OpenAI announces something called "GPT-4.5" or very close to it

'-4.5':6 'announc':2 'call':4 'close':9 'gpt':5 'openai':1 'someth':3

OpenAI announces something new related to Sora

'announc':2 'new':4 'openai':1 'relat':5 'someth':3 'sora':7

OpenAI announces something new related to Voice Engine

'announc':2 'engin':8 'new':4 'openai':1 'relat':5 'someth':3 'voic':7

OpenAI announces something new related to DALL-E

'announc':2 'dall':8 'dall-':7 'e':9 'new':4 'openai':1 'relat':5 'someth':3

'altman':2 'magic/magical/etc':4 'sam':1 'say':3

OpenAI announces something new related to music generation

'announc':2 'generat':8 'music':7 'new':4 'openai':1 'relat':5 'someth':3

Immediate or scheduled public rollout (even if gradual) of Sora

'even':6 'gradual':8 'immedi':1 'public':4 'rollout':5 'schedul':3 'sora':10

'chatbot':6 'gpt2':5 'gpt2-chatbot':4 'mention':2 'someon':1

Update on the partnership with Figure/humanoid robots

'figure/humanoid':6 'partnership':4 'robot':7 'updat':1

OpenAI announces an improvement to GPT-4 similiar to GPT-4-Turbo, which is not major enough to be called GPT-4.5 or GPT-5

'-4':7,11 '-4.5':22 '-5':25 'announc':2 'call':20 'enough':17 'gpt':6,10,21,24 'improv':4 'major':16 'openai':1 'similiar':8 'turbo':12

Temporary technical issues (e.g. mic, presentation, lights, ..., not working)

'e.g':4 'issu':3 'light':7 'mic':5 'present':6 'technic':2 'temporari':1 'work':9

Actual presentation (person talking, video, ...) starts before 10:05 AM

'05':9 '10':8 'actual':1 'person':3 'present':2 'start':6 'talk':4 'video':5

"Regulation" is mentioned, in a legal/policy context

'context':7 'legal/policy':6 'mention':3 'regul':1

A tangible product/device is demonstrated

'demonstr':5 'product/device':3 'tangibl':2

Someone from another (not OAI) company joins on stage

'anoth':3 'compani':6 'join':7 'oai':5 'someon':1 'stage':9

Coalition for Content Provenance and Authenticity Steering Committee mentioned

'authent':6 'coalit':1 'committe':8 'content':3 'mention':9 'proven':4 'steer':7

'mention':3 'partnership':2 'stackoverflow':1

'competitor':3 'googl':1 'mention':4 'search':2

Any mention of the redirect from chat.openai.com to chatgpt.com

'chat.openai.com':7 'chatgpt.com':9 'mention':2 'redirect':5

Mention of AI used by researchers (use cases, how it can help them, etc.)

'ai':3 'case':8 'etc':14 'help':12 'mention':1 'research':6 'use':4,7

The LMSys Chatbot Arena is referenced or shown 

'arena':4 'chatbot':3 'lmsys':2 'referenc':6 'shown':8

At least one specific industry mentioned as userbase or a use case for AI tools

'ai':14 'case':12 'industri':5 'least':2 'mention':6 'one':3 'specif':4 'tool':15 'use':11 'userbas':8

Someone says "patience" or "patient" (not as a noun)

'noun':9 'patienc':3 'patient':5 'say':2 'someon':1

Altman lied about what would be featured at the event in his X posts

'altman':1 'event':10 'featur':7 'lie':2 'post':14 'would':5 'x':13

In the market creator's opinion, the average reaction to the event by journalists is "mixed" or negative

'averag':8 'creator':4 'event':12 'journalist':14 'market':3 'mix':16 'negat':18 'opinion':6 'reaction':9

Any type of new software, AI or not, usable by anyone in the United States willing to pay, is released by OpenAI on the day of the event

'ai':6 'anyon':11 'day':25 'event':28 'new':4 'openai':22 'pay':18 'releas':20 'softwar':5 'state':15 'type':2 'unit':14 'usabl':9 'will':16

In the market creator's opinion, a live demonstration of an OpenAI product is marred by a bug

'bug':18 'creator':4 'demonstr':9 'live':8 'mar':15 'market':3 'openai':12 'opinion':6 'product':13

Any of this words: "Agent", "Agents" or "Agentic" are mentioned

Sam Altman says something along the lines of “what we have now is nothing/laughable compared to what is coming later this year/next year.” 

'along':5 'altman':2 'come':19 'compar':15 'later':20 'line':7 'nothing/laughable':14 'sam':1 'say':3 'someth':4 'year':23 'year/next':22

The start of the stream is delayed at least 5 minutes

'5':10 'delay':7 'least':9 'minut':11 'start':2 'stream':5

GPT-4 becomes cheaper (API price reduced)

'-4':2 'api':5 'becom':3 'cheaper':4 'gpt':1 'price':6 'reduc':7

Presenters give away $10,000 or more in total of free stuff as part of presentation

'000':5 '10':4 'away':3 'free':11 'give':2 'part':14 'present':1,16 'stuff':12 'total':9

The official stream pictures someone wearing a mask (for COVID-19 prevention)

'-19':11 'covid':10 'mask':8 'offici':2 'pictur':4 'prevent':12 'someon':5 'stream':3 'wear':6

An OpenAI employee makes a major gaffe (in the market creator's opinion) that is featured on the front page of the New York Times

'creator':11 'employe':3 'featur':16 'front':19 'gaff':7 'major':6 'make':4 'market':10 'new':23 'openai':2 'opinion':13 'page':20 'time':25 'york':24

Something related to phone calls and ChatGPT

'call':5 'chatgpt':7 'phone':4 'relat':2 'someth':1

Any mention of deepfakes (including any reference to audio or video identity spoofing)

'audio':9 'deepfak':4 'ident':12 'includ':5 'mention':2 'refer':7 'spoof':13 'video':11

Films or the film industry (or related topics) mentioned

'film':1,4 'industri':5 'mention':9 'relat':7 'topic':8

Announcement of software that can automatically make phone calls on user's behalf

'announc':1 'automat':6 'behalf':13 'call':9 'make':7 'phone':8 'softwar':3 'user':11

Announcement of generally intelligent model using speech as input and speech as output, with no intermediate models

'announc':1 'general':3 'input':9 'intellig':4 'intermedi':16 'model':5,17 'output':13 'speech':7,11 'use':6

Greg Brockman creates a new AI voice assistant/personality during the demo and gives it a custom name

'ai':6 'assistant/personality':8 'brockman':2 'creat':3 'custom':16 'demo':11 'give':13 'greg':1 'name':17 'new':5 'voic':7

Someone says “magic” during the livestream

'livestream':6 'magic':3 'say':2 'someon':1

OpenAI announces improved memory abilities for ChatGPT

'abil':5 'announc':2 'chatgpt':7 'improv':3 'memori':4 'openai':1

'altman':2 'blue':6 'sam':1 'shirt':9 't-shirt':7 'wear':4

'altman':2 'pink':6 'sam':1 'wear':4 'wig':7

Something with an anthropomorphic name is announced

'announc':7 'anthropomorph':4 'name':5 'someth':1

The word "or" is said in the first 3 minutes of the livestream (with someone speaking)

'3':9 'first':8 'livestream':13 'minut':10 'said':5 'someon':15 'speak':16 'word':2

Llama, Claude, or Gemini mentioned by name

'claud':2 'gemini':4 'llama':1 'mention':5 'name':7

A less censored model, or an option to make existing models less censored, is announced.

'announc':15 'censor':3,13 'exist':10 'less':2,12 'make':9 'model':4,11 'option':7

On X, during May 13, Eliezer Yudkowsky criticizes OpenAI or its employees for endangering human civilization

'13':5 'civil':16 'critic':8 'eliez':6 'employe':12 'endang':14 'human':15 'may':4 'openai':9 'x':2 'yudkowski':7

A model is announced with a name like GPT-[Word] or GPT-4-[Word]

'-4':13 'announc':4 'gpt':9,12 'like':8 'model':2 'name':7 'word':10,14

Something is announced which is not specifically a search engine but which could still be reasonably viewed as a significant new threat to Google Search 

'announc':3 'could':13 'engin':10 'googl':24 'new':21 'reason':16 'search':9,25 'signific':20 'someth':1 'specif':7 'still':14 'threat':22 'view':17

Something related to live translation with voice in/out

'in/out':8 'live':4 'relat':2 'someth':1 'translat':5 'voic':7

Someone says "Accelerate", "Acceleration", "Accelerated", "Accelerating", or "E/Acc"

'acceler':3,4,5,6 'e/acc':8 'say':2 'someon':1

There is a reveal in the form of "Actually, this [voice/video/script] your thought came from a person was made with generative AI"

'actual':9 'ai':22 'came':14 'form':7 'generat':21 'made':19 'person':17 'reveal':4 'thought':13 'voice/video/script':11

OpenAI does not reveal GPT-5, but does say we should expect it before 2025 (Includes phrasing like "Our next flagship model" or "GPT-4's successor")

'-4':25 '-5':6 '2025':15 'expect':12 'flagship':21 'gpt':5,24 'includ':16 'like':18 'model':22 'next':20 'openai':1 'phrase':17 'reveal':4 'say':9 'successor':27

'3':1 'one':6 'peopl':2 'stage':4 'time':7

'basic':4 'comput':5 'say':2 'someon':1 'univers':3

OpenAI announces an existing model will be open-sourced

'announc':2 'exist':4 'model':5 'open':9 'open-sourc':8 'openai':1 'sourc':10

A rumour is mentioned/addressed and confirmed untrue

'confirm':6 'mentioned/addressed':4 'rumour':2 'untru':7

Something related to education/academia/students (of any age), or framed as useful for users for these purposes

'age':7 'education/academia/students':4 'frame':9 'purpos':16 'relat':2 'someth':1 'use':11 'user':13

In general, resolution will err towards resolving things YES in ambiguous situations. If it seems like something might count according to the spirit of the answer but the letter is unclear, it will probably count. If the event is postponed, this market will be extended until the event occurs.

Prop Bets

🤖 AI

🤖 OpenAI

🖥️ Technology

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

'announc':4 'chines':1 'gpt':3 'languag':2

'announc':4 'gpt':3 'korean':1 'languag':2

Chat GPT-5 release date or timeframe information given

'-5':3 'chat':1 'date':5 'given':9 'gpt':2 'inform':8 'releas':4 'timefram':7

Something related to DALL-E / Image generation

'dall':5 'dall-':4 'e':6 'generat':8 'imag':7 'relat':2 'someth':1

Someone is announced in a new role / position 

'announc':3 'new':6 'posit':8 'role':7 'someon':1

Announcing a partnership with another company

'announc':1 'anoth':5 'compani':6 'partnership':3

'ai':4 'pornographi':5 'relat':2 'someth':1

Something related to economics / finance / stock trading / money management 

'econom':4 'financ':5 'manag':9 'money':8 'relat':2 'someth':1 'stock':6 'trade':7

'chatbot':5 'detail':1 'gpt2':4 'gpt2-chatbot':3

Partnership with the government of a country

Something Manifold finds disappointing (in a poll created after the announcement)

'announc':11 'creat':8 'disappoint':4 'find':3 'manifold':2 'poll':7 'someth':1

A new physical product in the same niche as the Rabbit R1 or the Humane AI Pin

'ai':16 'human':15 'new':2 'nich':8 'physic':3 'pin':17 'product':4 'r1':12 'rabbit':11

Announcement of model with a new "personality", less refusals, or different RLHF

'announc':1 'differ':11 'less':8 'model':3 'new':6 'person':7 'refus':9 'rlhf':12

Partnership on data collection with other companies

'collect':4 'compani':7 'data':3 'partnership':1

'-4':2,9 'cheaper':5 'gpt':1,8 'lite':3 'version':6

Access to a GPT-4 level model is now free

'-4':5 'access':1 'free':10 'gpt':4 'level':6 'model':7

The ability for ChatGPT to make or be integrated with phone calls

'abil':2 'call':12 'chatgpt':4 'integr':9 'make':6 'phone':11

ChatGPT with a knowledge cutoff that is always recent (within 48h)

'48h':11 'alway':8 'chatgpt':1 'cutoff':5 'knowledg':4 'recent':9 'within':10

Live translation through voice input/output

'input/output':5 'live':1 'translat':2 'voic':4

https://twitter.com/sama/status/1788989777452408943

Sam Altman

🤖❤️ AI Alignment

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

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".

GPT-5 Speculation

ChatGPT

The bet resolves YES if OpenAI offers GPT-4 (including variants like "GPT-4-lite" or similar) for free to ChatGPT users in 2024 under any of these conditions:

ChatGPT 

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.

Technical AI Timelines

AI Alignment 

Mechanistic interpretability 

New York Mets beat Milwaukee Brewers - Mar 29

'29':8 'beat':4 'brewer':6 'mar':7 'met':3 'milwauke':5 'new':1 'york':2

Baltimore Orioles beat Los Angeles Angels - Mar 28

'28':8 'angel':5,6 'baltimor':1 'beat':3 'los':4 'mar':7 'oriol':2

Philadelphia Phillies beat Atlanta Braves - Mar 29

'29':7 'atlanta':4 'beat':3 'brave':5 'mar':6 'philadelphia':1 'philli':2

Chicago White Sox beat Detroit Tigers - Mar 28

'28':8 'beat':4 'chicago':1 'detroit':5 'mar':7 'sox':3 'tiger':6 'white':2

Cincinnati Reds beat Washington Nationals - Mar 28

'28':7 'beat':3 'cincinnati':1 'mar':6 'nation':5 'red':2 'washington':4

Houston Astros beat New York Yankees - Mar 28

'28':8 'astro':2 'beat':3 'houston':1 'mar':7 'new':4 'yanke':6 'york':5

Kansas City Royals beat Minnesota Twins - Mar 28

'28':8 'beat':4 'citi':2 'kansa':1 'mar':7 'minnesota':5 'royal':3 'twin':6