IE "make me a 120 minute Star Trek / Star Wars crossover". It should be more or less comparable to a big-budget studio film, although it doesn't have to pass a full Turing Test as long as it's pretty good. The AI doesn't have to be available to the public, as long as it's confirmed to exist.

Scott Alexander's 5 year predictions

Technology

AI Media Creation

Resolves YES if GPT-5 is publicly released before Jan 1, 2025; resolves NO otherwise.

Update from March 18: If the next OpenAI flagship model deviates from the GPT-N naming scheme 

 that is going to be released from OpenAI (e.g. they completely changed the trajectory of their naming scheme) then I will count it as "GPT-5" regardless.

However, if the next flagship model is not called GPT-5 but still conforms to the OpenAI GPT-N naming scheme (e.g. GPT-4.5) 

there will eventually be a model called GPT-5

 that will be released (i.e. they are going to continue with the same naming scheme) then I will not count it as "GPT-5".

🤖 AI

🖥️ Technology

🤖 OpenAI

GPT-5 Speculation

ChatGPT

This question resolves to YES if during at least one of the main (non-VP) 2024 general presidential election debates in the United States, the moderator asks at least two separate questions directly related to artificial intelligence, such as the potential for job losses due to AI automation, or existential risk from AI. This question also resolves to YES if AI is otherwise brought up by a candidate in the context of a question unrelated to AI and ALL candidates talk about AI at least once during the same debate, with at least one of the candidates saying "AI", "artificial intelligence", "robots" or a close synonym at least three separate times, at least five seconds apart, during the same debate. Otherwise, this question resolves to NO.

Asking "at least two separate questions" means that the moderator must pose two semantically distinct and individual queries, each intended to elicit a unique response, specifically about artificial intelligence. In other words, simply repeating the same question doesn't count as two separate questions, even if the question is asked twice, first to one candidate, and then a second time to the other candidate. Moreover, asking two distinct questions without waiting for the answer to the first one counts as two separate questions for the purpose of this question. For example, a clear positive instance would be asking first, "How do you plan to manage the existential risks from AI?" and then later asking, "What are your proposals to mitigate job loss due to automation from AI?" These are two separate questions because they cover different facets of artificial intelligence—existential risk and employment. A clear negative example would be if the moderator simply asked, "How do you plan to manage the existential risks from AI?" to one candidate, and then later said "Same question to you" to the other candidate. Even though the question is asked twice, it's not considered two "separate" questions because it's a repetition of the same query.

When it comes to questions related to automation and job losses, the key factor that determines whether the question is "directly related to artificial intelligence" is whether the moderator mentions AI technologies or AI companies, such as ChatGPT, OpenAI, generative AI, or neural networks; any question will automatically qualify if the moderator says the word "AI" or "artificial intelligence" in the question. General questions about automation or job losses that don't clearly link back to AI won't be considered "directly related to artificial intelligence". For example, the following questions would NOT count as being directly related to artificial intelligence:

If there are no 2024 general presidential election debates, this question resolves to N/A.

US Politics

🗳️ Politics

presidential debate

I will resolve this based on some combination of how much it gets talked about in elections, how much money goes to interest groups on both topics, and how much of the "political conversation" seems to be about either.

Politics

Medicine 

Abortion

https://www.lesswrong.com/posts/sWLLdG6DWJEy3CH7n/imo-challenge-bet-with-eliezer

 Eliezer has this at >16%; Paul at <8%. Resolves to YES if either Eliezer or Paul acknowledge that an AI has succeeded at this task.

https://manifold.markets/MatthewBarnett/will-a-machine-learning-model-score-f0d93ee0119b

Update: As noted by Paul, the qualifying years for IMO completion are 2023, 2024, and 2025.

Technical AI Timelines

IMO Grand Challenge

Mathematics

Third Party Validated, Predictive Markets: AI

If a large language models beats a super grandmaster (Classic elo of above 2,700) while playing blind chess by 2028, this market resolves to YES.

I will ignore fun games, at my discretion. (Say a game where Hiraku loses to ChatGPT because he played the Bongcloud)

Some clarification (28th Mar 2023): This market grew fast with a unclear description. My idea is to check whether a general intelligence can play chess, without being created specifically for doing so (like humans aren't chess playing machines). Some previous comments I did.

1- To decide whether a given program is a LLM, I'll rely in the media and the nomenclature the creators give to it. If they choose to call it a LLM or some term that is related, I'll consider. Alternatively, a model that markets itself as a chess engine (or is called as such by the mainstream media) is unlikely to be qualified as a large language model.

2- The model can write as much as it want to reason about the best move. But it can't have external help beyond what is already in the weights of the model. For example, it can't access a chess engine or a chess game database. 

I won't bet on this market and I will refund anyone who feels betrayed by this new description and had open bets by 28th Mar 2023. This market will require judgement.

♟️ Chess

In "Moving the Eiffel Tower to ROME", a paper claimed to have identified where the fact "The Eiffel Tower is in France" had been stored within GPT-J-6B, in the sense that you could poke the GPT there and make it believe the Eiffel Tower was in Rome.

"The Eiffel Tower is in France" seems (in my personal judgment) like the sort of fact that early AI pioneers could and did represent within GOFAI systems. GPT-J probably does more with that fact - it can for example answer how to get to the Eiffel Tower from Berlin, believing that the Eiffel Tower is in Rome.  But the paper didn't offer neural transparency into how GPT-J gives directions, we don't know the stored patterns for answering that part - just a neural representation of the brute idea that GOFAI pioneers might've represented with in(Eiffel-Tower, Rome).

This market reflects the probability that, in the personal judgment of Eliezer Yudkowsky, anyone will have uncovered any sort of data, pattern, cognitive representation, within a text transformer / large language model (LLM), whose semantic pattern and nature 

 familiar to AI and cognitive science in 2006 (to pick an arbitrary threshold for "before the rise of deep learning").

Also in 2006, somebody might've represented "the Eiffel Tower is in France" by assigning spatial coordinates to the Eiffel Tower and a regional boundary to France.  Idioms like that appear in eg video games long predating 2006.  Nobody has yet identified emergent environmental-spatial-coordinate representations inside a text transformer model, so far as I know; but even if someone did so before the end of 2026 - as mighty a triumph as that would be - it would not (in the personal judgment of Eliezer Yudkowsky) be an instance of somebody finding a cognitive pattern represented inside a text transformer, which pattern was unknown to cognitive science in 2006.

2006 similarly knew about linear regression, k-nearest-neighbor, principle components analysis, etcetera, even though these patterns were considered "statistical learning" rather than "Good-Old-Fashioned AI".  Identifying an emergent kNN algorithm inside an LLM would again 

constitute "understanding via transparency, within an LLM, some pattern and representation of cognition not known in 2006 or earlier".  Likewise for TD-learning and other biologically inspired algorithms, including those considered the domain of neuroscience (from 2006 or earlier).

GOFAI and kNN and similar technologies did not suffice to, say, invent new funny jokes, or carry on a realistic conversation, or do any sort of intellectual labor.  The 

of this proposition, if relevant, is to assert that by end of 2026 we will not be able to grasp any inkling of the cognition inside of LLMs by which they do much 

than AIs could do in 2006; we will not have decoded any cognitive representations inside of LLMs supporting any

 cognitive capabilities original to the era of deep learning

.  We will only be able to hunt down internal cognition of the sort that lets LLMs do more trivial and old-AI-ish cognitive steps, like localizing the Eiffel Tower to France (or Rome); on their way to completing larger and more impressive tasks, incorporating other cognitive steps; whose representations inside the LLM, even if we have some idea of which weights are involved, have not yet been decoded in a way semantically meaningful to a human.

AI Alignment 

Mechanistic interpretability 

This will resolve to the Twitter/X account that has the post from the year 2024 with the most likes on January 1, 2025 at 12:00am ET.

Feel free to add any account you want, but remember that the answer has to be a single Twitter account. "Joe Biden" would not be an eligible answer, while @JoeBiden and @POTUS would be eligible.

🍿 Entertainment and Pop Culture

👴🏼 Joe Biden

👴 Trump

✖️ X (Twitter)

Social Media

🚗 Elon musk

👸 Taylor Swift

Greta Thunberg

If TikTok is divested from its China, and TikTok or its assets are transferred to a U.S. company, "divest it" resolves YES.

OR, if ByteDance disallows U.S. users from accessing TikTok, or otherwise exits the U.S. market, "shut it down" resolves YES.

Alternatively, if the bill is overturned and TikTok is allowed to operate in a structure substantially similar to its current structure, this question resolves N/A and all traders get their mana back.

Fine print: The law gives TikTok up to a year to be divested. If this timeline is extended due to legal challenges, the question won't resolve until ByteDance makes (or is forced to make) its final decision. For example, if ByteDance disallows U.S. users from accessing TikTok but they have credible ongoing legal proceedings to overturn the law, this question will stay open until the legal disputes are resolved and TikTok finalizes its decision.

TikTok

🇺🇸 US Politics

This market resolves to the streaming service that has higher viewership in Nielsen's April 2024 viewership report (whatever looks similar to that link).

🎥 Movies

💼  Business

🎨 Culture

Companion paper here with description of synthesis:

Updating to answer some more common questions. If anything here differs from the 2023-07-27 update, this update should supersede that one. I believe what I have written below maintains the spirit of the market and preserves its predictivity against weird edge cases. 

This question is specifically about LK-99. In the absence of overwhelming expert consensus (see below about resolution caveats), adding materials not in the original synthesis or characterization (replacing Cu with Au, Ag, etc) will not count toward a YES resolution. 

original XRD is OK, up to a point. I think I would be more accepting of this is it’s convincingly shown that the impurities are important to the superconductivity of LK-99. 

Synthesis of LK-99 by other means is totally fine, as long as the material characterization satisfies the criteria above. 

The original intent of this market was to bet on whether LK-99 exhibited the hallmarks of traditional superconductivity, but there have been a few questions on what happens if LK-99 is regions of 1D superconductivity in an insulating matrix, so it doesn’t display the expected R = 0 behavior below Tc. 

At this point, I’m going to hew to the original resolution criteria, in order of importance. 

While superconductor-insulator transitions have been observed, I’m not really familiar with them. Additionally, we don’t have good evidence that it is impossible to observe R=0 in LK-99, since at least 

 measurements show decreasing resistivity as a function of temperature. If we end up in a world where LK-99 convincingly displays a Meissner effect, but doesn’t show the expected resistivity behavior, I will solicit opinions from subject matter experts. After all, it would seem perverse to resolve NO if the majority of superconductivity experts think this is a YES.

I reserve the right to resolve to a probability if the experts I ask are split on the question of superconductivity.

I reserve the right to resolve according to whatever Wikipedia says about this material by Jan 1, 2025, especially if the experts I ask blow me off.

I reserve the right to revisit the resolution criteria if LK-99 or LK-99-like materials end up rewriting what we thought we knew about superconductivity. 

********************************************

Pasting my comment on resolution criteria

We're clearly all here because a) it's fun, and b) we're interested in room temperature superconductivity, not whether some other experimental group gets the same kinda sus data as the original paper. So, when I write 'replicate' in the question I am specifically asking: is the room temperature, ambient pressure superconductivity of the compound LK-99 convincingly demonstrated?

Specifically, replications should convincingly demonstrate:

If synthesizing the compound, there should be evidence that they did make something essentially the same as what is reported in the original paper. 

 not be required for a quantum well superconductor. I think I see how this could be the case. Willing to adjust this criterion after receiving more info from relevant theorists/experimentalists. 

I don't intend to require that replications be published in a peer-reviewed journal. The arXiv is sufficient for me. However, I do intend to wait a few weeks/months to resolve so that any pre-print can be adequately investigated for data manipulation, fraud, etc. In my utopia, labs that claim to have confirmed/disconfirmed this effect would also publish their raw data with their arXiv submissions, but I'm not holding my breath. 

Since high Tc superconductivity is not my specific field of expertise, I'm willing to defer to a consensus of subject matter experts on whether a pre-print is convincing or not, and I am willing to contact some beyond the usual twitter personalities.

Science

superconductivity

LK-99

2023 is trending to be the hottest year on record.

🪸 Climate

Global Warming 

🧪 Science

'2':4 '7':6 'leg':3 'may':5 'psg':1 'win':2

Borussia Dortmund wins ⚽ (Leg 2 | May 7)

'2':5 '7':7 'borussia':1 'dortmund':2 'leg':4 'may':6 'win':3

'2':7 '7':9 'draw':5 'end':2 'leg':6 'match':1 'may':8

Over 2.5 goals scored ⚽ (Leg 2 | May 7)

'2':6 '2.5':2 '7':8 'goal':3 'leg':5 'may':7 'score':4

Over 4.5 goals scored ⚽ (Leg 2 | May 7)

'2':6 '4.5':2 '7':8 'goal':3 'leg':5 'may':7 'score':4

Borussia Dortmund first to score ⚽ (Leg 2 | May 7)

'2':7 '7':9 'borussia':1 'dortmund':2 'first':3 'leg':6 'may':8 'score':5

'2':6 '7':8 'goal':2 'leg':5 'may':7 'score':4

Over 7.5 corner kicks ⚽ (Leg 2 | May 7)

'2':6 '7':8 '7.5':2 'corner':3 'kick':4 'leg':5 'may':7

Over 3.5 yellow cards given ⚽ (Leg 2 | May 7)

'2':7 '3.5':2 '7':9 'card':4 'given':5 'leg':6 'may':8 'yellow':3

Penalty kick is missed ⚽ (Leg 2 | May 7)

'2':6 '7':8 'kick':2 'leg':5 'may':7 'miss':4 'penalti':1

Match is tied at halftime ⚽ (Leg 2 | May 7)

'2':7 '7':9 'halftim':5 'leg':6 'match':1 'may':8 'tie':3

Both teams score 1+ goals in 2nd half ⚽ (Leg 2 | May 7)

'1':4 '2':10 '2nd':7 '7':12 'goal':5 'half':8 'leg':9 'may':11 'score':3 'team':2

Goal is scored during stoppage time ⚽ (Leg 2 | May 7)

'2':8 '7':10 'goal':1 'leg':7 'may':9 'score':3 'stoppag':5 'time':6

Sancho scores 1+ goals ⚽ (Leg 2 | May 7)

'1':3 '2':6 '7':8 'goal':4 'leg':5 'may':7 'sancho':1 'score':2

Füllkrug scores 1+ goals ⚽ (Leg 2 | May 7)

'1':3 '2':6 '7':8 'füllkrug':1 'goal':4 'leg':5 'may':7 'score':2

'1':3 '2':6 '7':8 'goal':4 'leg':5 'may':7 'reus':1 'score':2

Hummels scores 1+ goals ⚽ (Leg 2 | May 7)

'1':3 '2':6 '7':8 'goal':4 'hummel':1 'leg':5 'may':7 'score':2

Adeyemi scores 1+ goals ⚽ (Leg 2 | May 7)

'1':3 '2':6 '7':8 'adeyemi':1 'goal':4 'leg':5 'may':7 'score':2

Dembélé scores 1+ goals ⚽ (Leg 2 | May 7)

'1':3 '2':6 '7':8 'dembélé':1 'goal':4 'leg':5 'may':7 'score':2

Mbappé scores 1+ goals ⚽ (Leg 2 | May 7)

'1':3 '2':6 '7':8 'goal':4 'leg':5 'may':7 'mbappé':1 'score':2

Kang-in scores 1+ goals ⚽ (Leg 2 | May 7)

'1':5 '2':8 '7':10 'goal':6 'kang':2 'kang-in':1 'leg':7 'may':9 'score':4

Ramos scores 1+ goals ⚽ (Leg 2 | May 7)

'1':3 '2':6 '7':8 'goal':4 'leg':5 'may':7 'ramo':1 'score':2

Vitinha scores 1+ goals ⚽ (Leg 2 | May 7)

'1':3 '2':6 '7':8 'goal':4 'leg':5 'may':7 'score':2 'vitinha':1

Hakimi gets a yellow card ⚽ (Leg 2 | May 7)

'2':7 '7':9 'card':5 'get':2 'hakimi':1 'leg':6 'may':8 'yellow':4

PSG first to score in 2nd half ⚽ (Leg 2 | May 7)

'2':9 '2nd':6 '7':11 'first':2 'half':7 'leg':8 'may':10 'psg':1 'score':4

'2':6 '7':8 'card':4 'get':2 'leg':5 'may':7 'player':1 'red':3

Player scores a hat trick ⚽ (Leg 2 | May 7)

'2':7 '7':9 'hat':4 'leg':6 'may':8 'player':1 'score':2 'trick':5

Match ends in a draw (regular time) ⚽ (Leg 2 | May 7)

'2':9 '7':11 'draw':5 'end':2 'leg':8 'match':1 'may':10 'regular':6 'time':7

Match goes to extra time ⚽ (Leg 2 | May 7)

'2':7 '7':9 'extra':4 'goe':2 'leg':6 'match':1 'may':8 'time':5

Match ends with penalty shoot-out ⚽ (Leg 2 | May 7)

'2':9 '7':11 'end':2 'leg':8 'match':1 'may':10 'penalti':4 'shoot':6 'shoot-out':5

2024 UEFA Champions League Matches & Prop Bets

⚽ Soccer

🏟️ Sports

UEFA Champions League

⚽ 🏈 Football

Sports Betting

Resolves YES if Verstappen qualifies 1st, then goes on to win the Miami Grand Prix with the fastest lap by race end.

If Verstappen qualifies first but is moved down the order, that will not affect the ‘qualify first’ element of this market, and if he wins the race with the fastest lap after that, the market will still resolve YES.

Resolves N/A if Verstappen doesn’t attend qualifying and the race.

🏎️ Formula 1

Auto Racing

Motorsports

Resolves YES if the gap between Perez and Verstappen in qualifying is 0.350 seconds or lower at the Miami Grand Prix.

Resolves YES if Perez out-qualifies Verstappen with any gap of time.

'beat':3 'citi':2 'dortmund':4 'manchest':1

Real Madrid beats Manchester City ⚽ (Leg 1) Apr-9

'-9':9 '1':7 'apr':8 'beat':3 'citi':5 'leg':6 'madrid':2 'manchest':4 'real':1

Atlético de Madrid beats Borussia Dortmund ⚽ (Leg 1) Apr-10

'-10':10 '1':8 'apr':9 'atlético':1 'beat':4 'borussia':5 'de':2 'dortmund':6 'leg':7 'madrid':3

'-10':7 '1':5 'apr':6 'barcelona':3 'beat':2 'leg':4 'psg':1

Arsenal beats Bayern Munich ⚽ (Leg 1) Apr-9

'-9':8 '1':6 'apr':7 'arsenal':1 'bayern':3 'beat':2 'leg':5 'munich':4

Manchester City beats Real Madrid ⚽ (Leg 2) Apr-17

'-17':9 '2':7 'apr':8 'beat':3 'citi':2 'leg':6 'madrid':5 'manchest':1 'real':4

 Borussia Dortmund beats Atlético de Madrid ⚽ (Leg 2) Apr-16

'-16':10 '2':8 'apr':9 'atlético':4 'beat':3 'borussia':1 'de':5 'dortmund':2 'leg':7 'madrid':6

'-16':7 '2':5 'apr':6 'barcelona':1 'beat':2 'leg':4 'psg':3

Bayern Munich beats Arsenal ⚽ (Leg 2) Apr-17

'-17':8 '2':6 'apr':7 'arsenal':4 'bayern':1 'beat':3 'leg':5 'munich':2

Real Madrid beats Manchester City ⚽ Aggregate

'aggreg':6 'beat':3 'citi':5 'madrid':2 'manchest':4 'real':1

Atlético de Madrid beats  Borussia Dortmund ⚽ Aggregate

'aggreg':7 'atlético':1 'beat':4 'borussia':5 'de':2 'dortmund':6 'madrid':3

'aggreg':4 'barcelona':3 'beat':2 'psg':1

Arsenal beats Bayern Munich ⚽ Aggregate

'aggreg':5 'arsenal':1 'bayern':3 'beat':2 'munich':4

Borussia Dortmund beats PSG ⚽ Aggregate

'aggreg':5 'beat':3 'borussia':1 'dortmund':2 'psg':4

Bayern Munich beats Real Madrid ⚽ Aggregate

'aggreg':6 'bayern':1 'beat':3 'madrid':5 'munich':2 'real':4

FINAL 🚧 Borussia Dortmund/PSG beats Bayern Munich/Real Madrid

'bayern':5 'beat':4 'borussia':2 'dortmund/psg':3 'final':1 'madrid':7 'munich/real':6

'9':6 'april':5 'card':4 'get':2 'match':7 'player':1 'red':3

'4.5':2 '9':6 'april':5 'goal':3 'match':7 'score':4