5 ms·

RIP google. There's a joke that the best place to hide a murder is on page 2 of google search. The fact is GPT has at bare minimum parsed the top page of google

RIP google. There's a joke that the best place to hide a murder is on page 2 of google search. The fact is GPT has at bare minimum parsed the top page of google and reasoned about the top results to give the best answer with as much context of why it reasoned so. GPT is basically a superset of google with added reasoning. I can't see it giving worse results than Google. Both will produce a bogus answer if you ask a bogus question, but GPT will never produce a bogus answer when google has the correct one.

jchmrt 4y ago
I think it's really problematic to use the word "reason" to describe what GPT does. As others have commented in this thread, GPT is a language model, which in no way reasons about anything it writes. It is simply writing whatever fits statistically to the prompt and context. Attributing reasoning or too much intelligence to this could prove quite dangerous.
- astrange 4y ago
  > As others have commented in this thread, GPT is a language model, which in no way reasons about anything it writes. It is simply writing whatever fits statistically to the prompt and context. Those people are wrong; statistics are a kind of reasoning (intuition) and if you try it, it obviously does the other kinds of reasoning too. I have all kinds of answers from it that are logically coherent and can’t possibly have appeared on the internet already. (Try asking it to write new SCP wiki entries.)
  - jchmrt 4y ago
    Just because some of its answers are logically coherent due to it following statistical patterns, does not mean that the model is reasoning. Reasoning implies some type of discipline which prevents nonsensible arguments (even in uncommon situations).
    - astrange 4y ago
      No such thing as “is reasoning”, only “does reasoning”. Are you an unreasoning being if you make a math mistake? You can see it “does” reasoning if you say “think step by step” after a question. It can attend over its steps then.
tsimionescu 4y ago
> but GPT will never produce a bogus answer when google has the correct one. What are you basing this assumption on? GPT can always generate a bogus answer, because its only notion of whether an answer is good or not is whether your prompt + that answer are statistically similar to text in its training corpus. The more text from the training corpus it resembles the better GPT will think the answer is. For more obscure topics, this means that the chances of it hallucinating based on similar sounding much more popular topics increases greatly. This way of working also means that facts that appear more in the training corpus have a higher chance of being part of the model's outputs. For example, if the training corpus was not carefully curated to exclude vaccine conspiracy theories, there would be a good chance that the answer GPT gives to "autism is caused by ___" would be "vaccines", as this would likely be the most common answer in its training corpus. In contrast, Google's algorithms have more sophisticated biasing especially for these kinds of sensitive topics to ensure that more high-confidence sources are likelier to be top results, even if they are a minority of all the pages on the internet on that topic.
- astrange 4y ago
  > GPT can always generate a bogus answer, because its only notion of whether an answer is good or not is whether your prompt + that answer are statistically similar to text in its training corpus. “Statistically similar” is doing a ton of work even for the original GPT model - just printing the letter “e” is “statistically similar” to English text, if it can even be grammatically correct you should be surprised. And it’s certainly not true for the last two generations (InstructGPT and ChatGPT) which weren’t trained with that objective at all but used human raters.
noitpmeder 4y ago
This comment is based on such a large amount of BS. Reasoning over facts, as a concept, is completely divorced from the function of these GPT models.

jchmrt 4y ago

I think it's really problematic to use the word "reason" to describe what GPT does. As others have commented in this thread, GPT is a language model, which in no way reasons about anything it writes. It is simply writing whatever fits statistically to the prompt and context. Attributing reasoning or too much intelligence to this could prove quite dangerous.

astrange 4y ago
> As others have commented in this thread, GPT is a language model, which in no way reasons about anything it writes. It is simply writing whatever fits statistically to the prompt and context. Those people are wrong; statistics are a kind of reasoning (intuition) and if you try it, it obviously does the other kinds of reasoning too. I have all kinds of answers from it that are logically coherent and can’t possibly have appeared on the internet already. (Try asking it to write new SCP wiki entries.)
- jchmrt 4y ago
  Just because some of its answers are logically coherent due to it following statistical patterns, does not mean that the model is reasoning. Reasoning implies some type of discipline which prevents nonsensible arguments (even in uncommon situations).
  - astrange 4y ago
    No such thing as “is reasoning”, only “does reasoning”. Are you an unreasoning being if you make a math mistake? You can see it “does” reasoning if you say “think step by step” after a question. It can attend over its steps then.

tsimionescu 4y ago

> but GPT will never produce a bogus answer when google has the correct one. What are you basing this assumption on? GPT can always generate a bogus answer, because its only notion of whether an answer is good or not is whether your prompt + that answer are statistically similar to text in its training corpus. The more text from the training corpus it resembles the better GPT will think the answer is. For more obscure topics, this means that the chances of it hallucinating based on similar sounding much more popular topics increases greatly. This way of working also means that facts that appear more in the training corpus have a higher chance of being part of the model's outputs. For example, if the training corpus was not carefully curated to exclude vaccine conspiracy theories, there would be a good chance that the answer GPT gives to "autism is caused by ___" would be "vaccines", as this would likely be the most common answer in its training corpus. In contrast, Google's algorithms have more sophisticated biasing especially for these kinds of sensitive topics to ensure that more high-confidence sources are likelier to be top results, even if they are a minority of all the pages on the internet on that topic.

astrange 4y ago
> GPT can always generate a bogus answer, because its only notion of whether an answer is good or not is whether your prompt + that answer are statistically similar to text in its training corpus. “Statistically similar” is doing a ton of work even for the original GPT model - just printing the letter “e” is “statistically similar” to English text, if it can even be grammatically correct you should be surprised. And it’s certainly not true for the last two generations (InstructGPT and ChatGPT) which weren’t trained with that objective at all but used human raters.

noitpmeder 4y ago

This comment is based on such a large amount of BS. Reasoning over facts, as a concept, is completely divorced from the function of these GPT models.