He said AI will be incredibly sophisticated in 20 years and appears to be accelerating, which makes exponential forecasts hard.
"I think like AI is going to be incredibly sophisticated in 20 years"
When will AI surpass humans?
Q649 answers2015-2026view changed
Short answer
He predicted AI software would be "Stockfish level" good at some point next year, so no human can compete in writing software.
How his answer changed: The 'exceeds all humans combined' date has drifted later: ~2029-2030 (Feb 2025) vs. '5 or 6 years' (Feb 2026) and 'around 5 years' (Jul 2026), i.e. ~2031.
All 49 answers, oldest to newest
2015 · 1 answer
2017 · 2 answers
If it reaches a threshold where it is as smart as the smartest, most inventive human, it really could be only a matter of days before it is smarter than the sum of humanity.
"it really could be only a matter of days before it's smarter than the sum of humanity"
Imagine being very confident we will be visited by super-intelligent aliens in 10 years or 20 years at the most. Digital superintelligence will be like an alien.
"digital super intelligence will be like an alien"
2019 · 2 answers
He said people underestimate the capability of AI. They think it is a smart human, but it will be much more than that — much smarter than the smartest human, more like a chimpanzee trying to understand humans.
"it'll be much smarter than the smartest human"
He said computers are already much smarter than people on many dimensions; we keep moving the goalposts. There is a smaller and smaller corner of intellectual pursuits where humans are better, and soon we will be far surpassed in every single way — guaranteed, or civilizational end.
"soon we will be far far surpassed in every single way"
2020 · 2 answers
He says the people most wrong about AI are very smart people who cannot imagine a computer being way smarter than them.
"The people I see being the most wrong about AI are the ones who are very smart because they can't imagine that a computer could be way smarter than them."
AI will be able to do everything better than humans over time, everything.
"AI will be able to do everything better than human over time"
2022 · 1 answer
Even if Tesla stopped doing AI, he thinks we are still on a track to artificial general intelligence, meaning intelligence smarter than the smartest human.
"intelligence smarter than the smartest human"
2023 · 14 answers
He said but, all things considered with regard to AGI existential angst, I would prefer to be alive now to witness AGI than be alive in the past and not
"But, all things considered with regard to AGI existential angst, I would prefer to be alive now to witness AGI than be alive in the past and not"
It is very difficult to predict — the singularity, like a black hole. Be cautious with AI.
"it's called the singularity"
Humans' advantage is being smarter than other creatures; now there will be something way smarter than the smartest human.
"way smarter than smartest human"
He said it’s not AGI until it can solve at least one fundamental physics problem
"It’s not AGI until it can solve at least one fundamental physics problem"
In July 2023 he said his AGI prediction roughly matched Ray Kurzweil's 2029.
"my prediction for AGI would, you know, roughly match that which I think Ray Kurzweil at one point said 2029"
He predicts that once AGI is solved, people will look back and wonder why it seemed so hard.
"once AGI is solved, we'll look back on it and say actually why do we think it was so hard?"
He says AGI will not arrive overnight but gradually.
"It's not going to be like one day, it's, you know, not AGI next. It is, it's going to be gradual."
He says biological compute could become less than 1% of all compute if trends continue.
"at some point, if this trend continues, biological compute will be less than 1% of all computer"
He said i think we may have figured out some aspects of AGI. The car has a mind. Not an enormous mind, but a mind nonetheless.
"I think we may have figured out some aspects of AGI. The car has a mind. Not an enormous mind, but a mind nonetheless."
He describes AI progress as a series of stacked log curves.
"Series of stacked log curves. Each log curve gets higher than the last one."
Less than three years from AI that could write a novel as good as JK Rowling or discover new physics.
"we are less than three years from that point"
For the first time we will have something smarter than the smartest human. There will come a point where no job is needed — you can have a job for personal satisfaction, but AI will be able to do everything.
"we will have the first time something that is smarter than the smartest human"
He predicts AI will come to understand fundamental physics and invent new technologies.
"I think what's coming is AI understanding fundamental physics and AI inventing new technologies."
In March 2024 he said more than 99% of all compute would be digital rather than biological, if not already then within a year or two.
"more than 99% of all compute will be digital instead of biological. Right. And if we're not there already, we will be within a year or two."
2024 · 2 answers
AI smarter than any one human around the end of next year (2025); total sentient AI compute exceeding all humans in about 5 years.
"smarter than any any one human probably to around the end of next year"
Asked Do we reach the critical Fermi on Mars before or after we reach AGI super intelligence ..., he said AGI will come first, but there are many levels of AGI. As @waitbutwhy’s charts suggest, we are at the very beginning of intelligence. Humans are the biological bootloader for digital superintelligence and we may serve...
"AGI will come first, but there are many levels of AGI. As @waitbutwhy’s charts suggest, we are at the very beginning of intelligence. Humans are the biological bootloader for digital superintelligence and we may serve as a backup plan for intelligence, given that humans are far more resilient than circuit boards on Earth."
2025 · 18 answers
Smarter than the smartest human maybe next year or in a couple of years; smarter than all humans combined around 2029 or 2030.
"smarter than all humans combined which frankly is around 2029 or 2030"
In 10 years AI could do anything better than a human cognitively; based on current improvement it will be smarter than the smartest human.
"in 10 years based on the current rate of improvement AI will be smarter than the smartest human"
He said they are close to AGI or digital superintelligence and seeing an explosion in digital superintelligence.
"we are close to what you might call AGI"
He is not sure; the first crossing may be less profound than people expect. AI has not yet invented useful new technologies or discovered new physics, but he thinks that will happen.
"At least the first will be less profound"
He said AI is already much smarter than most humans, but not yet smarter than any human at anything; probably smarter than any single human at anything in under 2 years and smarter than all humans combined in about 5 years, and it must be maximally truth-seeking.
"AI is already much smarter than most humans, just not yet smarter than any human at anything and much less smart than a group of smart humans using computers (aka “a company”). But it will be. Probably smarter than any single human at anything in <2 years and smarter than all humans combined in ~5 years. Make sure AI is maximally truth-seeking, even if there are some bumps along the way."
He said we are seeing extremely rapid AI take-off, but movement of people and ideas means the leading AI companies still appear in similar positions for now.
"If you zoom out a bit, we are actually seeing extremely rapid take-off of AI, but there is enough movement of people/ideas that the leading AI companies – for now – appear to be in similar positions. As a side note unrelated to your post, and at the risk of starting a war, I think the em dash is too long esthetically and is particularly ugly when attached directly to letters. We should use the en dash instead, at least on this platform – where italics for emphasis – do not appear in primary text."
He said intelligence still scales logarithmically with compute, and natively multimodal systems should just be called models, not LLMs.
"False, intelligence still scales logarithmically with compute. And it doesn’t make sense to call them LLMs when they’re natively multimodal. Just models – it’s cleaner."
He said Grok 5 has a shot at being true AGI, which he had not felt about anything before.
"Wait until you see Grok 5. I think it has a shot at being true AGI. Haven’t felt that about anything before."
Asked about Grok 5, he said it has a non-trivial chance of achieving AGI and that xAI is close to having all the pieces in place.
"It has a non-trivial chance of achieving AGI. xAI is close to having all the pieces in place."
Human intelligence is plateauing with population decline. He guesses AI may be smarter than any single human at anything as soon as next year (2026), and smarter than all humans combined around 2030.
"within five like say 2030 probably AI is smarter than the sum of all humans"
He said he now assigns about a 10% and rising probability that Grok 5 reaches AGI, versus 0% for all prior Grok releases.
"Just a chance. But I thought the probability was 0% for all prior Grok releases and now my neural net predicts ~10% probability and rising for Grok 5."
He said he now thinks xAI has a chance of reaching AGI with Grok 5, which he had never thought before.
"I now think @xAI has a chance of reaching AGI with @Grok 5. Never thought that before."
He said early Grok 4.20 experiments show the ability to generalize from Python to other languages, and that software is just logic at scale.
"Our early experiments with Grok 4.20 show the ability to generalize from Python to other languages. Software is just logic at scale."
He defined the nearer target as anything a human with a computer can do, probably three to five years away, and said Grok 5 will be better at AI engineering than Andrej Karpathy.
"Capable of doing anything a human with a computer can do, but not smarter than all humans and computers combined. That’s probably three to five years away. Grok 5 will be better at AI engineering (called “research” by the pompous) than @karpathy, for example."
He said Grok 5 will be AGI or something indistinguishable from AGI.
"Grok 5 will be AGI or something indistinguishable from AGI"
He said his estimate of the probability of Grok 5 achieving AGI is now 10% and rising.
"My estimate of the probability of Grok 5 achieving AGI is now at 10% and rising"
Grok 5, due in Q1, is the first time he assigns a non-zero (about 10%) chance of AGI. He says it should be a 6-trillion-parameter multimodal model and the smartest AI by a significant margin, though he might be wrong.
"we have a non-zero chance of achieving artificial general intelligence"
He said car AI is photons in, controls out, just like humans, and that this is the path to AGI.
"The car AI is photons in, controls out. Just like humans. This is the path to AGI."
2026 · 7 answers
He doesn't know 10 years out. AI might be smarter than any human by the end of this year, and no later than next year; by 2030 or 2031 (about 5 years) smarter than all of humanity collectively.
"AI that is smarter than any human by the end of this year"
He said we are already in the singularity, expects AGI next year in 2026, and is confident that by 2030 AI will exceed the intelligence of all humans combined. After AGI, thousand- to 10,000-fold algorithmic gains would come quickly.
"I'm confident by 2030 um AI will exceed the intelligence of all humans combined."
Maybe in five or six years AI exceeds the sum of all human intelligence.
"maybe in five or six years, AI will exceed the sum of all human intelligence"
He said by the end of this year things may move to where you don't even bother doing coding.
"I think actually things will move maybe even by the end of this year to where you don't even bother doing coding"
AI may exceed the sum of human intelligence in around 5 years, and be far greater within 10.
"I think AI may exceed the sum of human intelligence in about in around 5 years"
He predicted AI software would be "Stockfish level" good at some point next year, so no human can compete in writing software.
"at some point next year is my prediction, software will be so good, AI software will be so good that it will be Stockfish level good"
Replying to a report that Anthropic's Jack Clark now thinks full recursive self-improvement is more likely in 2027 than 2028, he said that was obvious.
"Obv"