3 ms·

While comparing ordinary human activity to language modeling is perhaps the greatest intellectual cliche of 2024, what PG describes in the essay-writing process

While comparing ordinary human activity to language modeling is perhaps the greatest intellectual cliche of 2024, what PG describes in the essay-writing process itself essentially boils down to beam search. Autocomplete, find wrongness, backtrack, predict more consistent series of next tokens. The fact that only one of those options is actually reflected on the screen doesn't change the fact that the writer is probability-weighting continuations in their head. I remember running across a tweet that said, in essence, you always have access to your own optimal policy if you simply ask. A similar technique might be the key to wresting good ideas from an ambiguous starting point: "What would be the most consistent answer to the original question, given what I've learned and written so far?"

Der_Einzige 3y ago
Maybe that's the process PG describes themselves, but I think that the amount of different "processes" or "human equivalent to LLM decoding/sampling techniques" used by humans is likely very large - far larger than the number of techniques available to LLMs today. Today within LLMs, there typicality sampling, contrastive search, top-p/top-k (nucleus sampling), and a whole massive amount of other more obscure techniques which didn't make it into huggingfaces model.generate() function. I believe that the ontological configuration of humans has wide variety. As such, I believe that humans go through a diverse and vibrant amount of different writing processes - and often radically different processes can still result in excellent works! Specifically, beam search is really only good at dealing with sequence-sequence tasks, i.e. summarization or translation. It's also horribly, horribly inefficient in the worst case scenarios. That said, it's also the only good way we know of for enforcing sequence level constraints (i.e. like this https://huggingface.co/blog/constrained-beam-search https://huggingface.co/blog/constrained-beam-search)

Der_Einzige 3y ago

Maybe that's the process PG describes themselves, but I think that the amount of different "processes" or "human equivalent to LLM decoding/sampling techniques" used by humans is likely very large - far larger than the number of techniques available to LLMs today. Today within LLMs, there typicality sampling, contrastive search, top-p/top-k (nucleus sampling), and a whole massive amount of other more obscure techniques which didn't make it into huggingfaces model.generate() function. I believe that the ontological configuration of humans has wide variety. As such, I believe that humans go through a diverse and vibrant amount of different writing processes - and often radically different processes can still result in excellent works! Specifically, beam search is really only good at dealing with sequence-sequence tasks, i.e. summarization or translation. It's also horribly, horribly inefficient in the worst case scenarios. That said, it's also the only good way we know of for enforcing sequence level constraints (i.e. like this https://huggingface.co/blog/constrained-beam-search https://huggingface.co/blog/constrained-beam-search)