OpenAI will be announcing something every day for the next 12 days. What will be announced?
https://www.theverge.com/2024/12/4/24312352/openai-sora-o1-reasoning-12-days-shipmas
"Available to use" means available to some members of the public, not just to private alpha testers, but not necessarily to all free users either.
I'm sure there will be no controversy about the resolution of any of these markets 🤞 if you add your own, you're responsible for determining resolution
Update 2024-19-12 (PST): - Announcements about models that can only read data from windows but cannot perform actions or use the whole computer do not count as valid announcements for this market (AI summary of creator comment)
Update 2024-20-12 (PST): - o3 and similar names that are successors to o1 will not be considered original names (AI summary of creator comment)
Update 2024-20-12 (PST): - Names following the o_ naming scheme (like o3) will not be considered original names since they follow an existing pattern, similar to how GPT-5 is not considered original (AI summary of creator comment)
Update 2024-20-12 (PST): - The deliberative alignment technique described at openai.com/index/deliberative-alignment/ will be considered a research advancement specifically focused on safety/alignment, separate from the fact that access is being given for safety/alignment reasons (AI summary of creator comment)
1,000
3.00Resolved most things to NO.
I think https://openai.com/index/deliberative-alignment/ seems pretty clearly like a "research advancement specifically focused on safety/alignment." This option is trading really low right now though, so I'll leave it up to people to make their case either way. The best argument I can think of for NO is that it was "part of the o3 announcement" so it's not "specifically focused on safety/alignment." But this seems to me like its own thing that could generalize to models beyond o3, even if it happened to be announced on the same day.
The "New small model" question should be resolved by @AndrewMackenzie69e9. I guess this one isn't a slam dunk for NO since we actually don't know how much o3-mini costs per token, but it is probably NO. It's up to Andrew though
I would resolve the "research advancement specifically focused on safety/alignment" thing no (makes sense, I bought a bunch of no shares) bc it's an announcement of a really really good model and they only said they're giving access to people for safety / alignment reasons, not that it's like a breakthrough in alignment research or something like that. I think it's much more "research advancement specifically focused on displaying increased reasoning abilities of ai models with inference scaling" than anything about safety / alignment.
@Bayesian The fact that they're "giving access to people for safety / alignment reasons" is separate from the "deliberative alignment" technique described at the link I posted. That's what I'd resolve the market YES for. It's not necessarily a huge advancement in alignment and is maybe somewhat derivative of Constitutional AI, but it is a new technique for safety/alignment specifically
@CDBiddulph I read this as saying that “o3” would count as an original name, since it’s not Sora, o1, Orion, or anything with GPT in it. Is that an accurate reading?
@CharlesFoster I was about to ask the same thing. My initial thought was no, since o3 is a successor to o1, but it's a little ambiguous.
@CharlesFoster Yep! I should have clarified earlier, but I would not consider o3 to be an original name, given what I said about "GPT"
This one jumped up today - I'm just going to clarify that today's announcement doesn't count in my opinion, because the model can't actually use your whole computer or perform actions. It can only read data from individual windows, and can't even do things in those windows directly, IIUC. Sorry, this may be the most ambiguous market here