Inductive or Deductive? Rethinking the Fundamental Reasoning Abilities of LLMs

Re:Inductive or Deductive? Rethinking the Fundamental Reason... by godelski on Monday September 02, 2024 @01:13AM (#41422782)

  > They are statistical text generators, whose results are defined by their training data set

Honestly I'm pissed at the research community. It's fraud. If you don't know what's in the training data you simply cannot differentiate reasoning from memorization.

Beyond that, the experiments are almost certainly in the training data. Like come on! I feel like I'm going crazy here. How can any coder not think LLMs are training on oct and hex‽

https://news.ycombinator.com/item?id=41422751

[ Parent | on HN ]

Re:Inductive or Deductive? Rethinking the Fundamental Reason... by og_kalu on Monday September 02, 2024 @09:04AM (#41425076)

Of course they've been trained on oct and hex.

The question is would the results be largely valid if this was done on a human who had learnt how to perform base 8, 9, 11 etc arithmetic instead ?

I mean, they're clearly not trying to test the ability to derive base arithmetic from scratch.

[ Parent | on HN ]