Resolves YES if by the end of the year we demonstrate on the podcast that we can get GPT-4 to write better resolutions than the market creator for at least 50% of markets selected by a panel of top market creators.
1,000🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ167 | |
| 2 | Ṁ113 | |
| 3 | Ṁ111 | |
| 4 | Ṁ81 | |
| 5 | Ṁ53 |
People are also trading
Is the comparison against some set of high-quality markets, or against everything? I feel like I have seen a LOT of lazy/missing resolution criteria. Though, many markets e.g. sports events look like they are created from a template, and might zerg a count-based comparison. On the other hand, GPT-4 will lack context on extrapolations from 2023 current events -- Gigacasting's example below already reveals it didn't seem to recognize Manifold conventions or terminology.
(I'll sit this out for now since 29% doesn't seem obviously mispriced either way given the above)
It’s alright. It makes some semantic mistakes like “better resolution” instead of “better resolution criteria” and having a podcast seems pointless. I think gpt-4 3-shot could do it though but this market is too vague
Can you provide some more context? It seems like market creator + GPT should always be equal or better than market creator alone, assuming there are no restrictions on how GPT is used. Does it have to write all the criteria alone from just the question? Is it an iterative process? Multi-shot? One-shot?
@JimHays Yes I will add details from laptop. Typing out resolution criteria on my phone is annoying.