I don’t think they passed it in a full sense. Before LLM, there was a 5 minute Turing test, and some chatbots were passing it. I think 5 minutes is not enough. I bet that if you give me 10 hours, any currently existing LLM and human, we will communicate only via text, I will be able to figure out who is who (if both will try hard to persuade in their humanity). I don’t think LLM can come up yet with a consistent non-contradicting life story. It would be an interesting experiment :)
Valentin2026
Karma: 260
Hyperbolic trend with upcoming singularity fits METR capabilities estimates.
Would you mean similarity on the outer level (e.g. Turing test) or at inner (e.g. neural network structure should resemble brain structure?
If the first—would it mean that when AI passes Turing test it would be sentient?
If the second—what are the criteria for similarity? Full brain emulation or something less complicated?
How AI researchers define AI sentience? Participate in the poll
Are you working with SOTA model? Here, mathematicians report a quite different story https://www.scientificamerican.com/article/inside-the-secret-meeting-where-mathematicians-struggled-to-outsmart-ai/
I guess “good at” was improper wording. I did not mean that they do not produce nonsense. I meant that sometimes they can produce a correct solution. It is like the person may be not fit for running 100 meters in 10 seconds every day, but even if they do it in 5% of cases this is already impressive, and shows that it is possible in principle. And I guess “Ph.D. level” sounded like they can write a Ph.D. thesis from scratch. I just meant that there are short well-formulated problems that would require Ph.D. student a few hours, if not few days, which current LLM can solve in non negligible number of cases.
Can you expand your argument why LLM will not reach AGI? Like, what exactly is the fundamental obstacle they will never pass? So far they successfully doing longer and longer (for humans ) tasks https://benjamintodd.substack.com/p/the-most-important-graph-in-ai-right
I neither can see why in a few generations LLM won’t be able to run a company, as you suggested. Moreover, I don’t see why it is necessary to get to AGI. LLM are already good at solving complicated, Ph.D. level mathematical problems, which improves. Essentially, we just need an LLM version of AI researcher. To create ASI you don’t need a billion of Sam Altmans, you need a billion of Ilya Sutskevers. Is there any reason to assume LLM will never be able to become an excellent AI researcher?
Mechanistic Interpretability Via Learning Differential Equations: AI Safety Camp Project Intermediate Report.
I agree, they have a really bad life, but Eliezer seems to talk here about those who work 60 hours/week to ensure their kids will go to a good school. Slightly different problem.
And on homeless people, there are different cases. In some UBI indeed will help. But, unfortunately, in many cases the person has mental health problems or addiction, and simply giving them money may not help.
I feel that one of the key elements of the problem is misplaced anxiety. If the ancient farmer stops working hard he will not not get enough food. So all his family will be dead. In modern Western society, the risk of being dead from not working is nearly zero. (You are way more likely to die from exhausting yourself and working too hard). When someone works too hard, usually it is not fear of dying too earlier, or that kids will die. It is a fear of failure, being the underdog, not doing what you are supposed to, and plenty of other constructs that ancient people simply did not reach—first they needed to survive. In this sense, we are way better than even one hundred years ago.
Can UBI eliminate this fear? Maybe partially it can help, but people will still likely work hard to preserve their future and the future of their children. Maybe making psychotherapy (t address the fear itself) more available for those with low income is a better solution. I understand that it would require training way more specialists than we have now. However, some people report a benefit from talking with GPT as a therapist https://x.com/Kat__Woods/status/1644021980948201473 , maybe it can help.
Mentorship in AGI Safety: Applications for mentorship are open!
What is the application deadline? I did not find it in the post. Thank you!
Yes, absolutely! We will open the application for mentee later
Mentorship in AGI Safety (MAGIS) call for mentors
So far nothing, was distracted by other stuff in my life. Yes, let’s chat! frombranestobrains@gmail.com
After the rest of the USA is destroyed the very unstable situation (especially taking into account how many people have guns) is quite likely. In my opinion countries (and remote parts of countries) that will not be under attack at all are much better
What to do if a nuclear weapon is used in Ukraine?
[Question] Would “Manhattan Project” style be beneficial or deleterious for AI Alignment?
Thank you for your research! First of all, I don’t expect the non-human parameter to give a clear power-law, since we need to add humans as well. Of course, close to singularity the impact of humans will be very small, but maybe we are not that close yet. Now for the details:
Compute:
1. Yes, Moore’s law was a quite steady exponential for quite a while, but we indeed should multiply it.
2. The graph shows just a five years period, and not the number of chips produced, but revenue. The five years period is too small for any conclusions, and I am not sure that fluctuations in revenue are not driven mainly by market price rather than by produced amount.
Data storage:
Yes, I saw that one before, seems more like they just draw a nice picture rather than real data.
General remarks:
I agree with the point that AGI appearance can be sufficiently random. I can see two mechanisms that potentially may make it less random. First, we may need a lot of computational resources, data storage etc. to create it, and as a lab or company reaches the threshold, it happens easily with already existing algorithms. Second, we may need a lot of digitalized data to train AGI, so the transition again happens only as we have that much data.
Lastly, notice that cthe reation of AGI is not a singularity in a mathematical sense yet. It will certainly accelerate our progress, but not to infinity, so if the data will predict for example singularity in 2030, it will likely mean AGI earlier than that.
How trustworthy would this prediction be? Depends on the amount of data and noise. If we have just 10-20 datapoints scattered all around the graph, so you can connect the dots in any way you like—not really. If, instead, we are lucky and the control parameter happened to be something easily measurable (something such that you can get just-in-time statistics, like the number of papers on arXiv right now, so we can get really a lot of data points) and the parameter continues to change as theory predicts—it would be a quite strong argument for the timeline.
It is not very likely that the control parameter will be that easily measurable and will obey power-law that good. I think it is a very high risk—very high gain project (very high gain, because if the prediction will be very clear it will be possible to persuade more people that the problem is important).
6 votes
Overall karma indicates overall quality.
0 votes
Agreement karma indicates agreement, separate from overall quality.
Thank you very much for catching the mistake! I checked, you are completely right.