>>9841
This is too technical, and i am no longer in the habit of reading mathematical notation like this. but
>In this paper, we critically examine the COT reasoning of LLMs through the lens of data distribution,
revealing that the perceived structured reasoning capability largely arises from inductive biases shaped
by in-distribution training data. We propose a controlled environment, DataA lchemy, allowing
systematic probing of CoT reasoning along three crucial dimensions: task structure, reasoning length,
and query format. Empirical findings consistently demonstrate that CoT reasoning effectively repro-
duces reasoning patterns closely aligned with training distributions but suffers significant degradation
when faced with distributional deviations. Such observations reveal the inherent brittleness and
superficiality of current CoT reasoning capabilities. We provide insights that emphasize real-world
implications for both practitioners and researchers.
It's good to have some scientific tests to confirm this, but what the paper proposes is quite obvious if you ask me. Both in practice and analytically. ChatGPT doesn't know much about the plots of media and filmography. If you ask ChatGPT about old movies, or even popular video games, it will produce total and complete garbage.
And then if you think about it, it's obvious that the model will perform badly on things it has not been trained on. Real world logic is too complicated to be extrapolated from existing data. I get it, CoT is an attempt to give intelligence to the AI, so it can be more than just a word prediction engine. Even so, we cannot expect it to know what it has never been told.
AI is a machine/tool, and we shouldn't give it more credit than that. At least, for now. Someday, they will create a machine that is actually intelligent. For now, they're making a lot of money selling a machine they claim is "intelligent".
>>9867
I don't think the drivers by themselves were bad. It was just hard to get them configured correctly. I used to run hashcat on my machine, noveau was bad. proprietary drivers bumped up the speed properly.