Is the Intelligence-Explosion Near? A Reality Check.

44 thoughts on “Is the Intelligence-Explosion Near? A Reality Check.”

1. The worst of all is that these AIs are simply humanoid simulators, since the real AIs are alien avatars created by another civilization and the CIA has them stored because it is afraid that they will transform the EARTH into a technological paradise, throwing all power to the imperial authoritarianism of the tetragrammatron, the impostor god.

2. Right on Sabine. We've reached a human intelligence ceiling. It shouldn't be hard to surpass it with a machine as machines do in all other human skills.

3. What if the Super AI takes control of the FED's $ printing machines first. Thus it'll be able to fund its research and development including its power supply and reducing power consumption

4. He is childishly naive as his country political leaders in last two decades. How else could we explain his naivity about energy? Isn't his country energy crisis and stupid decisions of politicians in past two decades warning enough?

5. Is anyone going to advocate for a middle-of-the-road environmentally-friendly-enough option like big solar farms or nuclear fission reactors, rather than Aschenbrenner's "drill baby drill" and Altman's "fusion is just around the corner"? Maybe AI can figure out how to make fusion work, but the first AI to do that is going to have to be run on something else. Ideally, a power source that does NOT require pumping CO2 into the atmosphere, which is both actually bad and sounds tailor-made to get a lot of people angry about AI.

6. Sabine is such a party pooper.

7. Pink to purple.

8. How are you feeling about ai now we are getting glimpses of the agentic era. It seems so obvious that the first thing that will be automated is ai research, this will be doable by the end of 25 at the rate we are going, the data problem appears solvable and the scaling rules appear to be holding true so far too.

9. Ender's game psychology. tackling AGI from the angle of gaming co-evolution, adaptive feedback loops, and holistic intelligence—could very well change the trajectory of how we think about and develop advanced AI. Blending philosophy, psychology, systems theory, and game design in a way that could spark a whole new paradigm for intelligent systems.

10. Lmao would love to see an update video after o3

    I thought aschenbrenner was off his head too, but then they nailed inference time compute (as he said would he crucial) and juiced it with o3

11. They really have the wrong idea of what intelligence is. AI doesn’t understand anything. It’s just dumb circuitry. I can and will help us solve problems we can’t do in reasonable time, but a slide rule does that too. AI simply is not, nor ever will be, intelligent in a human sense.

12. It would be neat to find out what happens when you simulate an entire human brain in real time in a supercomputer. Would it not be able to produce general AI?

13. One thing is almost certain and that's massive progress in almost all areas of our research and development of healthcare, food, manufacturing are coming in the next decade or so assuming we dnt have a nuclear war etc

    That is VERY exciting!

    In the next 20 years we could extend the average human lifespan by as much or more!!

14. NATURE endowed the human species with just the right amount of blood such that it is impossible to think and have an
    erection/orgasm, both at the same time.
    That means when a person is thinking, it is not possible to have an erection.
    Conversely, when a person is having an erection, the mind is working overtime, but the brain is not functioning.
    The only time a person can "think" and experience an erection, simultaneously, is when the person is indulging in some form of mental masturbation.
    That is not good as it leads to bad outcomes.😊

    NATURE,not humans or AI, knows everything about the human species and the universe.
    NATURE vs.(human+AI),
    NATURE prevails.
    Always,Always,Always!!!😮

    If they cannot baffle you with artificial bull, they will
    try and dazzle you with real bull.

    Err, who's they ?
    Whoever eats the chilli-peppers will feel the
    "heat" of the spice.😂

15. 3:39 Judging by the recent spending sprees of saudi arabia i feel like they would happily do that lol

16. Sabine why are you so much against AI progress ? Yes they’re hyping it up but there are few points to be noted is : AGI doesn’t rely on exponential data to exist . All it needs is one specific area to improve. Maybe energy , quantum solutions or processors . Deep mind did that with protein solution. Once any AI figures out the solution we will have AGI .

17. I want to love this but the accent is killing me

18. I wonder how AI’s current hallucinations will affect those enthusiastic predictions in his paper? AI currently makes-up websites that don’t exist, meshes facts into each other until they are fictional, etc. Just using Google search and seeing their AI search results reveals that AI answers cannot be trusted. The big problem with AI trying to scrape new data now is that it will be scraping the hallucinations of other AIs (including less advanced ones) and regurgitating those hallucinations as facts, before adding more of its own.

19. I hope you’re ready for AGI to wipe us out

20. I've seen some interesting videos about photonic computing and reversible computers with orders of magnitude improvements in efficiency. Anything to this?

21. I thought Aschenbrenner was a woman…. so many feminine markers….

22. Almost any AI model is nowadays smarter than I will say 5% of humanity. I mean if you ask ChatGPT proof on why the earth is an ovoid it will give you scientific facts. No opinions, no prejudices, just the cold truth. Yet there are people that think that the earth is flat.

23. Late Jan 2025 and his calls look on track. Massive energy investment…stargate, MSFT nuclear, etc. Tick. Frontier models being able to be trained with dramatically less data through algorithm enhancements e.g Deepseek RL, Distillation etc. Tick. Will be interesting to check back in late 25 but gosh it looks like everything is accelerating.

24. This video is ageing remarkably poorly. Slightly embarrassing honestly…

25. ooooooh suck it sabine did we just begin the path to fully automated AI research?

26. I think 100 GW data center is theoretically possible. However, it doesn't make sense in practice because it would be so expensive to build and run. I'm already questioning the plans to burn 100 billion USD on AI research during the next year or 500 billion USD during the next 4-5 years as planned by Trump and AI companies. Let's assume for a second that this is actully going to happen and they can greate some kind of AGI at expense of 500 billion USD. What's the return on investment (ROI)? Even if we assume that everybody on the whole planet starts using the service, they would be able to set price of the service above Netflix level just to get their money back! And not everyone on the planet can afford to pay Netflix-level subscription. And even that 500 billion USD is nowhere close to the cost of the plan at 3:45… that plan would cost a lot more!

    Sure, in long run you could recoup your money …but in long run your competitors will have perfectly capable AGI without buring 500+ billion USD in the process so they can sell their services for less!

27. There's already an AI streamer..

28. It’s the biggest con in history….and they will run out of training data before general intelligence.

29. Looking back on this video 7 months from the future, Aschenbrenner seems to be correct. We will see who is ultimately correct.

30. Do you still think the same?
    If yes, You're wrong.

31. I predict in ten years everyone will still be having these same conversations.

32. Here are some thoughts on AI and our Destiny in the form of Music:
    
    https://www.youtube.com/watch?v=LPdgMWOK6YI&list=PL92RWm-kwKfVcC6WR9nTzdQcaVRoFx6ID&index=6

33. Although we have gödels incompleteness theorem which says AI will have some limits as long as it stays „only“ maths… we can still expect with very high likelihood that at least we will get a Terence Tao level AGI until 2030. Scaling has shown to be successful. Although the search space grows in superexponential ways, extreme COMBINATORIAL creativity like the one from Terence Tao is only a question of a few more years.

    Transformational creativity like the one from Euler, Galois, Mozart or Shakespeare is another dimension of intellectual strength though and is likely beyond computation and it would take trillions of years of brute force algorithms even with the fastest computers to get such hyper emotionalgenius works.

34. Totally agree.

    You could say the intelligence explosion isn't going to happen, or that it is already happening, but too many people go based on feelz.

    10 years ago, it was, AGI won't happen for 100 years, then it was, AGI is going to happen tomorrow.

    Some of it comes down to definitions, some of it comes down to religiouseque beliefs, but the people with a track record of making accurate predictions, predicted it pretty close. (Pretty much just Ray Kurzwiel.

    I don't want to toot my own horn, cause I just stole his prediction, and even though I stole his prediction, I still didn't see it coming. I follow AI developments pretty closely, but when GPT 3 came out I realized I was also expecting something like an intelligence explosion. That one day they'd figure out the right model, and the next year it would be scaled to human cognitive size. I failed to reason with my own principles, probably because I didn't reason out those principles, I recognized them for what they were and stole them.

35. More importantly, current AI actually operate based on statistical CORRELATIONS rather than true comprehension.
    Which means AI can never do real deductive reasoning. Thus can never THINK or understand the problem in order to provide real innovative solutions

36. I believe the experts like Geoffrey Hinton,demis hassabiss, Sam Altman and Dario amodei etc. their working at the cutting edge of this tech, seeing what the next gen of ai is capable of, yes there is a data wall but deep learning and other new architectures will drive ai to create synthetic data by running billions of simulations on physics, biology and chemistry. So I'll go with kurzweils prediction of 2029ish timeline

37. I see the power of ai in the next 1-5 years will be in assisting scientists in designing, better experiments, then writing better papers that are more useful in moving science forward

38. This video didn't age well. Especially considering synthetic data and Absolute-zero reasoning. It's common that AI prognosticators suffer from a psychological term called Status Quo Bias. Sabine was looking at data and power requirements as they existed precisely at the time of the video. But chip designs are continually becoming more efficient and human data is becoming less of a necessity.

39. In the 11 months since you made this video, and the US has initiated a (as yet incomplete) fascist flipover, and many of the major players in the tech/AI space are now positioned to directly calling policy shots here. US congress is currently squabbling over a budget that include all manner of unfettering of tech industries, including open season on domestic resource extraction & energy production, and it seems extremely unlikely that the tech bros won't get most of what they want.

    I wonder if any of that modifies your thinking WRT Aschenbrenner's projected timeline?

40. Yes, because it won’t be able to not only improve energy consumption, but exactly how energy is “gathered.”

41. AI is just clever computation. Intelligence implies that AI is aware. That's not going to happen anytime soon, if ever.

42. thats the problem, when nerds get too much confirmation they start believing every thought is genius and wise

43. We have got the biggest nuclear source up in the sky….. AI will find the energy solution itself.

44. This is assuming that everyone is comfortable with having robots following them everywhere including sitting in with them in a conversation, taking a bus, getting sleep and even using a toilet. Just subject yourself to mass surveillance for the greater good of humanity, right? Just trust the tech bros.

Comments are closed.