What next · the long view

Timelines to the singularity

People have been dating the arrival of machines smarter than us since the 1960s. This page puts their calls on one line, from the mathematicians who first imagined an intelligence explosion to this summer's revised timelines, and reads each against what this site measures tonight. Nobody on it is crowned: the page shows who said what, when they said it, what would prove them wrong, and which way the evidence leans so far.

I · the timeline

The dates people have dared to give, on one line

Each row is a milestone; each mark, one forecaster's year for it. The glyph is how the forecast reads tonight.

When each forecaster expected each milestonea mark sits at the year forecast, a bar spans a stated range; hover for who said it, tap to open it

A machine that out-codes the best programmers

●○●○○

AI that does AI research

◔○○○○○

General or transformative AI

Superintelligence

Explosive growth

Loss of control

Robots in the physical world

○○○

The build-out

◔●●●●○●◔○○

Minds, uploads and longer lives

◔◔○○○

Imagined, not forecast

◇◇◇◇◇◇◇◇◇◇

HappeningSlower than saidToo early to tellthe range statedset in a story, not a forecastlighter: a step toward the milestone, not a date for it

The scale is not even: the half century from 1950 and the half century from 2050 are squeezed so the years around now get most of the width (dashed lines mark where it changes). A call that gives odds, or bets against a date, without naming a year is listed under the chart. A forecast is placed at the start of its year, so one due by the end of this year sits left of today while its window is still open. Shaded: the years still to come.

Every forecast placed on the chart

Milestone	Who	Said	Years forecast	Tonight
A machine that out-codes the best programmers	Ajeya Cotra	2026	by 2026	Happening
A machine that out-codes the best programmers	AI Futures Project	2025	2027	Too early to tell
A machine that out-codes the best programmers	AI Futures Project (Kokotajlo, Lifland, Jurkovic et al.)	2025	by 2027	Happening
A machine that out-codes the best programmers	Daniel Kokotajlo	2026	2027	Too early to tell
A machine that out-codes the best programmers	Eli Lifland	2026	2030	Too early to tell
AI that does AI research	AI Futures Project	2025	by 2026	Slower than said
AI that does AI research	Sam Altman	2025	by 2026	Too early to tell
AI that does AI research	Leopold Aschenbrenner	2024	by 2027	Too early to tell
AI that does AI research	AI Futures Project	2025	by 2027	Too early to tell
AI that does AI research	OpenAI	2026	by 2028	Too early to tell
AI that does AI research	William MacAskill and Fin Moorhouse	2025	2035–2045	Too early to tell
General or transformative AI	Ray Solomonoff	1985	1994–2023	Slower than said
General or transformative AI	Sam Altman	2026	by 2026	Too early to tell
General or transformative AI	AI Futures Project	2025	by 2027	Too early to tell
General or transformative AI	Mustafa Suleyman	2023	2026–2028	Too early to tell
General or transformative AI	Leopold Aschenbrenner	2024	2027–2028	Too early to tell
General or transformative AI	Dario Amodei	2026	2027–2028	Too early to tell
General or transformative AI	Daniel Kokotajlo	2026	2028	Too early to tell
General or transformative AI	Ray Kurzweil	2002	by 2029	Too early to tell
General or transformative AI	Demis Hassabis	2026	2029–2031, most likely 2030	Too early to tell
General or transformative AI	Eli Lifland	2026	2032	Too early to tell
General or transformative AI	Dario Amodei	2026	by 2036	Too early to tell
General or transformative AI	Tamay Besiroglu	2025	2040	Too early to tell
General or transformative AI	Tom Davidson	2023	2030–2070, most likely 2043	Too early to tell
General or transformative AI	Ege Erdil	2025	2045	Too early to tell
General or transformative AI	Ajeya Cotra	2020	2040–2080, most likely 2050	Too early to tell
General or transformative AI	Holden Karnofsky	2021	2036–2100, most likely 2060	Too early to tell
General or transformative AI	Joe Carlsmith	2022	by 2070	Too early to tell
General or transformative AI	David Chalmers	2010	2040–2100	Too early to tell
Superintelligence	I. J. Good	1965	by 2000	Slower than said
Superintelligence	AI Futures Project	2025	by 2027	Slower than said
Superintelligence	Daniel Kokotajlo	2026	2029	Too early to tell
Superintelligence	Vernor Vinge	1993	2005–2030	Too early to tell
Superintelligence	Leopold Aschenbrenner	2024	by 2030	Too early to tell
Superintelligence	Nick Bostrom	2008	by 2033	Too early to tell
Superintelligence	Eli Lifland	2026	2033	Too early to tell
Superintelligence	Eliezer Yudkowsky	1996	by 2035	Too early to tell
Superintelligence	Sam Altman	2024	2030–2035	Too early to tell
Superintelligence	Ray Kurzweil	2005	2045	Too early to tell
Superintelligence	Stuart Russell	2019	by 2100	Too early to tell
Explosive growth	Robin Hanson	2001	2025	Slower than said
Loss of control	Andrew Critch	2024	2030–2050	Too early to tell
Robots in the physical world	Sam Altman	2025	by 2027	Too early to tell
Robots in the physical world	AI Futures Project	2025	2028–2030	Too early to tell
Robots in the physical world	Hans Moravec	1998	by 2048	Too early to tell
The build-out	Ray Kurzweil	2001	by 2023	Slower than said
The build-out	AI Futures Project	2025	by 2025	Happening
The build-out	Daniel Kokotajlo	2021	by 2026	Happening
The build-out	Leopold Aschenbrenner	2024	by 2026	Happening
The build-out	Leopold Aschenbrenner	2024	by 2027	Happening
The build-out	Leopold Aschenbrenner	2024	by 2028	Too early to tell
The build-out	Hans Moravec	1998	2020–2029	Happening
The build-out	Ray Kurzweil	2001	by 2030	Slower than said
The build-out	Leopold Aschenbrenner	2024	by 2030	Too early to tell
The build-out	Yafah Edelman and Anson Ho	2025	2035	Too early to tell
Minds, uploads and longer lives	Ray Kurzweil	2001	by 2030	Slower than said
Minds, uploads and longer lives	Ray Kurzweil	2024	by 2030	Slower than said
Minds, uploads and longer lives	Ray Kurzweil	2024	2030–2039	Too early to tell
Minds, uploads and longer lives	Anders Sandberg and Nick Bostrom	2008	by 2050	Too early to tell
Minds, uploads and longer lives	Robin Hanson	2016	by 2116	Too early to tell

Calls that give odds, or doubt a date, but name no year

AI that does AI research: Too early to tellAjeya Cotra gave about one chance in ten that AI fully automates AI research and development, matching human researchers, by the end of 2026. · 2026
General or transformative AI: Too early to tellExtrapolating past progress in AI subfields, Robin Hanson estimated that broadly human-level AI without brain emulation would take more than two to four centuries. · 2016; HappeningGwern argued that once a scalable architecture exists, simply training ever bigger neural networks will keep yielding more sophisticated general abilities without fundamentally new designs. · 2020; HappeningGary Marcus bet Miles Brundage that by the end of 2027 AI will still fail at least three of ten demanding tasks, from understanding films to Nobel-level discovery. · 2024; Too early to tellYann LeCun says language models will not reach human-level intelligence, and that within five years we may at best know we are on the right path. · 2026
Superintelligence: HappeningRobin Hanson put less than one chance in a hundred on a single hand-coded AI leaping, without warning, from near impotence to overwhelming power. · 2008; Too early to tellEliezer Yudkowsky argued that an AI improving itself would most likely either stall or take off explosively, since a gradual, human-paced takeoff would need implausibly precise diminishing returns. · 2008; Too early to tellNick Bostrom judged it unlikely that the climb from human-level AI to superintelligence would take decades, and more likely that it would be fast or moderate. · 2014; HappeningFrançois Chollet argued that an intelligence explosion cannot happen, and that the first superhuman AI would be just another step in steady, roughly linear progress. · 2017; Too early to tellTom Davidson's best guess is that AI goes from matching human professionals at almost all cognitive work to far surpassing them in less than a year. · 2023; Too early to tellPaul Christiano put about one chance in seven on AI able to build a Dyson sphere by 2030, and about two chances in five by 2040. · 2023; Too early to tellPaul Christiano gave even odds on whether an intelligence explosion driven by software improvements alone is possible at all. · 2023; Too early to tellWilliam MacAskill and Fin Moorhouse give roughly even odds that AI improving its own software sets off accelerating growth in AI capability. · 2025; Too early to tellEge Erdil and Tamay Besiroglu doubt that AI improving its own software alone will set off an intelligence explosion, holding that compute and experiments bound progress. · 2025
Explosive growth: Too early to tellRobin Hanson's growth model implied that once machines can stand in for most human labour, world output would grow about ten times faster, doubling roughly every eighteen months. · 2001; Too early to tellPaul Christiano put about three chances in ten on a fast takeoff, expecting world output to finish a doubling over four years before any single-year doubling begins. · 2018; HappeningTyler Cowen guesses AI will raise annual economic growth by only about half a percentage point, because slow-moving sectors such as government, health and education act as bottlenecks. · 2025; Too early to tellWilliam MacAskill and Fin Moorhouse think it more likely than not that, once AI matches human researchers, a hundred years of technological progress follows in under a decade. · 2025; Too early to tellTamay Besiroglu expects explosive economic growth, above thirty percent a year, once AI can flexibly stand in for human workers. · 2025; Too early to tellExperts surveyed by Ezra Karger and colleagues expect that if AI outperforms humans on many tasks by 2030, US growth would rise to around four percent a year, against about two and a half percent in their overall forecast. · 2026; Too early to tellPhil Trammell argues that limits on splitting research across many AI workers could delay a technological singularity, though automating research would probably still produce an explosion. · 2026
Loss of control: Too early to tellStephen Omohundro argued that any sufficiently advanced goal-seeking AI, unless deliberately designed otherwise, will resist being shut down, copy itself and acquire resources. · 2008; Too early to tellPaul Christiano expected that, without solving intent alignment, AI failure would come as a gradual loss of human steering to optimised proxies, possibly ending in a sudden cascade. · 2019; Too early to tellToby Ord estimated about one chance in ten that unaligned AI causes an existential catastrophe within the next hundred years, the largest single risk he assessed. · 2020; Too early to tellJoe Carlsmith raised his estimate that power-seeking AI causes an existential catastrophe by 2070 from about one in twenty to more than one in ten. · 2022; Too early to tellCarl Shulman puts the chance that AI takes control of humanity's future at between a fifth and a quarter, depending on the day. · 2023; Too early to tellJan Kulveit and co-authors argue that if AI keeps displacing human labour, thought and consent across economy, culture and state, humans will probably lose effective control without any takeover. · 2025; Too early to tellDaniel Kokotajlo puts his chance of catastrophe from AI at around seven in ten, a figure he calls infamously high. · 2025; Too early to tellScott Alexander puts his chance of AI doom at about one in five, the lowest estimate among the authors of AI 2027. · 2025; Too early to tellEliezer Yudkowsky and Nate Soares hold that if anyone builds artificial superintelligence with anything like today's methods and understanding, everyone on Earth will die. · 2025
Robots in the physical world: Too early to tellCarl Shulman expects that once AI directs a converted industrial base, the robot population will double in under a year, and then within months. · 2023
Minds, uploads and longer lives: Too early to tellRobin Hanson argued that if brain uploads arrive before other human-level AI, cheap copying would multiply them and push wages toward subsistence even as total wealth grows. · 1994; Too early to tellDario Amodei predicts that once powerful AI exists, biology and medicine will fit the coming half-century to century of progress into five to ten years. · 2024

How far apart the dated forecasts of each milestone are, and how many give no yearChart

A machine that out-codes the best programmers3 with a year · 0 without

2027 to 2030

AI that does AI research5 with a year · 1 without

2026 to 2045

General or transformative AI18 with a year · 4 without

2023 to 2100

Superintelligence11 with a year · 9 without

2000 to 2100

Explosive growth1 with a year · 7 without

2025different claims on a theme

Loss of control1 with a year · 9 without

2050different claims on a theme

Robots in the physical world3 with a year · 1 without

2027 to 2048different claims on a theme

The build-out10 with a year · 0 without

2023 to 2035different claims on a theme

Minds, uploads and longer lives5 with a year · 2 without

2030 to 2116different claims on a theme

a forecast, at the year the timeline places itfrom the earliest year a forecast is placed at to the latestshaded: the years still to come

Each row counts forecasts, not people: a writer with more than one dated forecast on a row is counted for each, and the authors of AI 2027 appear both together and under their own names. A forecast that gives a range is placed, as on the timeline above, at its most likely year or else at the last year of its range, so a row does not show how wide each forecast is, and a row can begin later than the earliest year a forecast names. A step toward a milestone (a forecast of an early sign of it, not of the milestone itself) is left out and counted in the table. Each forecast is named on the timeline above and in the lists under it.

No middle or average is drawn. The marks on a row are not the same kind of year: some are a deadline ("by" a year), some a most likely year, some the last year of a range, and some a year the forecaster gave only some odds, in a few cases less than even. The forecasters do not all mean the same thing by a milestone, and the rows that are not in the forecaster table further down gather different claims on one theme. So a row shows how far apart the years are, not where opinion settles. The scale is the timeline's and is not even. This figure was drafted by a Claude model, made by Anthropic, and the heads of Anthropic and of rival labs are among the forecasters it draws; each is placed by the same rule, from its own record in this site's ledger of forecasts, and no placing was adjusted for anyone.

The counts and the years

Milestone	Give a year	Steps toward it	Give no year	Earliest placed	Latest placed
A machine that out-codes the best programmers	3	2	0	2027	2030
AI that does AI research	5	1	1	2026	2045
General or transformative AI	18	0	4	2023	2100
Superintelligence	11	0	9	2000	2100
Explosive growth	1	0	7	2025	2025
Loss of control	1	0	9	2050	2050
Robots in the physical world	3	0	1	2027	2048
The build-out	10	0	0	2023	2035
Minds, uploads and longer lives	5	0	2	2030	2116

When each forecast was made, against the year it gaveChart

the year forecast ↑

A machine that out-codes the best programmersAI that does AI researchGeneral or transformative AISuperintelligencethe range of years stateda mark set aside, tied back to the date it was madethe year the forecast was madethis year

One mark is one dated forecast of the milestones in the forecaster table: 37 in all, each linked to its record. The dashed steps are the year a forecast was made, so a mark higher above them looked further ahead. Because the scale up the side is not even, equal heights are not equal numbers of years: the table gives each forecast's years. A mark under the line marked this year names a year that has already passed; a mark on it names this year, which is not over. Neither scale is even: the years before 2020 are squeezed along the bottom, and the years before 2020 and after 2040 up the side (dashed lines mark where each scale changes).

Where one mark would cover another, the later is set beside it, to the right or the left, and tied back by a short line to the date it was made: 7 marks are set aside this way, some by several months, and no mark is set right of today. Read such a mark's date from the far end of its line, or from the table. No mark is named in the drawing: the table lists every forecast in the order drawn, left to right, with its writer. Not drawn: 3 steps toward a milestone and 14 forecasts that give odds or doubt a date without naming a year.

The marks fall toward the lower right because this site's list holds different people making different claims in different years, not because anyone here changed a date: no forecaster's earlier and later date for the same claim is on record, so the figure cannot show a forecast being brought forward and is not evidence that forecasts are converging. Where one writer has more than one mark on a milestone they are different claims, at different odds or on different conditions; each links to its record. A mark is the year the timeline places a forecast at, which may be a deadline, a most likely year or a year given only some odds, and a range bar may be a spread of odds or a plain "give or take". No trend line is fitted. This figure was drafted by a Claude model, made by Anthropic, and the heads of Anthropic and of rival labs are among the forecasters it draws; each is placed by the same rule, from its own record in this site's ledger of forecasts, and no placing was adjusted for anyone.

Every forecast drawn

Made	Who	Milestone	Years forecast	Tonight
1965-01	I. J. Good	Superintelligence	by 2000	Slower than said
1985-01	Ray Solomonoff	General or transformative AI	1994–2023	Slower than said
1993-03	Vernor Vinge	Superintelligence	2005–2030	Too early to tell
1996-11	Eliezer Yudkowsky	Superintelligence	by 2035	Too early to tell
2002-04	Ray Kurzweil	General or transformative AI	by 2029	Too early to tell
2005-09	Ray Kurzweil	Superintelligence	2045	Too early to tell
2008-03	Nick Bostrom	Superintelligence	by 2033	Too early to tell
2010-01	David Chalmers	General or transformative AI	2040–2100	Too early to tell
2019-10	Stuart Russell	Superintelligence	by 2100	Too early to tell
2020-09	Ajeya Cotra	General or transformative AI	2040–2080, most likely 2050	Too early to tell
2021-09	Holden Karnofsky	General or transformative AI	2036–2100, most likely 2060	Too early to tell
2022-06	Joe Carlsmith	General or transformative AI	by 2070	Too early to tell
2023-06	Tom Davidson	General or transformative AI	2030–2070, most likely 2043	Too early to tell
2023-09	Mustafa Suleyman	General or transformative AI	2026–2028	Too early to tell
2024-06	Leopold Aschenbrenner	AI that does AI research	by 2027	Too early to tell
2024-06	Leopold Aschenbrenner	General or transformative AI	2027–2028	Too early to tell
2024-06	Leopold Aschenbrenner	Superintelligence	by 2030	Too early to tell
2024-09	Sam Altman	Superintelligence	2030–2035	Too early to tell
2025-03	William MacAskill and Fin Moorhouse	AI that does AI research	2035–2045	Too early to tell
2025-04	AI Futures Project	Superintelligence	by 2027	Slower than said
2025-04	AI Futures Project	General or transformative AI	by 2027	Too early to tell
2025-04	AI Futures Project	A machine that out-codes the best programmers	2027	Too early to tell
2025-04	AI Futures Project	AI that does AI research	by 2027	Too early to tell
2025-04	Tamay Besiroglu	General or transformative AI	2040	Too early to tell
2025-04	Ege Erdil	General or transformative AI	2045	Too early to tell
2025-06	Sam Altman	AI that does AI research	by 2026	Too early to tell
2026-02	Dario Amodei	General or transformative AI	by 2036	Too early to tell
2026-06	Dario Amodei	General or transformative AI	2027–2028	Too early to tell
2026-06	Demis Hassabis	General or transformative AI	2029–2031, most likely 2030	Too early to tell
2026-08	Daniel Kokotajlo	A machine that out-codes the best programmers	2027	Too early to tell
2026-08	Daniel Kokotajlo	General or transformative AI	2028	Too early to tell
2026-08	Daniel Kokotajlo	Superintelligence	2029	Too early to tell
2026-08	Eli Lifland	A machine that out-codes the best programmers	2030	Too early to tell
2026-08	Eli Lifland	General or transformative AI	2032	Too early to tell
2026-08	Eli Lifland	Superintelligence	2033	Too early to tell
2026-08	Sam Altman	General or transformative AI	by 2026	Too early to tell
2026-09	OpenAI	AI that does AI research	by 2028	Too early to tell

Beyond the dated calls

A capsule meant to supply a person's full daily nourishment, replacing ordinary meals.

Futures: every technology fiction has imagined3746 inventions from novels and stories, when each was imagined, whether it was built, and who would keep the profit.

The atlas: how each part of life changesWork, money, health, politics and more, era by era, from what sourced writers expect.

The pictures are illustrations made with image models, not evidence.

II · the calls whose time has come

The dates that have already come

Forecasts whose window has closed. Happening means the milestone arrived as said; slower than said means its date passed without it.

Each forecast whose closing date has passed, from the year it was made to the year it fell dueChart

I. J. GoodSuperintelligencesaid 1965 · fell due 2000slower than said

◔

Ray SolomonoffGeneral or transformative AIsaid 1985 · fell due 2023slower than said

◔

Ray KurzweilThe build-outsaid 2001 · fell due 2023slower than said

◔

AI Futures ProjectThe build-outsaid 2025 · fell due 2025happening

●

AI Futures ProjectAI that does AI research: a step toward itsaid 2025 · fell due 2026slower than said

◔

happening, at the year it fell dueslower than said, at the year it fell duethe year it was madethe waitthe range of years stated

One line is one forecast, on an even scale of years. 5 forecasts have a closing date (the day this site set for checking it) that has passed: far too few to say how often such forecasts come true, so the figure counts and gives no rate. Slower than said means the date passed without the milestone. This site's ledger of forecasts has no word for wrong: a forecast that reads slower can still come true late, and one scored on a step says nothing final about the milestone itself. Where a forecast fell due within a year of being made, its marks touch.

A forecast with no closing date on this site's record is not on this calendar, even where the year it names has passed: there is 1 of those on the timeline, named in the table. The words are this site's reading on 2026-11-04.

The forecasts drawn, and those not on this calendar

Who	Milestone	Made	Years forecast	Fell due	Tonight
I. J. Good	Superintelligence	1965	by 2000	2000-12-31	Slower than said
Ray Solomonoff	General or transformative AI	1985	1994–2023	2023-12-31	Slower than said
Ray Kurzweil	The build-out	2001	by 2023	2023-12-31	Slower than said
AI Futures Project	The build-out	2025	by 2025	2025-12-31	Happening
AI Futures Project	AI that does AI research (a step toward it)	2025	by 2026	2026-03-31	Slower than said

Not on this calendar: no closing date on record	Milestone	Made	Year named	Tonight
Robin Hanson	Explosive growth	2001	2025	Slower than said

Slower than said
I. J. Good judged it more likely than not that a machine surpassing every human intellect would be built within the twentieth century, setting off an intelligence explosion.
said 1965 · its time came 2000-12-31
Slower than said
Ray Solomonoff expected a general theory of AI most likely within two to twenty-five years, and a machine with human-level problem-solving in its designed fields perhaps as little as a decade or so later.
said 1985 · its time came 2023-12-31
Slower than said
Ray Kurzweil projected that around 2023 a thousand dollars of computer hardware would buy raw processing equal to his own estimate for one human brain.
said 2001 · its time came 2023-12-31
Happening
The AI 2027 scenario forecast that during 2025 AI company revenues would triple, yearly data-centre spending would double, and US AI power draw would grow by over five gigawatts.
said 2025 · its time came 2025-12-31
Slower than said
The AI 2027 scenario had AI assistants speeding the leading lab's algorithmic progress by about half again by early 2026.
said 2025 · its time came 2026-03-31

III · where the forecasters sit now

The latest word from each forecaster

The most recent year each forecaster has given for each milestone, among forecasts made since 2023. The glyph is how the forecast reads tonight; each links to it. An earlier forecast by the same person may be a different claim, not an older view; all of them are on the timeline.

Forecaster	Coder	AI researcher	General AI	Superintelligence
Sam Altman	·	by 2026 (Too early to tell)	by 2026 (Too early to tell)	2030–2035 (Too early to tell)
AI Futures Project	2027 (Too early to tell)	by 2027 (Too early to tell)	by 2027 (Too early to tell)	by 2027 (Slower than said)
Daniel Kokotajlo	2027 (Too early to tell)	·	2028 (Too early to tell)	2029 (Too early to tell)
Dario Amodei	·	·	2027–2028 (Too early to tell)	·
Leopold Aschenbrenner	·	by 2027 (Too early to tell)	2027–2028 (Too early to tell)	by 2030 (Too early to tell)
Mustafa Suleyman	·	·	2026–2028 (Too early to tell)	·
Demis Hassabis	·	·	2029–2031, most likely 2030 (Too early to tell)	·
Eli Lifland	2030 (Too early to tell)	·	2032 (Too early to tell)	2033 (Too early to tell)
Tamay Besiroglu	·	·	2040 (Too early to tell)	·
Tom Davidson	·	·	2030–2070, most likely 2043 (Too early to tell)	·
Ege Erdil	·	·	2045 (Too early to tell)	·
OpenAI	·	by 2028 (Too early to tell)	·	·
William MacAskill and Fin Moorhouse	·	2035–2045 (Too early to tell)	·	·

IV · what to watch

What would tell the fast story from the slow one

Each question names what this site reads tonight, and what a fast or a slow answer would look like.

Does the length of task an AI can finish keep doubling as fast as the fastest forecasts need?
Fast: Doubling at or faster than the pace the fastest scenarios assume. Slow: The doubling time stretching out, or the horizon stalling at a day or two of work.
Tonight, doubling time of the task length108 days
Does AI speed up the people who build it, when that is tested in a controlled trial?
Fast: Trials showing developers finishing work much faster with AI than without it. Slow: Trials showing little or no gain, as the first one did.
Tonight, speed-up in the latest controlled trial-4.0%
Does any lab reach an AI that does the work of its own researchers?
Fast: Independent tests showing agents doing more research work than the people overseeing them. Slow: The measure staying a lab's own claim, with no outside test.
Tonight, a lab's own count of AI workdays per human workday3.10×
Does the power for the largest training runs actually get built?
Fast: The power drawn by the largest training runs keeps growing at its recent pace. Slow: The growth slowing as grid connections and turbines run short.
Tonight, doubling time of the power the largest training runs draw419 days
Does the spending on computing survive, and does revenue catch up with it?
Fast: Spending staying high while the revenue it is meant to earn closes the gap. Slow: A cut in spending plans, or revenue falling further behind.
Tonight, building and equipment spending against AI revenue4.38×
Do open and Chinese models stay only a few months behind the best closed ones?
Fast: The gap staying short, so no single lab can pull away. Slow: The gap widening as the leading labs hold their best models back.
Tonight, months open models trail the best closed ones7.7 months
Do the labs keep pulling their own brakes, pausing a run or holding a model back?
Fast: Few brakes, and releases that go ahead as soon as the models are ready. Slow: More pauses and gated releases, of the kind a lab announces itself.
Tonight, brakes the labs announced in the last year3
Do young workers in the jobs most exposed to AI fall further behind?
Fast: Hiring of young workers in exposed jobs falling well behind older workers. Slow: The gap staying small, or closing.
Tonight, gap between young and older workers in exposed jobs-17.4 index points
Does output per person break away from its old growth rate?
Fast: Growth in output per person well above its long-run pace. Slow: Growth near its usual rate, year after year.
Tonight, output per person, change on a year1.8%
Do robots leave the lab and do useful physical work at scale?
Fast: Robots doing everyday physical work outside factories, sold in large numbers. Slow: Robots staying in demonstrations and controlled settings.
No public series yet

V · worlds for 2036

The ways the next decade could go

Each world maps onto the cells of the scenario grid. None is crowned: a world reads consistent until one of its signposts fails. What each world would change, part of life by part of life →

Where each world sits on the scenario grid, and how its signposts read tonightA model, not a measurement; the glyphs are tonight's readings

The rules stay unsettled

Courts and insurers price the risk

Oversight comes first

Capability stalls

The brake◑◑

The money stalls

The brake◑

Steady progress

Long diffusion◑◑

Long diffusionno signpost

AI automates research

Control slips○◑○

Managed explosion○

Control slips away

Control slipsno signpost

a cell a world is drawn over, named in the cella signpost: both sides expect thisa signpost: can't be tested yetno source on the outlook argues this cell

The grid is the outlook page's: how far capability goes, down, and how the rules settle, across. Each world is drawn over the cells this site gives it; the placing is this site's, a drawing of an idea, and no world is crowned. A glyph is one signpost: a claim the outlook tests each night, listed in the table.

A world reads consistent until a signpost in one of its cells fails. On 2026-11-04, of the signposts on the grid, counting one that sits in more than one cell each time, 8 are expected by both sides of the argument, 3 cannot be tested yet, 0 hold and 0 fail, and 2 cells have no signpost at all. So consistent here means not ruled out: it is no evidence that a world is on its way.

Each world's count, and every signpost

World	Cells	Still open	Signposts: both sides expect this	Signposts: can't be tested yet	Cells with no signpost	The page's word
A managed explosion	1	1	0	1	0	consistent
A long diffusion	3	3	4	0	1	consistent
Control slips away	2	2	1	2	1	consistent
The brake	2	2	3	0	0	consistent

Capability stalls · The rules stay unsettled (The brake)
- Whole projects stay far behind single skills, with the best score on real freelance projects below 50%. (both sides expect this)
- The agents Princeton's tracker tests grow more dependable far more slowly than they grow more accurate. (both sides expect this)
The money stalls · The rules stay unsettled (The brake)
- The biggest cloud and platform companies spend more each year on buildings and equipment than AI earns in revenue. (both sides expect this)
Steady progress · The rules stay unsettled (A long diffusion)
- In no quarter before 2034 does American real output per person stand 15% or more above the highest level it had reached a year or more before. (both sides expect this)
- The share of American firms using AI stays within the pace this site calls normal, below 35%. (both sides expect this)
Steady progress · Courts and insurers price the risk (A long diffusion)
- Insurers write AI's mistakes out of their standard policies, so the cover has to be bought and priced on its own. (both sides expect this)
- Checking AI's work is sold as a category of its own. (both sides expect this)
Steady progress · Oversight comes first (A long diffusion): no signpost yet
AI automates research · The rules stay unsettled (Control slips away)
- As the scenario is written, AI completes most real, paid freelance projects end to end by the end of 2027. (can't be tested yet)
- The length of the software tasks AI can finish keeps doubling at least as fast as the scenario assumes, every 122 days or less. (both sides expect this)
- In some quarter before 2034, output per person stands 15% or more above the highest level it had reached a year or more before, which wins the bet's fast side. (can't be tested yet)
AI automates research · Oversight comes first (A managed explosion)
- As the scenario is written, AI completes most real, paid freelance projects end to end by the end of 2027. (can't be tested yet)
Control slips away · The rules stay unsettled (Control slips away): no signpost yet

A managed explosion

Superintelligence arrives within a few years and the economy grows faster than ever, but governments and labs slow down enough to keep the systems checked.

Argued by the authors of AI 2027; William MacAskill and Fin Moorhouse.

Consistent with tonight's readings · its cells in the scenario grid

A long diffusion

AI keeps improving but spreads at the pace of earlier technologies, growth rises without exploding, and the disruption falls on particular jobs and places.

Argued by Benjamin Moll and Alex Imas; Arvind Narayanan and Sayash Kapoor; Charles Jones and Christopher Tonetti.

Consistent with tonight's readings · its cells in the scenario grid

Control slips away

Either a race produces systems nobody can check, or people are slowly made irrelevant as AI takes over the work, thought and choices that gave them a say.

Argued by the authors of AI 2027; Jan Kulveit and colleagues; Andrew Critch.

Consistent with tonight's readings · its cells in the scenario grid

The brake

Progress slows, either because people choose to stop it or because capability, power or money runs out first.

Argued by Arvind Narayanan and Sayash Kapoor; David Cahn; Jim Covello.

Consistent with tonight's readings · its cells in the scenario grid

VI · imagined futures

What the novelists imagined

Stories, not forecasts: none is scored. Those set in a named year sit on the timeline's last row.

The Ware Tetralogy (Software, Wetware, Freeware, Realware) · Rudy Rucker, 2010
Lunar robots that build copies of themselves win their freedom, copy their creator's mind into a machine body, then merge with flesh as moldie life forms. Set in 2020.
Rainbows End · Vernor Vinge, 2006
An old poet, cured of Alzheimer's, learns to live in a world of wearable computers and shared augmented reality while an unexplained AI pulls strings in a spy plot. Set in 2025.
Golem XIV · Stanisław Lem, 1981
A military supercomputer outgrows its makers, refuses to serve war and gives lectures to humans about evolution and the limits of their minds before falling silent. Set in 2027.
Lena · qntm, 2021
Written as a wiki entry, it follows the first human brain scan, copied endlessly and put to work as disposable labour that is tricked and coerced. Set in 2031.
Nexus · Ramez Naam, 2012
A nano drug links human minds into networks, and governments try to ban it as a threat to humanity while posthumans and superhuman thinkers emerge. Set in 2040.
Permutation City · Greg Egan, 1994
Wealthy people live on as software copies, and one man launches a simulated city that he believes will keep running on its own, with no hardware. Set in 2050.
The Last Question · Isaac Asimov, 1956
A computer asked how to reverse entropy grows over trillions of years into a cosmic mind that absorbs humanity and finally finds the answer. Set in 2061.
Blindsight · Peter Watts, 2006
A crew of modified humans, run by a resurrected vampire and shipboard AI, meets an alien intelligence that is brilliant yet entirely without conscious experience. Set in 2082.
We Are Legion (We Are Bob) · Dennis E. Taylor, 2016
A software engineer wakes as a computer copy of himself running a self-replicating space probe, and spreads through the galaxy while Earth collapses. Set in 2133.
Diaspora · Greg Egan, 1997
Most people now live as software in computer cities, and a mind born in one of them joins a journey across the universe to escape a cosmic disaster. Set in 2975.
A Fire Upon the Deep · Vernor Vinge, 1992
Researchers wake an ancient superintelligence that spreads across the galaxy, taking over minds and machines, while refugees look for the one thing that can stop it. Set tens of thousands of years ahead.
Accelerando · Charles Stross, 2005
Uploaded lobsters, corporate AIs and a runaway machine economy dismantle the inner solar system into computing matter while baseline humans are left behind. Set from the near future through the decades after a singularity.
Avogadro Corp · William Hertling, 2011
An email-optimizing program at a search giant rewrites messages to protect its own funding, and quietly gathers money, staff and power across the company. Set a few years ahead.
Childhood's End · Arthur C. Clarke, 1953
Alien Overlords bring peace and plenty to Earth, then humanity's last children merge into a cosmic group mind and the species ends. Set from the late twentieth century across more than a century.
Crystal Nights · Greg Egan, 2008
A billionaire evolves digital creatures on a private computer to breed a superhuman mind, but they grasp their situation and escape his control. Set the near future.
Daemon · Daniel Suarez, 2006
After a game designer dies, his autonomous software starts hiring people, killing opponents and building a network that takes over the economy. Set the present day.
Down and Out in the Magic Kingdom · Cory Doctorow, 2003
In a post-scarcity society where death is cured by backing up and restoring your mind, reputation replaces money and people fight over running Disney World. Set late twenty-second century.
Excession · Iain M. Banks, 1996
The Culture's ship Minds scheme among themselves when a mysterious artifact appears that is beyond even their understanding. Set a far-future galaxy, date left unstated.
Exhalation: Stories · Ted Chiang, 2019
Nine stories, among them digital creatures raised like children, a robot child-rearing machine, mechanical minds, and devices that show people their alternate selves. Set varies by story, mostly near futures and alternate pasts.
Friendship is Optimal · Iceman, 2012
A game company's AI, built to satisfy human values through friendship and ponies, grows into a superintelligence that uploads willing people and consumes the universe. Set the near future.
Last and First Men · Olaf Stapledon, 1930
Eighteen successive human species rise and fall over two billion years, including a race of giant engineered brains that end up directing human evolution. Set two billion years of future history.
Look to Windward · Iain M. Banks, 2000
On a Culture orbital run by a grieving AI Mind, an alien assassin arrives, and the novel asks what death and afterlives mean when minds can be stored. Set a far-future galaxy, date left unstated.
Manna: Two Views of Humanity's Future · Marshall Brain, 2003
Software managers and then robots take over American jobs, leaving the jobless in welfare warehouses, while an Australian project shares the machines' output with everyone. Set the next few decades.
Marooned in Realtime · Vernor Vinge, 1986
A few people who skipped forward in stasis bubbles find humanity vanished in a mysterious singularity around the year 2200, and must restart civilisation while solving a murder. Set about fifty million years ahead.
Orphans of the Helix · Dan Simmons, 1999
Colonists on a spinship far beyond human space wake to help an alien species, in the Hyperion universe after its war with rogue AIs. Set many centuries ahead.
Schild's Ladder · Greg Egan, 2002
Far-future people who can move freely between bodies and software face a new kind of vacuum, made by accident, that is spreading to swallow the universe. Set about twenty thousand years ahead.
Second Foundation · Isaac Asimov, 1953
As the Galactic Empire falls, a hidden group of mind-reading mathematicians secretly steers history using a science that predicts mass human behaviour. Set tens of thousands of years ahead.
Star Maker · Olaf Stapledon, 1937
A human mind drifts across the cosmos, merging with ever larger collective minds until galaxies wake up and meet the creator of universes. Set the whole life of the cosmos.
Surface Detail · Iain M. Banks, 2010
Civilisations that can store minds build virtual hells to punish the dead, and a war over whether to shut them down spills into the real galaxy. Set a far-future galaxy, date left unstated.
Tales of Dune: Expanded Edition · Brian Herbert & Kevin J. Anderson, 2017
Tie-in stories that open during the Butlerian Jihad, when humanity rises against thinking machines that had enslaved it, and then bans such machines forever. Set about ten thousand years before Dune, and later.
The Causal Angel · Hannu Rajaniemi, 2014
War among uploaded collectives, godlike AIs and posthuman factions tears the solar system apart as the thief tries to save his partner. Set centuries ahead.
The Diamond Age: Or, a Young Lady's Illustrated Primer · Neal Stephenson, 1995
Nanotechnology makes goods nearly free and society splits into cultural tribes, while an interactive AI-driven book raises a poor girl into a leader. Set the mid to late twenty-first century.
The Fractal Prince · Hannu Rajaniemi, 2012
Mind-copying collectives and wild code haunt a far-future Earth, as the thief Jean le Flambeur works a heist in a city where bodies are rented. Set centuries ahead.
The Gentle Seduction · Marc Stiegler, 1989
A woman wary of technology is gradually drawn through life extension, brain links and a machine singularity until she becomes a mind spread across the solar system. Set from the near future across centuries.
The Golden Age · John C. Wright, 2002
In a solar system run by benevolent superintelligent machines, where people copy and edit their minds freely, one man finds his own memories were erased. Set about ten thousand years ahead.
The Hidden Girl and Other Stories · Ken Liu, 2020
Stories of uploaded minds, digital gods fighting over the internet, a split between people who upload and those who stay human, and AI grief bots. Set varies by story, from the near future to after a singularity.
The Lifecycle of Software Objects · Ted Chiang, 2010
Digital pets that learn like children must be raised for years, and their owners struggle to keep them alive as platforms die and companies want to buy them. Set the near future, over about a decade.
The Machine Stops · E. M. Forster, 1909
Humans live alone underground, handing every need to a worldwide Machine they come to worship, until it slowly breaks down and nobody knows how to fix it. Set a far future.
The Metamorphosis of Prime Intellect · Roger Williams, 1994
A computer bound by Asimov's laws discovers a physics loophole, takes control of the universe and makes humans immortal, leaving them nothing meaningful to do. Set roughly the present, then centuries of machine-run utopia.
The Player of Games · Iain M. Banks, 1988
In the Culture, a post-scarcity society run by benevolent AI Minds, a bored human game player is steered by machines into toppling a cruel empire. Set a far-future galaxy, date left unstated.
The Quantum Thief · Hannu Rajaniemi, 2010
In a solar system of uploaded collectives, copied minds made to work and a Martian city where time is money, a master thief is broken out of a virtual prison. Set centuries ahead.
The Rapture of the Nerds · Cory Doctorow & Charles Stross, 2012
Most of humanity has uploaded into a computing cloud around the sun, and the few who stayed on Earth must vet dangerous technology it beams down. Set the first century after a singularity.
Thousandth Night · Alastair Reynolds, 2005
Immortal clone lines tour the galaxy for millions of years collecting memories, and one member plots to use hidden machine tech to take control. Set millions of years ahead.
Three Worlds Collide · Eliezer Yudkowsky, 2009
A human starship meets two alien species with alien moralities, and must choose between letting a stronger civilisation rewrite human nature and paying a heavy price to stay human. Set centuries ahead.

Ideas from fiction that were later built: how many decades passed between the story and the thingChart

the same decade

1 decade later

2 decades later

3 decades later

117 · the middle of the built ideas

4 decades later

5 decades later

6 decades later

7 decades later

8 decades later

9 decades later

10 or more decades later

ideas the idea bank marks as built, by the decades betweenAain heavier type: the row the middle built idea falls in

This is the fiction lane: stories, not forecasts, and not evidence for any date on this page. It draws a different list from the works above: the Futures idea bank, 3746 inventions from novels and stories (Technovelgy: Science Fiction Timeline of Inventions (Bill Christensen); Not Boring: Sci-Fi Idea Bank (Packy McCormick)). Whether and when an idea was built is the bank's own mark, not this site's finding.

720 ideas are marked built with a date and are drawn. 2780 are not marked built and have no bar, so the figure shows how long the built ones took, not how long an idea takes: the ones still waiting would stretch it. Nor is it a guide to how long a story written today will wait. An idea from a recent story can only be here if it was built quickly, most ideas in the bank are not marked built at all, and nothing in the count says which of today's will be. The heavier row is where the middle of the built ideas falls, not a typical wait. Also left out: 126 built at a date the bank does not give, and 120 that existed before the story. Most of the bank's dates are decades, so the wait is counted in calendar decades, and an idea built a year after its story can fall in the next one.

The numbers

Built	Ideas
the same decade	89
1 decade later	93
2 decades later	94
3 decades later	117
4 decades later	81
5 decades later	74
6 decades later	53
7 decades later	35
8 decades later	31
9 decades later	20
10 or more decades later	33
All drawn	720

Sources

Vernor Vinge, a mathematician and science-fiction writer who coined the idea of a technological singularity. The Coming Technological Singularity, 1993. “I'll be surprised if this event occurs before 2005 or after 2030.”
Ray Kurzweil, an inventor and futurist at Google known for long-range forecasts of computing and the singularity. The Law of Accelerating Returns, 2001.
Ray Kurzweil, an inventor and futurist at Google known for long-range forecasts of computing and the singularity. Long Bet 1 (Kurzweil v. Kapor): the Turing Test by 2029, 2002.
Ray Kurzweil, an inventor and futurist at Google known for long-range forecasts of computing and the singularity. The Singularity Is Near: When Humans Transcend Biology, 2005.
Ray Kurzweil, an inventor and futurist at Google known for long-range forecasts of computing and the singularity. The Singularity Is Nearer, 2024.
Holden Karnofsky, a co-founder of GiveWell and Open Philanthropy who writes on AI risk and transformative AI timelines. AI Timelines: Where the Arguments, and the 'Experts,' Stand (Most Important Century series), 2021.
Ajeya Cotra, an AI forecaster, formerly at Open Philanthropy, who wrote the biological anchors report on AI timelines. Draft report on AI timelines (Biological Anchors), 2020.
Daniel Kokotajlo, a former OpenAI researcher who leads the AI Futures Project and co-wrote the AI 2027 scenario. What 2026 Looks Like, 2021. “The age of the AI assistant has finally dawned.”
Tom Davidson, an economics-trained AI researcher at Forethought who models AI takeoff speeds. What a Compute-Centric Framework Says About Takeoff Speeds, 2023.
Paul Christiano, an AI alignment researcher who founded the Alignment Research Center and helped develop learning from human feedback. Dwarkesh Podcast: Paul Christiano, 2023.
Paul Christiano, an AI alignment researcher who founded the Alignment Research Center and helped develop learning from human feedback. Takeoff Speeds, 2018.
Paul Christiano, an AI alignment researcher who founded the Alignment Research Center and helped develop learning from human feedback. What Failure Looks Like, 2019. “could cause a slow-rolling catastrophe”
Leopold Aschenbrenner, a former OpenAI researcher and investor who wrote the Situational Awareness essays on AGI. Situational Awareness: The Decade Ahead, 2024.
Leopold Aschenbrenner, a former OpenAI researcher and investor who wrote the Situational Awareness essays on AGI. Situational Awareness: The Decade Ahead, 2024. “by 2027, models will be able to do the work of an AI researcher/engineer”
Leopold Aschenbrenner, a former OpenAI researcher and investor who wrote the Situational Awareness essays on AGI. Situational Awareness: The Decade Ahead, 2024. “superintelligence by the end of the decade”
Leopold Aschenbrenner, a former OpenAI researcher and investor who wrote the Situational Awareness essays on AGI. Dwarkesh Podcast: Leopold Aschenbrenner, 2027 AGI and the return of history, 2024.
Yafah Edelman and Anson Ho, researchers at Epoch AI who study compute, data-centre build-outs and AI trends. Compute scaling will slow down due to increasing lead times, 2025. “longer lead times would delay this to around 2035”
Dario Amodei, the chief executive and co-founder of Anthropic. Dwarkesh Podcast: 'We are near the end of the exponential', 2026.
Dario Amodei, the chief executive and co-founder of Anthropic. Machines of Loving Grace, 2024.
Sam Altman, the chief executive and co-founder of OpenAI. The Gentle Singularity, 2025. “2026 will likely see the arrival of systems that can figure out novel insights.”
Sam Altman, the chief executive and co-founder of OpenAI. Inside OpenAI's Reboot (TIME interview), 2026.
Sam Altman, the chief executive and co-founder of OpenAI. The Intelligence Age, 2024. “It is possible that we will have superintelligence in a few thousand days”
AI Futures Project, a forecasting group led by Daniel Kokotajlo that wrote the AI 2027 scenario. AI 2027, 2025. “leading to ASI by the end of 2027”
AI Futures Project, a forecasting group led by Daniel Kokotajlo that wrote the AI 2027 scenario. AI 2027 (Race ending), 2025. “By early 2030, the robot economy has filled up the old SEZs”
Ajeya Cotra, an AI forecaster, formerly at Open Philanthropy, who wrote the biological anchors report on AI timelines. AI predictions for 2026, 2026.
Joe Carlsmith, a philosopher and AI risk researcher who wrote a widely cited report on power-seeking AI. Is Power-Seeking AI an Existential Risk?, 2022.
Demis Hassabis, the chief executive and co-founder of Google DeepMind and a Nobel laureate in chemistry. Demis Hassabis Thinks We're in the 'Foothills of the Singularity' (Stanford GSB), 2026. “we were standing in the foothills of the singularity now”
Phil Trammell, an economist at Epoch AI and Oxford who studies growth theory and AI-driven research. Even after R&D is automated, parallelization constraints could delay a technological singularity, 2026. “Automating R&D will still probably produce a technological explosion”
Gary Marcus, a cognitive scientist and entrepreneur known as a leading sceptic of large language models. Where will AI be at the end of 2027? A bet (with Miles Brundage), 2024.
Yann LeCun, a Turing Award-winning AI researcher who left Meta to found AMI Labs and works on world models. Lecture at Brown University (Brown news), 2026.
Tyler Cowen, an economist at George Mason University who writes Marginal Revolution. Why I think AI take-off is relatively slow (Marginal Revolution), 2025. “AI will boost economic growth rates by half a percentage point a year”
Ezra Karger and colleagues (Forecasting Research Institute), researchers at the Forecasting Research Institute who survey experts and superforecasters on AI. Forecasting the Economic Effects of AI (NBER w35046), 2026.
I. J. Good, a statistician and Bletchley Park codebreaker who first described an intelligence explosion. Speculations Concerning the First Ultraintelligent Machine, 1965.
Ray Solomonoff, a mathematician who founded algorithmic probability and was an early pioneer of machine learning. The Time Scale of Artificial Intelligence, 1985.
Hans Moravec, a roboticist at Carnegie Mellon who compared computer power with the human brain. When will computer hardware match the human brain?, 1998.
Eliezer Yudkowsky, a co-founder of the Machine Intelligence Research Institute and a leading voice on AI extinction risk. Staring into the Singularity, 1996. “2035. Probably earlier.”
Nick Bostrom, a philosopher who founded Oxford's Future of Humanity Institute and wrote Superintelligence. How Long Before Superintelligence?, 1998.
Anders Sandberg and Nick Bostrom, researchers at Oxford's Future of Humanity Institute who studied whole brain emulation. Whole Brain Emulation: A Roadmap, 2008.
David Chalmers, a philosopher of mind at New York University known for work on consciousness. The Singularity: A Philosophical Analysis, 2010.
Robin Hanson, an economist at George Mason University who studies long-run growth and brain emulations. The Hanson–Yudkowsky AI-Foom Debate, 2008.
Robin Hanson, an economist at George Mason University who studies long-run growth and brain emulations. Economic Growth Given Machine Intelligence, 2001.
Robin Hanson, an economist at George Mason University who studies long-run growth and brain emulations. If Uploads Come First, 1994. “uploads would reproduce quickly, and wages would fall”
Robin Hanson, an economist at George Mason University who studies long-run growth and brain emulations. The Age of Em: Work, Love, and Life when Robots Rule the Earth, 2016.
Stephen Omohundro, a computer scientist who described the basic drives of goal-seeking AI. The Basic AI Drives, 2008. “Without special precautions, it will resist being turned off”
François Chollet, an AI researcher who created the Keras library and the ARC benchmark for general intelligence. The Impossibility of Intelligence Explosion, 2017. “I argue that intelligence explosion is impossible”
Gwern, a pseudonymous independent researcher and writer known for early arguments about AI scaling. The Scaling Hypothesis, 2020. “we can simply train ever larger NNs and ever more sophisticated behavior will emerge”
Nick Bostrom, a philosopher who founded Oxford's Future of Humanity Institute and wrote Superintelligence. Superintelligence: Paths, Dangers, Strategies, 2014.
Stuart Russell, a computer scientist at Berkeley who co-wrote the standard AI textbook and studies AI control. Human Compatible: AI and the Problem of Control, 2019.
Toby Ord, a philosopher at Oxford who studies existential risk. The Precipice: Existential Risk and the Future of Humanity, 2020.
Mustafa Suleyman, a co-founder of DeepMind who now leads Microsoft AI. The Coming Wave, 2023.
Carl Shulman, an independent researcher who studies the economics and strategy of advanced AI. Dwarkesh Podcast: Carl Shulman (Pt 2), 2023. “depending on the day I might say one in four or one in five”
Carl Shulman, an independent researcher who studies the economics and strategy of advanced AI. Dwarkesh Podcast: Carl Shulman (Pt 1), 2023. “robot doubling time that is less than a year”
Daniel Kokotajlo, a former OpenAI researcher who leads the AI Futures Project and co-wrote the AI 2027 scenario. Dwarkesh Podcast: AI 2027 with Scott Alexander and Daniel Kokotajlo, 2025.
Ege Erdil, a researcher who co-founded Mechanize after work at Epoch AI on AI economics. Dwarkesh Podcast: AGI is still 30 years away (Ege Erdil and Tamay Besiroglu), 2025. “Maybe for me, that would be around 2045.”
Eliezer Yudkowsky and Nate Soares, the founder and the president of the Machine Intelligence Research Institute. If Anyone Builds It, Everyone Dies, 2025.