🏅 Top traders

What?? How is GPT-4o not close to GPT-4.5?

@Hazel Was a judgement call like many resolutions in these kinds of markets, but that answer was intended to cover specifically the name GPT-4.5 or close variations like GPT-4.5T.

@ManifoldAI I see

?

@maxhumber they apparently decided that the back of his head counted :/

@ChrisPrichard They didn’t decide anything. He was on the livestream.

@Predictor I agree that that is a possible interpretation, indeed apparently the interpretation of the resolver. That said, I'm annoyed by the resolution in any case.

First off, it's not clear that it even was Sam Altman, given that you couldn't see his face. Secondly, the natural expectation of "Sam Altman is on the livestream" would be that he's up on stage, or even included in the camera shot on purpose. This seems more like the angle of the camera happened to catch what might have been the back of his head and on a technicality the way it is worded means it resolves yes.

Perhaps technically correct is the best kind of correct, but I'm annoyed anyway.

@ChrisPrichard

yes, when we bet on this option it was about "will sam altman present, appear, etc." in a meaningful way. It wasn't about whether a pixel from his hair would appear. We got zero words, emotions, or anything else from this.

So, I'd resolve NO since the entire "NO" camp's predictions all came true: there were no words, sounds, expressions, emotions, ideas, etc from SamA. Really, "YES" generally means "we got something of sama as a human". Since we didn't, this isn't YES.

This belief is due to judging cases where people push for resolution based on meaningless corner cases and seeing how frustrating this kind of resolution is.

@Ernie You got a lot out of “Sam Altman is on the livestream.” You have a very big imagination and I’m proud of you for trying so hard. ⭐️

@Predictor https://manifold.markets/Ernie/boat-owners-only-sign-correct-inter?r=RXJuaWU

@Ernie how many pixels of Sam, for how many frames, qualify? How much uncertainty will you tolerate?

Take this example:

We take 100 men dressed identically, one is Sam. We randomly split them into two groups of fifty and send one group home. We take the other group to the desert, at night, and they hold candles at a distance of 1 mile from the camera standing close together. Video camera turns on, spins around fast, glints of light show up, smeared pixels, which probably represent the fifty faces lit by candles, but super blurry. We do not know if it is Sam but it definitely could be. Regardless we gain no info from those pixels.

Does that count?

To answer, I think you need to make some kind of decision based on information transmission. I can vary the parameters to cause a clear yes (zoom in on just sam, visible emotion, he is holding a sign with a message etc) and a clear no (weaken the pixels, lower the probability he's even in the image, make the world one such where he has 100 identical twins who we may have mistakenly included. I'd keep weakening them til you admit we have reached NO - say a universe where there are 1billion fake Sam copies, real sam hasn't been seen for 100 years, but there are rumors he is immortal, I made a 1% probability that a single pixel is a possible real sam but likely a fake Sam, etc etc. At some point you cannot keep saying that no matter how low the probability, that any crappy single pixel in a single frame counts.)

Once you admit that, it means we're actually in a negotiation about what counts and what doesn't. And we have to figure out where to draw the line for what counts. So, how do you make the decision?

And I'm arguing that wherever we set it, the state of "nothing significant specific to Sam is visible, and we learn nothing at all, and it might not even be him" is on the NO side. It's not possible to perfectly define it, but what happened was very similar to all the "Sam did not appear" cases you can imagine, and doesn't resemble at all most of the "Sam appears" ones.

@Ernie No need to talk to me about it, you're not convincing me of anything. Add it to your manifesto and maybe we'll read about it someday.

@maxhumber That seems like a bad resolution

There is no certainty in reality. Sam could have a twin brother, Sam could be a VR effect, etc etc. everything is just a probability. Even if "Sam" appeared, someone could plausibly argue it wasn't really him. Would we NO in that case?

Typical traits of Sam on livestream

• we learn information from his voice, words, expression, action

• we see a face that looks like him

• we hear a voice that sounds like him

• he responds and interacts with other people, things, questions, etc.

Things that the alleged "back of head of Sam" did

• appear and have brown hair

Think of the space of YES-ish things, from "Sam declares himself god-emperor and transforms into an immortal meta-AI-cyborg" to "Sam confronts a revolutionary anti-ai activist" to "Sam announces products and answers questions" to "Sam introduces a new hire and hands off the mic"... etc

There are billions of scenarios like this, and they all would result in nearly every human saying "Sam appeared in the video".

Then after a long, long time, you get to a scenario like "Someone with brown hair is seen from the back, who might be Sam. Continuing on, we reach items like "The video is entirely conduced by Mira", "The video is CGI completely and has no people", "the video is nothing but AI text and hypnotic music"

So we have to draw the line between these two zones. Manifold can pick how it wants to do this, and I think it makes sense to match up YES and NO beliefs & expectations and scenarios to what happened. This isn't a court, and we aren't lawyers. What is the claim about and what traits does a typical case of YES and NO have? What happened is basically a canonical NO case, except for a back of brown head. If Sam had even turned around, we'd have seen his emotion, and that would have been a clear appearance. Or if we knew that the head was him, we could infer meaning from it being in the front row, but even that's in doubt. But that's not what happened. If all there is is just possible-Sam in the front row, our reality is basically identical to the NO holders description of the scenario: "Sam doesn't participate at all, Sam contributes nothing, nobody questions Sam, there is no Sam Voice, we learn nothing about what Sam feels like or hopes for from the product, etc.". Those NO-matching things all happened, yet the resolution is YES? That doesn't seem right.

To the "but, meh brown pixels, it's clear-cut" I can create scenarios like: Imagine a high school/baby picture of someone who might be Sam - would that count? Imagine Sam had died and we saw a stone bust of him, allegedly based on his final appearance, or we saw his last wave of the arm and nothing more, or we saw a tree branch swaying which was claimed to mystically represent the last wave of his arm, or we saw a stick which he was allegedly holding as he waved his arm for the last time. I can vary the params between "you see his actual face" to "you see the skin on his arm" to "you see the fabric of his alleged shirt" to "you see an alleged image of what he may have looked like" to "you see an alleged image of an imagined situation he was in" to "you see a hypothetical imagined version of him which takes huge liberties and doesn't resemble his actual physical self at all". You're going to have to decide, and you need a resolution system which can help in all those tough cases. Just looking for "meh pixels of sam's alleged body" doesn't cut it. The stable way to resolve is to compare YESish and NOish scenarios to reality.

What market bettors cared about here was whether Sam would clearly appear in a leadership role, what his emotions would be, what we would say, etc. None of that happened, it's clearly NO-ish.

@Ernie company literally creates Sora and people still trust its livestream 🤦

Should this not have resolved YES? Screenshots from YouTube + Transcript

@nicefryroll ChatGPT memory and continuity across conversations were announced/released back in Feb, and she's referring to that - afaict there aren't any particular improvements or expansion on memory abilities for ChatGPT?

hey look at how good I did (no live betting)

industries were mentioned but only in the context of stakeholders. she talked about users but not their industry (professors, students, podcasters, builders)

the video and transcript are on YT now and the delve, accelerate, universal basic compute options are all no.

not mentioned by name, but arguably the majority of the live stream (starting with Mira saying ChatGPT is far more useful and helpful because it has a sense of continuity across your conversations, and "Today, we have 100 million people and they use ChatGPT to create work, learn...") and then all of the demos ("I wonder if you can help me" "sure I'd be glad to" interactions) could count as AI assistants?

Story about robots doesn't count?

Sam Altman mentioned during his live tweeting of the live stream https://x.com/sama/status/1790066003113607626?s=46&t=8czzAcEeJp-MF1JheldZPQ

@StephenMWalkerII That's not in the livestream though, I don't think that should count.

@Joshua it's not in the video, but when the ceo live tweets the entire time, it's as good as the video IMO

Resolves YES.

@MugaSofer gpt-4o is free

@shankypanky So? It's still usable by anyone in the US.

@MugaSofer

usable by anyone in the United States willing to pay

Someone can be willing to pay but not required to pay.. er

I interpreted this as a paid product?

Imo both interpretations are perfectly reasonable and natural language doesn't disambiguates them

it is also paid, available as API / playground

@shankypanky I think it pretty clearly includes free products, but even if it didn't, there's a paid version via the API.

@MugaSofer paying for an upgrade in spec/capability doesn't change the fact that the product is available for free, though.

you're not paying for the type of software, you're paying for added abilities.

but tbh it's up to ManifoldAI to decide how to resolve - I need to log off and go back to work lol