Hacker News Reader

Comment by chaxor | Hacker News Reader

What's 375792*7385482? You can't use paper, or think about it more than a second. If you get it wrong, it's clear that you are not intelligent.

See how ridiculous that sounds?

CoT or <work> tokens are not 'cheating' - they allow the system to have (as an analogy) a sheet of paper to work out the problem, instead of relying on a single step to jump through thousands of steps of reasoning. It's trained on the algorithmic reasoning embedded within natural language (Transformers<=>GATs+PE⊂GNNs, which are dynamic programmers, and align well with algorithmic reasoning), so the steps are necessary to lay out in as much detail.

View on HN

The answer is 2775405051744.

I cheated, and used bc, the unix utility.

These three maneuvers (the correct answer, the agency shown by cheating, and the crystallized intelligence of unix) handily demonstrate both my own intelligence, and that of my species.

The LLMs aren't in the same category of being, yet. It's not clear they ever will be; while I think silicon is a viable substrate for intelligent agents, it still strikes me as doubtful they'll be any more descended from the current machine learning systems as they are from Symbolics Lisp.

Until an AI decides to take control of a robot, physically sequestrate a human, and under the threat of physical pain make the human perform the tasks the AI can't...

> It's not clear they ever will be;

It seems to me that the ability to reason ( though primitive it may still be) is an emergent property of sufficiently large differentiable hash/lookup tables (and if you think about the mechanism of reasoning, that seems... Reasonable).

According to "folks in the know", the scaling limits of GPT architecture seem to be nigh, but if it were a bit higher I wouldn't be surprised if LLMs would have made it.

I think you’re conflating reasoning with inference. And in any case, GP wasn’t using reasoning, but rather resourcefulness/invention. I’m not sure an LLM can pull that off beyond coming up with vague, derivative fiction.

Gp said "state of being" so I don't think the either of us can know for sure what was meant by that.

Are you sure? Have you read through enough literature on GNNs? Transformers are GNNs with attention message passing and positional embeddings. GNNs have also been *mathematically proven * to be dynamic programmers and to *align with algorithmic reasoning*. So tell me again how LLMs aren't capable of performing reasoning to be intelligent.

with a bit of glue code I can allow gpt4 to use those tools. and decide to do so. cheating is part of 'intelligence'.

I know I’m bad at multiplying six and seven digit integers in my head, and I don’t care at all if that makes me more or less intelligent (whatever the fuck that means) than a wrist watch.

What does that possibly have to do with GPT-4 shitting the bed if you draw the logic puzzle out of a deck of cards instead of cherry-picking and hinting it?

My point is that there are already mathematical proofs in the literature that show that LLMs literally are doing algorithmic reasoning. This has been proven by category theory, in the area of GNNs (for which transformers are a trivial subset). Sorry if I'm going over too much basics, as this is obvious and well understood by anyone actually in the field, to the point of being excessively boring.

Reply on news.ycombinator.com