Post

Given that LLMs are stochastic parrots of their training data, I assume the simple reason is that LLMs have no idea about machine code. Also the output is not reliable enough to omit the human in the loop, so do we really want that?

As opposed to the humans who are strictly logical? ;) I think we're beyond the point where we can claim that humans are superior at writing code. Also, doing this doesn't mean we would exclude testing.

I think it can[1] but we don’t do it because: 1) we don’t trust the LLM enough. We want to review the code. 2) high level languages give you a higher density of expression per token. i.e. it takes less tokens so you get faster answers [1] chatgpt.com/share/674db3...

1) isn't relevant. You're not speaking for everyone and there are plenty of people that trust the output from LLMs 2) might be relevant. I don't know enough of the limitations around LLMs to determine that. However, I can get it to write assembly code.

Yeah I get what you’re saying. But I’d caution against dismissing people because they don’t speak for _everyone_. I am an expert 😂 and while I trust LLMs for many things, me and most of my friends very much would not trust an LLM machine code output.

Yeah I think this will happen at some point. If you need to debug, you just convert from machine code to whatever language you want. Meta has this: ai.meta.com/research/pub...