4 ms·

Ask it to “reason through” a problem and then ask it to give you an answer. How’s that different from thinking?

by aqme28 4y ago

scotty79 4y ago
It's just wrong. That's how you can tell. Actual reasoning leads to sensible conclusions.
- aqme28 4y ago
  Coming to the wrong conclusion doesn’t mean I wasn’t thinking through the problem.
  - scotty79 4y ago
    It definitely means that it was thinking wrongly if at all. Just talk to GPT about math. You'll quickly change your mind about the possibility of it thinking.
    - stevenhuang 4y ago
      LLMs are bad at arithmetic due to tokenization limitations but they're actually pretty decent at mathematical reasoning. You don't know what you're talking about I'm afraid. https://www.lesswrong.com/posts/qy5dF7bQcFjSKaW58/bad-at-arithmetic-promising-at-math https://www.lesswrong.com/posts/qy5dF7bQcFjSKaW58/bad-at-ari...
      - scotty79 4y ago
        Please just try. It's horrible at mathematical reasoning. Use just words to avoid problems with tokenization. Alternatively just read through the link you provided. It has many examples of failures of GPT and gabage it produces when talked to about math.
        stevenhuang 4y ago
        Clearly you haven't read it then. The provided example directly shows ability in mathematical reasoning by coming up with a novel concept and example case, it is just poor in arithmetic. Math is not simply arithmetic abilities, you seem unable to comprehend this.
        scotty79 4y ago
        Sure, sure. "I want it to come up with a new idea. Its first attempt was to just regurgitate the definition of the set of zero-divisors (a very basic concept), and (falsely) asserted that they formed an ideal (among other false claims about endomorphism rings)." "I tried a few more times, and it gave a few more examples of ideas that are well-known in ring theory (with a few less-than-true modifications sometimes), insisting that they are new and original." "This in particular is quite an interesting failure. " "So there we have it. A new definition. One example (of a 4-cohesive ring) extracted with only mild handholding, and another example (of a 2-cohesive ideal) extracted by cherry-picking, error-forgiveness, and some more serious handholding." "Some errors (being bad at arithmetic) will almost certainly be fixed in the fairly near future." - and this opinion is based on absolutely nothing.
KyleLewis 4y ago
When a model "reasons through" a problem its just outputting text that is statistically likely to appear in the context of "reasoning through" things. There is no intent, consideration of the options available, the implications, possible outcomes. However, the result often looks the same, which is neat
Xelynega 4y ago
"thinking" and reasoning can be done by toddlers with a dataset a fraction of a fraction of the size that even the simplest language models are trained on. I don't understand this thinking that it's x because it looks like x(thinking, artistic creativity, etc.). I can prompt Google for incrementally more correct answers to a problem, does that mean there's no difference between "google" and "thought"?

Ask it to “reason through” a problem and then ask it to give you an answer. How’s that different from thinking?

by aqme28 4y ago

Ask it to “reason through” a problem and then ask it to give you an answer. How’s that different from thinking?

scotty79 4y ago

It's just wrong. That's how you can tell. Actual reasoning leads to sensible conclusions.

aqme28 4y ago
Coming to the wrong conclusion doesn’t mean I wasn’t thinking through the problem.
- scotty79 4y ago
  It definitely means that it was thinking wrongly if at all. Just talk to GPT about math. You'll quickly change your mind about the possibility of it thinking.
  - stevenhuang 4y ago
    LLMs are bad at arithmetic due to tokenization limitations but they're actually pretty decent at mathematical reasoning. You don't know what you're talking about I'm afraid. https://www.lesswrong.com/posts/qy5dF7bQcFjSKaW58/bad-at-arithmetic-promising-at-math https://www.lesswrong.com/posts/qy5dF7bQcFjSKaW58/bad-at-ari...
    - scotty79 4y ago
      Please just try. It's horrible at mathematical reasoning. Use just words to avoid problems with tokenization. Alternatively just read through the link you provided. It has many examples of failures of GPT and gabage it produces when talked to about math.
      - stevenhuang 4y ago
        Clearly you haven't read it then. The provided example directly shows ability in mathematical reasoning by coming up with a novel concept and example case, it is just poor in arithmetic. Math is not simply arithmetic abilities, you seem unable to comprehend this.
        scotty79 4y ago
        Sure, sure. "I want it to come up with a new idea. Its first attempt was to just regurgitate the definition of the set of zero-divisors (a very basic concept), and (falsely) asserted that they formed an ideal (among other false claims about endomorphism rings)." "I tried a few more times, and it gave a few more examples of ideas that are well-known in ring theory (with a few less-than-true modifications sometimes), insisting that they are new and original." "This in particular is quite an interesting failure. " "So there we have it. A new definition. One example (of a 4-cohesive ring) extracted with only mild handholding, and another example (of a 2-cohesive ideal) extracted by cherry-picking, error-forgiveness, and some more serious handholding." "Some errors (being bad at arithmetic) will almost certainly be fixed in the fairly near future." - and this opinion is based on absolutely nothing.

KyleLewis 4y ago

When a model "reasons through" a problem its just outputting text that is statistically likely to appear in the context of "reasoning through" things. There is no intent, consideration of the options available, the implications, possible outcomes. However, the result often looks the same, which is neat

Xelynega 4y ago

"thinking" and reasoning can be done by toddlers with a dataset a fraction of a fraction of the size that even the simplest language models are trained on. I don't understand this thinking that it's x because it looks like x(thinking, artistic creativity, etc.). I can prompt Google for incrementally more correct answers to a problem, does that mean there's no difference between "google" and "thought"?