Every time someone says 'current AI is not enough, we need new architectures', it turns out that current AI was enough and all we really needed was more scaling.
This lasted for 2005–2014. By 2015, the advantage humans added was negative.
I imagine this will be the case here as well, but I highly doubt it'll take that long.
The genius here is Claude.
It doesn't make sense to compare the quantity of work. Humans are significantly more efficient than AI, so little work can go a lot further.
I’m sure for every incredible insight which Levant credits to Claude, there’s 100x+ more bad insights that Levant had to filter out.
This is certainly my experience of using Claude every day, in a much less cutting edge domain.
And Claude itself cannot tell the difference between the two. The best it can do is use some sort of external constraint (e.g. verification) to filter them.
And then they are simply boldly wrong at times. About mathematical facts. That’s why we need formalisation when working with them. They are still stupid and can’t be trusted. But theyvare also smart and can come with amazing insight. It’s a tool, but definitely one which cuts both ways.
Do we even know what Alpöge had at his disposal? I think most people underestimate what working with an agent swarm even with the current public technology would be like. We can't assume he didn't have something more advanced either.
I don't feel confident my usage of AI on a personal level reflects any of what has been done here.
A non math person may not be able to comprehend the output, but even professional mathematicians have trouble reading through the huge “alien” AI generated LEAN formalised proofs.
On that topic, I'm surprised the usual anti-ai trolls aren't posting low-effort drive-by "ai slop", "stochastic parrot", "autocomplete on steroids" comments to this very discussion.
Edit: I stand corrected: RunSet, ToucanLoucan, luka598, and others have checked in.
We know the architectures of the open weight models as they include architecture diagrams, support in projects like llama.cpp, etc.
Companies like Anthropic and OpenAI don't release information about their models (aside from general things like context window size, whether they support reasoning, etc.). A lot of what we know is speculation. From press releases Anthropic have stated that they are getting Claude to assist with designing new iterations of the models, but no specific details.
The open weight models and other arXiv papers/research are iterating on and experimenting with the base/core transformer architecture:
- RoPE/NoPE/etc. for context window scaling;
- dense vs mixture of experts models;
- Mamba-based State Space Models (falcon models, etc.);
- reasoning, combining transformers with a residual stream, Attention Residuals (Kimi), etc.
- activation functions: ReLU; SwiGLU; etc.
- etc.
it does seem like (IMO) the biggest leaps have basically been perf & scaling + RL on high quality data + harnesses.
He just did the math and realized you just needed more and more compute to get more intelligence. People keep doubting it but it's showing no sign of being incorrect.
ChatGPT was the turning point of realizing we needed just more compute
Inference can be increased vertically (a single model thinking for longer), or horizontally (NS was solved by a 10,000-agent swarm).
And pretraining has two inputs that can be scaled independently - training data and model size. (Things like the Chinchilla 20:1 law are meant to establish the optimal way to train with a limited budget - you could just keep making either term bigger and bigger on its own if you had infinite money.)
I think creating and curating high-quality datasets is also part of scaling. Datasets like FineWeb are quite a bit better than the barely-filtered CommonCrawl dumps GPT-3 trained on, but someone had to invest in making them. (To say nothing of whatever exotic bespoke robotics/physics/etc datasets OpenAI and Anthropic are investing in.)
Ilya Sutskever has been saying more compute = better perf since at least 2012.
... the real conclusion now is that if you have a large big data set and you train a very large deep neural network success is guaranteed ...
https://www.youtube.com/watch?v=-uyXE7dY5H0&t=1185 In the end, the network's size is limited mainly by the amount of memory available on current GPUs and by the amount of training time that we are willing to tolerate ... All of our experiments suggest that our results can be improved simply by waiting for faster GPUs and bigger datasets ...
https://proceedings.neurips.cc/paper_files/paper/2012/file/c...As if drawing a line through few points on a chart is a mystic art.
Even the best agents still can’t strategize their way out of a paper bag, in my experience.
The human remains the most critical differentiator for non-trivial work. (Different story for trivial projects that agents can solve independently. Here the human becomes a fungible operator.)
Somewhat related: I dont think AI is coming for everyones jobs, and I will never understand the argument that it is, because I always have the ability to use the AI and add my "human ingenuity" to it. So why _wouldnt_ that be a better option than just raw AI?
Well, that hasn't held true for chess, so I'm not sure how universal the domain is where human ingenuity will always augment AI. Also, there is a cost floor to human ingenuity, in that it requires the time of one living human, and their metabolic sustenance over that period. So the added benefit of that human has to be substantially more than epsilon.
- AI might just be good enough, and beat you at price and speed
- you might not be the one to have access to state-of-the-art AI - many more
From my point of view, AI has actually already surpassed humans in 2 ends of the spectrum
- math (logic)
- digital arts (ingenuity, with some heavy lifting to call it that, but still)
Filling the space in between could be reasonable expected to be a matter of time, money, data.
Current AI is actually good at search, but I can see how people mistake one for the other in the context of math, for example.
1. Eventually your "ingenuity" will be redundant. Human-machine chess players were dominant for a brief period.
2. Cost. Even if you are adding something, it might not be worth it to whoever is paying you when you cost 100x what the AI alone does.
It is not clear it is always a better option. It may be for you, but not for someone who doesn't know what they are doing and drives the AI down the wrong path. For example, I don't think adding my "human ingenuity" to Claude would help solve the Riemann Hypothesis, and would likely be counterproductive, because I have a 3rd-level understanding of math.
Curiously, the timeline for this proof only starts roughly around the same time that Tao said this. IIRC, work on the disproof for the Navier-Stokes smoothness theorem only started a week before it was published. Maybe I'm wrong, but it seems like AI companies have only recently started to put real effort into LLM-assisted proofs.
Edit: I read the PDF, Levant claims 11/16 for GRH.
Very funny argument either way.
- All of humanity, unaware their pants have fallen around their ankles
Just to point out how ridiculous it sounds to claim that because X (Riemann) is possible that anything is possible.
Cancer, aging, mental disfunction.
Law, creative film making, market inefficiency of any kind.
Writing programs using something other than your voice and having it be completed faster than your mouth can complete the sentence.
Pants around ankles.
- someone else
So maybe not too far off.