes tries to use different naming, you could add an overload of the same API end point, as long as it doesn't clash with an existing one. So in the end for anyone using your library with an LLM it would appear to just work.

I experienced this in my automated GPT-4 programmer project. This was 1.5 ago back when prompt size was very limited. I couldn't put a large programs entire source code in the prompt so I auto-converted it to stubs.

I noticed that letting GPT-4 write more of the code had the advantage that it understood it better and less context was necessary.

Now when I design systems for GPT-4 itself to use, I just let GPT-4 design them. Ask it to use an imaginary system, and what it comes up with (consistently) should be the API.

Once AI is fully writing software autonomously and humans don’t need to see that code anymore, then why not just have AI write iteratively more efficient assembly? (Meaning make the software work then turn it over to an optimization agent that can drive out all inefficiencies until perfection.)

Those efficiencies will eventually be driven into all parts of the system. AI will be able to run the most efficient code on the most efficiently designed processors, that can be designed with the knowledge that only AI is going to use them. And then we can remove ALL the weird abstractions and accommodations we had to make for human brains.

I’m not saying it will happen tomorrow but language choice is just a temporary concern.

Since when is AI code efficient?

It’s based on human code, most of that isn’t efficient.

That sounds like the inverted world though, as in, AI is imperfect and makes mistakes, let's try and work with and around those mistakes instead of making AI better.

I think it makes sense to go at this problem from multiple different angles. One is making the AI better, the second angle is making the World easier for AI. We learn from both approaches. And you can try to make a really good AI that is very good with the current World, but at the same time an AI that is maybe much faster and efficient will still be very valuable if there's a World that is made easier for them. Because there will always be a trade off to an extent with accuracy vs speed.

It could be even something that the larger and slower AI model is defining and implementing the APIs and smaller and faster is the one who will actually make use of those.

Or some variations of those combined, where it's a single AI model with a capability to select between the 2. If it's complicated APIs, it will try to crawl the docs for accuracy and make "AI intuitive" wrappers around all of the APIs and their edge cases, and making it easier for the fast model.

Doesn't the AI do that pretty well already? Not sure what other language features you would need to improve on the way the AI writes code.

Some LLMs do it well, some do it worse, and are better at certain things, worse at others. I think there's a balance to be found, e.g. you could have a very small and fast model perform better if the coding paradigm was specifically dedicated to the AI.

E.g. right now I think AI is excellent with JavaScript/TypeScript, but worse with some other languages like e.g. Rust or Scala.

I would be very surprised if there aren't any serious efforts on this already underway.

Reply on news.ycombinator.com