4 ms·
did you not read https://arxiv.org/pdf/2410.05229 https://arxiv.org/pdf/2410.05229 "We hypothesize that this decline is due to the fact that current LLMs are n
by turtlesdown11 2y ago
did you not read https://arxiv.org/pdf/2410.05229 https://arxiv.org/pdf/2410.05229
"We hypothesize that this decline is due to the fact that current LLMs are not capable of genuine logical reasoning; instead, they attempt to replicate the reasoning steps observed in their training data."
- soulofmischief 2y agoThey've presented a hypothesis warranting further study. Here is an article where a model's learned ability to carry out modular addition is recovered. https://arxiv.org/abs/2301.05217 https://arxiv.org/abs/2301.05217 As you can see, the jury's still out.