How many parameters will GPT-4 have?
Mini
117
16k
resolved Jun 12
Resolved
1.8t

Open numeric answer for number of GPT-4 parameters. Market will resolve 100B if there's fewer than 100B, and 100T if there's more than 100T.

GPT-3 was 175 billion. There have been rumors that GPT-4 will be much bigger.

Similar markets:

https://manifold.markets/MaxGhenis/will-gpt4-have-at-least-100-trillio (requires >100T, which is currently unlikely)

Get Ṁ600 play money

#NameTotal profit
1Ṁ32
2Ṁ30
3Ṁ15
4Ṁ14
5Ṁ10
Sort by:

I don't really know how these markets work, but it seems weird that I bought HIGHER 1.1t -> 1.4t and lost my entire investment?

@Eliza just fyi.

@DanMan314 Yeah, it looks definitely wrong. The people with the most profit all bought 'Lower' on high values. I'll unresolve it for now?

I unresolved it and it paid people back, but the frontend UI is still wrong. I think I can re-resolve it with a different value and get the correct payouts.

I'm rich 😅

I unresolved a second time and re-resolved a third time. This time I am more confident in the result.

This is an 'old' Numeric market where it is just a skin over a regular binary market. There are two factors used to resolve it, one is a `value`, but this is literally just a number used for display purposes. The other is `probabilityInt`. The API docs say to calculate it this way:

`probabilityInt`: Required if `value` is present. Should be equal to

• If log scale: `log10(value - min + 1) / log10(max - min + 1)`

So I tried using a figure like 0.87xxx., but apparently it wanted 87.xxx instead. Now it looks much better.

Notice how everyone who played got back all their money and then like 5 extra mana or 5 less mana? I guess that's why they deprecated this market type 😂

@mods It looks like this needs admin intervention, it's not possible to resolve numeric markets to large values

@Gabrielle did you try it with the API also?

I have not tried it, so if you would like to try it that would be great

Alright, since no one else has been willing to step in we'll go with the "try it and see what happens" approach. I am going to attempt to resolve to 1800000000000 with the API. If it breaks the market we will have to try to unresolve it or else just call the admins. Now to re-count those zeroes....

I intend to resolve to 1.8T based on many other similar markets resolving similarly, and that being the market’s prediction (ie, you could buy and sell at the resolve price for a while now).

If you strongly disagree please reply. Otherwise, all the information we have may be the closest we’ll get.

@andrew I tried to resolve it with the API. It appears visually to have worked, but someone needs to see if it actually paid out the correct amounts. If there is an issue, we can unresolve it and wait for Ian to look at it.

sold Ṁ0 LOWER

NVIDIA may have accidentally revealed that it's (or perhaps the next one is) 1.8T params. Not treating as authoritative, but interesting.

@andrew It's not accidental. He clearly says:
> The latest, the state-of-the-art OpenAI model, is approximately 1.8 trillion parameters.

Calling it GPT-MoE-1.8T reoccurs multiple times throughout the presentation. Jensen knows what he's talking about, they are the ones providing the hardware for OpenAI.
Source

I think the only uncertainty here is exactly which model he is talking about, but I'd say it's pretty safe to resolve this.

The other nuance is how to deal with the fact that it's MoE. If each of the experts have 200 billion parameters, does this resolve to that?

@Shump Yeah — agreed he knows what he's saying. Question is whether "the latest" is the one released (GPT-4) or the next. Unless consensus here disagrees strongly.

As for how to resolve, I lean pretty solidly towards counting all trained parameters. The result is most clearly not a 200b model — it's a stack of 8 of them, each needing to be trained, and each being used at runtime. The fact that only a subset of experts get activated for a specific token's evaluation doesn't really change that.

predicted HIGHER

Rumors have been coming out, will wait for more confirmation. But looks like the market adjusted.

predicted LOWER
predicted HIGHER

There's a 48% chance that GPT-4 has over a trillion params acording to manifolders. This is very, very low.

predicted LOWER

@ShadowyZephyr That market is also idiotic

predicted HIGHER

@jonsimon Well yes, GPT4 params will probably not release within any reasonable time frame, but it's pretty likely more than 200B lol

predicted LOWER

@ShadowyZephyr It depend strongly on whether they incorporated chinchilla scaling laws. If they did then 200B is quite plausible

predicted HIGHER

@jonsimon Then why don't you correct it?

predicted LOWER

@rockenots I'm one of the largest No holders in that market

Oh, he says now it was "a figure of speech" https://twitter.com/SebastienBubeck/status/1644151579723825154

So it's pretty weak evidence but still (also I cannot delete previous comment anyway).