How does this resolve if gpt2-chatbot turns out to be Q* (since it was available to the public before June)? That’s my mainline justification for betting YES. Suppose we get confirmation in the form of a blogpost or paper from OpenAI.
@chrisjbillington Would like to hear your thoughts as well as top NO holder
@AdamK Just seems unlikely I suppose.
gpt2-chatbot is roughly GPT-4 level and talks like GPT-4, appears to have the same system prompt, just looks like GPT-4 in every way. I think if it were something radically different it wouldn't seem so similar. Why would they train it up to almost the exact same level of capabilities? That's weird, and seems much more likely explained by it being something incremental.
Last we heard of Q* (I'm not really convinced any such thing exists) it was a small model that was able to solve some basic logic/maths problems with less resources than it takes for larger models to do so. I threw some logic puzzles at gpt2-chatbot and it was not very good at them. I don't see any breakthrough - if it's something radically different and in principle more powerful underneath then we would have to explain why it is not very good in practice.
@chrisjbillington I believe otherwise, glad for the clarification. If you agree with a YES resolution in the case of confirmation by OpenAI this month, I can place a large limit order at 20%.
@AdamK Let's. In the meantime, there's a Ṁ40k NO limit order at 25%.
@chrisjbillington Also see limit orders at https://manifold.markets/Sss19971997/what-is-true-about-gpt2chatbot