The below seem like easily the three most difficult ones. My understanding is that they judges would need to grant at least one of these. And even if they were technically achievable, the judges claiming that they had been accomplished would require the judges to claim that human skills/creativity are not as valuable in some of these highly valued domains. This seems very unlikely to me by the end of 2027 if the judges have to be agreed upon by both people.

With little or no human involvement, write Pulitzer-caliber books, fiction and non-fiction.

With little or no human involvement, write Oscar-caliber screenplays.

With little or no human involvement, come up with paradigm-shifting, Nobel-caliber scientific discoveries.

@AndyMcKenzie yes I find the challenge a bit odd for this reason. At least if I were a judge, the degree of difficulty of these 3 would be massively higher than the rest, making the rest not-particularly-relevant. Then it in part devolves into a challenge of defining “Oscar-caliber” screenplays. If it’s like, “looks equivalent to a layperson to something that wins an Oscar”, the bar is very low (lay people aren’t equipped to judge this). If it’s like, “AI might generate something that’s as good as the worst Oscar winning screenplay according to an expert judge”, then that’s moderate difficulty, but taste varies so if they think some winner is trash then maybe a mediocre AI achieves this. My interpretation would be about it being able to routinely achieve quality that would be broadly recognized by experts as Oscar-caliber, which is an extremely high bar. (But there’s a distinction here between forecasting the judges and forecasting AI progress)

@AdamK want more at 25%?

uh i also want exit liquidity pls

@Bayesian people, don’t fill Bayesian

at 24% it’s just exit liquidity not a real bet. Skip right to 25% if you want a real counterparty

I removed the 24% one it seemed rude but if yall fill ziddletwix (you should it's a sure bet) you can also go ahead and fill my limit order at 27%, thanks everyone

@Bayesian dw either way accurate price discovery via market mechanisms Is never rude

@Ziddletwix but it wasn't that as much as you creating a comment that made me aware of this thing going on. accurate price discovery via market mechanism incentivizes you to just dm adamK instead of publicizing this limit order which is cursed

@Bayesian Sure, happy to fill everyone's orders here lol

Living meme

I created a multiple choice version of this to get mroe refined opinions, e.g. which tasks will be achieved. https://manifold.markets/LovkushAgarwal/possible-outcomes-of-the-bet-on-ai?

There is same market on metaculus. https://www.metaculus.com/questions/31246/2027-ai-bet-winner-between-gary-marcus-and-miles-brundage/

For posterity, here was the original dialog when GM and MB were pinning down the wager on Twitter:

MB: In my new blog post, I share a starter pack for feeling the AGI: resources and arguments indicating that AI is virtually certain to exceed human performance in most areas in the next few years, and that the time to act is now.

GM: What odds will give you me on that Miles Brundage? 19 to 1, if it is almost certain? My $10,000 versus your $190k? Do you really believe this? What’s “few years”? All humans? Some humans? Can we put this on Metaculus?

MB: I don't feel comfortable with such large amounts of money for various reasons (namely it just feels obscene given all the good we can do with our $). But sure, I'll take those odds. Let's say: - $1k vs. $19k - List of benchmarks that you have previously articulated, minus physical stuff - I'm right if Epoch or METR AI says over 80% of your non-physical benchmarks were surpassed by end of 2027

GM: My list is a lot narrower than what you said above, you realize? But good man for mostly standing behind what you say, and I accept.

MB: Could you share the exact list?

GM: My two lists are in the two bets I offered Elon: 2022: https://garymarcus.substack.com/p/dear-elon-musk-here-are-five-things?utm_source=publication-search… 2024: https://garymarcus.substack.com/p/superhuman-agi-is-not-nigh… (bit high on robotics because Elon was emphasizing) My view is that you should not entirely discount the physical, because one thing ordinary people can learn to do very quickly is teleop robots, and another is to learn about their environment and quickly adapt to it. Happy to at least thin them though (e.g., drop the mountain bike example).

MB: Again, I want to focus on just non-physical stuff (I'm already being "generous" in the 95% odds + 80% threshold, and not quibbling about e.g. the 10k lines of code thing which as I noted years ago, is a superhuman bar, and the Nobel thing etc. are all quite high).

GM: Note. I will reply in morning after reading your blog, to decide a fair way to resolve/how strong your claim was, and take on board your comments here. Love this reasonable back and forth!

METACULUS: Just messaged / emailed each of you. Happy to support this bet with a forecast question.

I'm curious where the community will land on this one, so I subsidized a bunch!

Why did Miles agree to these odds?

@Usaar33 to be fair, he did say "almost certain," and I think people these days tend to think that's at least 90%. I'm not sure what the most credible data is on this but this is the oldest I know of is from a 1977 DoD report:

@Usaar33 Yeah I'm confused; feels like many people would bet at at least even odds against some of the stricter criteria like Nobel prize

@Usaar33 Costly signaling. https://x.com/metastable_1/status/1874478836370608424

https://x.com/GaryMarcus/status/1873766399618785646

https://garymarcus.substack.com/p/where-will-ai-be-at-the-end-of-2027?r=8tdk6&utm_campaign=post&utm_medium=web&triedRedirect=true

Apparently the bet has been booked at 10:1 odds (i.e. Miles's $20k to Gary's $2k)

The linked x whatchamacallit doesn't include the terms of the bet it refers to. Can you link the bit it is in reply to

@JussiVilleHeiskanen Yeah, I can't figure out if they've actually made a bet. We can try to extract what the bet would've been even if they fail to pin it down.

PS, I've now extracted the dialog from Twitter. I think they may have fizzled out on pinning down a wager? PS, nope, they were super on top of it, just moved things to Substack.