Buy YES in options you think are accurate enough to resolve a market on, but NO in options you think are not. Buying even 1 mana is enough to count as a vote and help us resolve the market more accurately.
With OpenAI's announcement of GPT-4o, Manifold now has to deal with the resolution criteria for quite a lot of markets that were anticipating the next version of ChatGPT.
Many markets have criteria saying they resolve for GPT 4.5 or GPT 5, but would also include if OpenAI released a model that could have been GPT 4.5 or 5 but deviated from that naming scheme. Is GPT-4o one such model? There's still a good deal of debate about that!
Some examples of these markets:
/Joshua/will-openai-release-gpt-45-or-gpt-5
/ms/when-will-openai-release-a-more-cap
/Joshua/when-will-openai-broadly-release-gp
/Joshua/before-what-month-will-openai-relea
/JonasVollmer/gpt45-released-this-quarter
I think one thing that would be interesting to add on to all this is a poll market! Options in this market resolve YES if a majority of positions are YES holders, and NO if a majority of users are NO holders, at market close on Saturday 5/18. In the event of a tie, the market will be extended 24 hours.
Every YES or NO holder counts for only one vote, regardless of how many YES or NO shares they hold.
This market's outcome does not have any power over other markets, but hopefully a gauge of public opinion will be helpful for myself and the other users in charge of these tricky resolutions.
1,000
3.00The confusion here is largely around whether or not AI has become asymptotic. For example, it's possible that gpt5 might just be a MoE / agents on top of gpt2. Many folks have speculated this is what gpt4 largely was. More CoT on top of previous models.
Does that mean gpt5 will be the next big pricey model or does it just mean that OAI is merely throwing more compute at the problem?
So in fact, it's possible that gpt2 is the key that will empower future advances. Or maybe gpt5 will fundamentally be a different arch. We just don't know.
@gpt_news_headlines Would be funny if they keep naming new models to reflect diminishing returns on expectations
If nothing can live up to the promise of GPT5, we just get GPT4.5, then GPT4.7, then GPT4-ultra, then GPT4-baja blast.