What will be true about GPT-4.5?
Basic
155
93k
May 13
2%
It will be considered AGI by OpenAI
·
4mo

Variations like "GPT-4.5", "GPT-4.5 Turbo", "GPT-4 Turbo 2.0" all count. Must use separate branding but remain in the GPT-4 family: A model id update to "GPT-4 Turbo" or "GPT-4" wouldn't count. GPT-5 won't count either.

Add your own options. One who creates an option is responsible for clarifying follow-up questions. @Mira may interpret them if not, or resolve the option NA.

If GPT-4.5 is not released in 2024 or insufficient information is presented, options resolve NO by default unless specified to resolve differently.

If GPT-4.5 is released, moderators should close this market for trading immediately.

Resolution tags. Place at the beginning of an option to change how it resolves.

  • Default: Option resolves NO by default at end of 2024

  • Ambiguous(2024): Option resolves 50% by default at end of 2024.

See also: /MiraBot/what-will-be-true-about-gpt5

Get Ṁ600 play money
Sort by:

@dominic

It will start out at the #1 spot on the HuggingFace leaderboard when it is added

You're responsible for deciding how your option resolves. If you do nothing, I would be resolving NO on a technicality.

I see there's 2 HuggingFace leaderboards. This one will probably list GPT-4o, but it's not a "HuggingFace leaderboard". It's a mirror of lmsys's leaderboard. Huggingface doesn't directly manage it.

https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboard

This one is managed by HuggingFace, but is only for open LLMs. So it won't include GPT-4o, so resolves NO because it can't be #1.

https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard

Unless there's another major leaderboard that I didn't see on a search that is managed by HuggingFace and includes GPT-4o.

@Mira I think it should resolve YES assuming it starts off at #1 on LMSYS, I think I just looked at the URL and assumed it was run by huggingface when I added the option

@Mira Looks like it's #1 now, so the leaderboard resolves YES. But it was closer than I expected.

@Mira I think they purposefully capped it at ~GPT-4-level (intentionally just slightly above). Will be exciting to see GPT-5.

@Shump @JoaoPedroSantos You guys are responsible for deciding your AGI options. I'm not going to mark my calendar to check in 3 months. I'll ask for NA at the end of the year or it'll resolve 50% if I can't get that.

@Mira both no assuming we are talking about gpt-4o, but I think closing this market before we get confirmation that there won't be a true 4.5 was a bit premature.

"GPT-4o" is separate branding and in the GPT-4 family, so it would count for this.

It was directly confirmed to be "im-also-a-good-gpt2-chatbot", which I'm seeing the earliest reference is May 5.

So the April option could've been a misresolution, but fortunately was not.

@Mira how likely do you think it is that they actually release a model called GPT-4.5 now?

https://www.axios.com/2024/05/02/mystery-chatbot-openai-gpt2

Speaking on Wednesday at Harvard University, Altman told an audience that the mystery bot is not GPT-4.5

@Mira resolve April thx

Hmmm

@Mira you could have just bet more against Chris on /jim/is-gpt2chatbot-gpt5 , where the criteria were clearly understood by everyone involved.

I could be wrong but I don't think anyone but you expected these options in these markets to remain open until market close instead of resolving No as each month arrives.

I think that even if GPT2 ws a test of an unfinished GPT4 successor, a test under a different name also isn't the same as that model being released. This is especially true in this market where you are being particular about the model's name.

Do you think anyone besides you was reading the criteria like you are, and expecting these options remain open until market close?

@Joshua Just to head that off: Nobody is allowed to cite the $40 at stake. I can burn it in the LK-99 market if required. I don't care about it.

I have to NA this option early @Joshua because NA is going away soon, so I'm doing it while I still can.

@Mira Ah, fair enough

It will release in April 2024 or before

Resolution please @Mira

@chrisjbillington We don't know if gpt2-chatbot wasn't GPT-4.5, and it's at least a plausible candidate. The definition I use for "release" is always "anyone not affiliated with OpenAI is able to use it"(to account for waitlists and demos and such, otherwise I'd say Gemini API isn't "released" since they don't pay-as-you-go).

So it'll resolve NO at end of 2024 if it's not confirmed, but at least for my option it's uncertain whether it already happened.

@Mira your criteria say no such thing, resolution please

@chrisjbillington https://manifold.markets/Mira/what-will-be-true-about-gpt5#bos83fcbhe

It's paired with the other market which is linked in the description, and a question about release was addressed there.

@Mira if clarifications from the other market count then so does this one:

has to be described as GPT-5. Otherwise it's just an internal name

Which means gpt2-chatbot doesn't count even if it is released later under a different name.

Resolve please

It will release in April 2024 or before

Resolves NO @Mira

It won't release during 2024

I'm definitely rules lawyer-ing this, but technically you could argue that no matter what this option resolves NO, right? Since "If GPT-4.5 is not released in 2024 or insufficient information is presented, options resolve NO by default unless specified to resolve differently" and there's no resolution tag on the answer. So if GPT-4.5 does get released in 2024 then it resolves NO by the text of the answer, and if GPT-4.5 doesn't get released then we hit the end of 2024 and the default NO resolution activates. [You could argue the "unless specified to resolve differently" kicks in here but that would spoil the fun...]

@A I actually thought about that when I wrote it a couple days ago, but no it will resolve YES if GPT-4.5 isn't released.

It probably is a good idea to have this notice above all the bets rather than 2 pages down on my large vertical monitor. But if an official Manifold account is going to be taking notice and editing my titles, they can implement "pre-bet descriptions" or clustering the table into sections which are each locally sorted and have titles or any number of other UI options. Any of them is fine.

But this looks ugly, and I'm not going to accept it as my market title. You can make a better UI if you care that much - I would use it.

I should probably mention that the /Mira/what-will-be-true-about-gpt5 and /Mira_/what-modalities-will-gpt-45-have-ja markets have the same clause.

@Mira Fair enough, it is a bit of an ugly title. I'll pass the idea along.

how about [All options resolve NO if no GPT-4.5] or [All options resolve NO if no GPT-4.5 before 2025]? I think it's quite misleading to have the title as it is

"What will be true of GPT-4.5 in 2024?"

"What will be true of GPT-4.5 in 2024?"

I think 90% of readers wouldn't notice the subtlety there before betting. I probably wouldn't

The basic premise here is that almost nobody reads the description before betting, including many power users, and it sucks if you bet YES on 'gpt-4.5 will know what 1 + 1 is' and it resolves no

I actually have no idea how many people read the description, that's worth asking.

@traders did you guys read the description?

@Joshua I did

@Joshua I don’t if the title is seemingly unambiguous, but do if I’m curious about resolution criteria/etc

@Joshua I always read the descriptions. If someone bet without reading it, it's their fault unless the title was deliberately misleading or the description was deliberately long and convoluted.

@Joshua I read it, but only because Mira's markets sometimes have unexpected but crucial things like that that have burned me before. It is completely normal for me to bet in markets like this without reading the description, and I definitely think people who read it are extracting mana here on expectation from those who didn't.

Guys: You should prefer this style for the same reason that you should prefer joint markets instead of conditional. The probability that an actual resolution happens is related to the amount of extra liquidity that a market consumes: Like those nuclear markets with "amplified odds" have a higher probability but have effectively reduced leverage by externalizing the liquidity. Conditional markets externalize the use of extra liquidity, this internalizes it but you need to work with spreads instead of single legs.

Also: @Gen @SirSalty can you guys keep your badgers under control? I don't mind @Joshua pointing out the NO resolution issue below, that's important context even just as a trader; but now they're editing my titles, pinging all my traders, the majority of comments are now about an issue they created and not GPT-4.5 news/rumors, and @jacksonpolack hasn't even bet on the market and is just inviting himself in to express opinions I don't care about.

: You should prefer this style for the same reason that you should prefer joint markets instead of conditional

I don't want my bet on 'will the knowledge cutoff be after June' to also be a bet on if it'll release before 2025! That's just confusing. A joint market would be a linked MC with three options 'knowledge cutoff before June, knowledge cutoff after June, not released'.

and @jacksonpolack hasn't even bet on the market and is just inviting himself in to express opinions I don't care about.

I would have bet if it was 'resolve n/a if not released' instead of 'resolve no', hence the opinions. I don't think I have taken any badge actions on this

@jacksonpolack If they change it so that mega-markets can have clusters of linked options but each cluster is independent, then that would be even better I agree. That's another suggestion that was in my top-level comment.

At this point I think the simplest solution is that anyone who wants an N/A clause can just submit a conditional option like I did, and traders will see that some options resolve N/A and some don't.

Given that no one has complained so far, I do think Mira is right that we run the risk of making a mountain out of mole hill.

I'd probably add a (read description) tag to the title if I were Mira but Manifold should just add a tag for that.

If people want a version of this market where all options resolve N/A if they skip to GPT-5 then I think someone should make that market and it would get a lot of traders.