“Bet with Miles Brundage at 10:1 odds: if AI systems can perform 8 of the 10 specified tasks by the end of 2027, as determined by our panel of judges, Gary will donate $2,000 to a charity of Miles' choice; if AI can do fewer than 8, Miles will donate $20,000 to a charity of Gary's choice.”
Cognitive scientist, author and generative-AI skeptic; Professor Emeritus of Psychology and Neural Science at NYU
Tasks include watching an unseen movie and following plot twists, constructing bug-free code of more than 10,000 lines from natural-language spec, and similar; tracked on Metaculus and Manifold. Marcus's separate end-of-2025 checkpoint (no system solving more than 4 of the 10) was judged correct, but the bet itself does not resolve until after 2027.
“$100,000 says AGI will not arrive by 2029: if any single AI system satisfies at least three of my five criteria (e.g., watching a movie and reliably explaining plot and motivations; reading a novel and answering questions about it; working as a competent cook in an arbitrary kitchen; writing bug-free code of more than 10,000 lines from natural-language spec; deriving mathematical proofs from natural language) by then, Musk wins; otherwise I do.”
Cognitive scientist, Professor Emeritus of Psychology and Neural Science at NYU, founder of Geometric Intelligence and Robust.AI
Musk never accepted; others (including Vivek Wadhwa, who matched it) raised the wager to $500,000, and Kevin Kelly offered to host it at Long Bets. Metaculus opened markets on several of the five criteria. Resolution due after 2029.
“Dear @elonmusk, $1 million says your latest prediction – that AI will be smarter than any individual human by the end of 2025 – is wrong. Game?”
Cognitive scientist, Professor Emeritus of Psychology and Neural Science at NYU, author of 'Taming Silicon Valley'
Said Apr 9, 2024Deadline Dec 31, 2025Gary Marcus on X, 9 April 2024 No AI system exceeded the best individual humans across intellectual work by the end of 2025; even Marcus's more general 'we will not see AGI in 2025' prediction was graded correct by third-party reviewers. Musk never accepted the bet (which was raised to $10M by investor Damion Hankejh) despite press inquiries.
“We will not see artificial general intelligence this year [2025], despite claims by Elon Musk to the contrary. (People will also continue to play games to weaken the definition or even try to define it in financial rather than scientific terms.)”
Cognitive scientist, Professor Emeritus of Psychology and Neural Science at NYU
No AGI arrived in 2025; independent year-end gradings of dated 2025 AI predictions scored this one correct, and by late 2025 even prominent optimists were pushing AGI timelines out by years.
“No single system will solve more than 4 of the AI 2027 Marcus-Brundage tasks by the end of 2025. I wouldn't be shocked if none were reliably solved by the end of the year.”
Cognitive scientist, Professor Emeritus of Psychology and Neural Science at NYU
A year-end review of 2025 AI predictions judged this correct, concluding that at most one of the ten tasks (a reading-comprehension-style item) was arguably passed, and certainly not four by any single system.
“Profits from AI models will continue to be modest or nonexistent (chip-making companies will continue to do well though, in supplying hardware to the companies that build the models; shovels will continue to sell well throughout the gold rush.)”
Cognitive scientist, Professor Emeritus of Psychology and Neural Science at NYU
Through 2025 the leading model developers continued to run large operating losses while revenue growth was funded by external capital, and the hardware suppliers — Nvidia above all — captured the profits; the 'shovel-sellers do best' pattern held. Both halves of the prediction came true.
“AI "Agents" will be endlessly hyped throughout 2025 but far from reliable, except possibly in very narrow use cases.”
Cognitive scientist, Professor Emeritus of Psychology and Neural Science at NYU
2025 was saturated with agent marketing while reliability lagged: MIT's NANDA work produced the widely-cited finding that about 95% of internally built enterprise genAI pilots delivered no measurable return, McKinsey-style surveys found very few agents reaching production, and teams that shipped did so by narrowing scope. Independent year-end gradings scored this prediction correct.
“Humanoid robotics will see a lot of hype, but nobody will release anything to remotely as capable as Rosie the Robot. Motor control may be impressive, but situational awareness and cognitive flexibility will remain poor.”
Cognitive scientist, Professor Emeritus of Psychology and Neural Science at NYU
Humanoid-robot announcements and demos proliferated in 2025, including Tesla's Optimus, but no shipped system approached a general-purpose household robot; impressive locomotion and manipulation demos continued to coexist with brittle autonomy and heavy teleoperation or scripting.
“The US will continue to have very little regulation protecting its consumers from the risks of generative AI. When it comes to regulation, much of the world will increasingly look to Europe for guidance.”
Cognitive scientist, Professor Emeritus of Psychology and Neural Science at NYU
No comprehensive federal generative-AI consumer-protection statute was enacted in 2025; US federal policy moved toward deregulation and promotion of AI build-out, leaving substantive rules to individual states and to the EU AI Act.
“GPT-4 by itself won't be a general purpose artificial general intelligence capable of taking on arbitrary tasks. Without external aids it won't be able to drive a car reliably; it won't be able to reliably guide a robot like Optimus to be anything like as versatile as Rosie the Robot. It will remain a turbocharged pastiche generator; reasoning about the physical, psychological and mathematical world will still be unreliable; and 'alignment' will continue to be a critical, unsolved problem.”
Cognitive scientist, Professor Emeritus of Psychology and Neural Science at NYU
GPT-4 shipped in March 2023 and was not a general-purpose AGI: it could not drive a car or control a humanoid robot, hallucinations and reasoning errors persisted, and alignment remained unsolved. Marcus's seven GPT-4 predictions were widely regarded — including by later critics tracking his record — as having held, which is why he reissued the same text for GPT-5 in 2025.
“GPT-5 by itself won't be a general purpose artificial general intelligence capable of taking on arbitrary tasks. Without external aids it won't be able to drive a car reliably; it won't be able to reliably guide a robot like Optimus to be anything like as versatile as Rosie the Robot. Reasoning about the physical, psychological and mathematical world will still be unreliable, fluent hallucinations will still be common, and 'alignment' will continue to be a critical, unsolved problem.”
Cognitive scientist, Professor Emeritus of Psychology and Neural Science at NYU
GPT-5 launched in August 2025 and was not an AGI: it could not drive a car or run a humanoid robot, hallucinations and reasoning failures persisted, and alignment remained open. The rollout was poorly received — Sam Altman conceded OpenAI had 'totally screwed up' the launch — and Marcus's assessment that it 'came late and fell short' was echoed by many observers.
“AGI really could disrupt the world. But GPT-5 is not going to do any of that; it's not going to be smart enough to revolutionize fusion; it's not going to be smart enough to revolutionize nanotech, and starships most definitely aren't going to be here in 2024.”
Cognitive scientist, NYU professor emeritus, AI critic and author
GPT-5 did not even ship until August 2025, and through the end of 2024 there was no AI-driven revolution in fusion or nanotech and no interstellar starships. Marcus's deflationary call was borne out.
“They no longer have anything like a clear technical lead. GPT-5 is unlikely to be ahead of the pack for more than a couple months. (And Grok 4 Heavy is already better on the ARC-AGI-2 measure.)”
Cognitive scientist, NYU professor emeritus, AI critic and author
GPT-5 never held an uncontested lead: Grok 4 already beat it on ARC-AGI-2 at launch, and competing frontier releases within weeks to a few months (Anthropic's Claude 4.5 line in late September and Google's Gemini 3 Pro in November 2025) took the top of major leaderboards, prompting OpenAI's own GPT-5.1 refresh.