Hacker News Reader

Comment by uoaei | Hacker News Reader

I don't think the commenter's understanding of LLMs is relevant here. It's what the authors of TFA think LLMs are for and, arguably more importantly, what kinds of problems (nails) they think about when presented with the idea of applying LLMs (hammer).

It's a bit ridiculous that this is the best they can come up with, and I don't really think the idea is worth defending, especially by telling other people that they just don't get it.

View on HN

His/Her primary assertion is that it is impossible for a LLM to generate novel recipes without physically tasting it first. His/her assertion is wrong and one easily corrected by understanding a LLM is and does.

It's not a perfect predictor so it's not impeccable for all conceivable situations but GPT-4 can already generate novel recipes that taste nice. This is not some far flung science fiction ability.

Salt to taste, cook until brown, kneed until the dough is no longer sticky, everything about cooking is tasting and feeling. Moreover everybody has differences in their tastes and consistency preferences. Good recipe books teach you how to use ingredients and apply that knowledge to recipes and how to modify those recipes to your personal preferences. Sure an LLM can generate novel recipes but without the premise of taste, texture and food pairing tuned in to personal preference, the novel recipe is a crap shoot in the dark. The reality is, the more you learn to cook, the less you need recipes and a good cookbokk will teach you alot more than a pile of recipes.

This is true. On the other hand, the recipes aren't typically good, and generating novel ones isn't really what people mostly want to do... so the value seems limited.

That's part of it yes but people tend to forget that the road to capable prediction is not instantaneous.

It's possible the data is enough to get a good predictor but the one you tested simply wasn't good enough yet.

Possibly, but the easily available data itself seems is pretty poor for systemic reasons.

Reply on news.ycombinator.com