What modalities will GPT 4.5 have? (January 2024)
Basic
47
αΉ€36k
resolved Feb 2
Resolved
NO
Image(Input)
Resolved
NO
Image(Output)
Resolved
NO
Video(Input)
Resolved
NO
Video(Output)
Resolved
NO
Audio(Input)
Resolved
NO
3D(Output)
Resolved
NO
3D(Input)
Resolved
NO
Audio(Output)

Announcement of capabilities counts, even if the API on release doesn't have it. Refers to the actual model, not to any tools packaged with it: GPT-4 does not have image output, despite ChatGPT supporting API calls to DALL-E 3.

Every option resolves NO if no model named GPT 4.5 is released by end of January.

Screenshot from a supposedly fake leak:

Get αΉ€600 play money

πŸ… Top traders

#NameTotal profit
1αΉ€534
2αΉ€379
3αΉ€216
4αΉ€162
5αΉ€149
Sort by:

Why are these prices still above zero?

The "insiders" not only knew nothing, but they pretty much admitted by this point that they know nothing. So very probably all of these are going to resolve NO.

The address in the "supposedly fake" leak returns 404.

Of course an API can return anything it wants, but the norm would be to return 401 or 403 in those circumstances.

100% actually fake.

@DavidBolin I made this market to get better prices on some options(why is GPT 4.5 being released in December at 5% while image inputs here were 34%? Every option here should be strictly less than it being released at all, since they resolve NO not NA.)

But the average Manifolder has such poor reading comprehension that they all missed the free money and I didn't get a chance to buy "image output @ 2%" or something.

It might have worked if you had made these resolve NO if it doesn't come out by end of December, instead of end of January.

@DavidBolin I don't think so. People buying NO would be buying it because e.g. roon explicitly states "there is no GPT 4.5". So anyone with an investment thesis of "it's not coming out" should be buying NO regardless.

I don't want people arbitraging(because this takes liquidity when they buy YES on other markets), I want more NO bettors. So I have to make them not completely arbitragable.

@Mira You strongly implied that you knew for a fact that there is a GPT 4.5 and that it is coming out soon.

Assuming those things are true (and you were not just trying to get us to make bad bets), I would expect it to come out in January at the earliest, as I said before. That is precisely why I sold my NO bets here at a profit, while not selling the NO bets on the Christmas and EoY markets.

@Mira does it count as image output if it's calling text2im to make images with DALL-E as ChatGPT does currently? Or does this require image output somehow to be part of GPT-4 itself?

(Even if it's kind of just marketing to officially declare an image model to be part of GPT-4? Not really sure there is a principled way to draw lines between models other than how they're marketed)

@chrisjbillington The case of DALL-E is explicitly mentioned:

GPT-4 does not have image output, despite ChatGPT supporting API calls to DALL-E 3.

Supporting a modality means that modality is mapped to an embedding vector along with the text.

@Mira sorry, I didn't read, my bad. Thanks!

@chrisjbillington Just in case anyone else is confused, here are 3 more details:

  1. It's acceptable for a modality to be mapped to multiple soft tokens(like image inputs are).

  2. Video is not the same thing as multiple image inputs.

  3. ChatGPT supports audio input and output, but GPT-4 doesn't because it simply reads/writes the text.

Mira 🦚boughtαΉ€5Image(Input) NO

why would it not have input image input

@TheBayesian This doesn't N/A

@Joshua wdym?

@TheBayesian

Every option resolves NO if no model named GPT 4.5 is released by end of January.

@Joshua ah ty

Sneaky market

@Joshua Ah yes, my favorite source of news: random Reddit comments ❀️

sexy the walking dead GIF