I will run a Manifold poll 1 month after the official GPT-5 release asking whether or not it exceeded expectations. Resolves to results of that poll.
Update 2025-22-01 (PST): - The market closure date has been extended to July 1, 2025 following updates on the GPT-5 release schedule. (AI summary of creator comment)
@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.