I expect it to become more resource-efficient to run, rather than more capable

we should all be ashamed that this market isn't at 50%

@Bayesian people could interpret it as referring to our pre-GPT-4.5 expectations, or to society's expectations

extending until July 1 since Sam said GPT-5 is still coming

How is this affected by the release of strawberry? If we get, in order, gpt-4, gpt-4+strawberry, gpt-5, gpt-5+strawberry?

Will the expectations be surveyed beforehand, or will the poll ask if GPT-5 meets a user’s arbitrary expectations?

@thepurplebull One month after it's released I'll ask "Does GPT-5 exceed your pre-launch expectations?" I'll resolve this based on the results of that.

In every paper comparing various leading LLMs, you see that the refusal rate of OpenAI's various GPTs is literally an order of magnitude higher than competitors. And what's more is that it continues to increase with newer releases of GPT. It has gotten to the point that it will refuse to answer questions about math and science even when explicitly prompted that it's own best guess is okay.

I feel like GPT-5 hype has gone way too far.