I think that media coverage simply reflects the growing anxiety among the public about the capabilities of new models and what future awaits us. Personally, I think that this anxiety is justified.
If this leads to tough regulations about AGI/ASI (e.g., a permanent ban), I doubt that AI companies would benefit financially.
AI progress is linked to most of these dangers, actually:
- it will likely cause massive unemployment, leading to rampant wealth inequality
- we are now seeing some use of autonomous weapons in real conflicts
- increasingly relying on LLMs is arguably a form of technology dependence (and cognitive dependence)
- datacenters have a non negligible environmental impact
I agree these dangers are connected. The danger comes from human institutions and incentive structures which (having existed long before now) are being exploited and exacerbated by LLMs, making them harder to ignore.
It will be too late by then! At the very least, an immediate pause of the training of frontier AI models is necessary to figure things out. Personally, I am not particularly thrilled to rush into the bleak future that few AI companies have dreamed for the rest of humanity.
Yeah, I see a huge disconnect between what enthusiasts see as 'the future' with AI, and what the actual owners of the technology are saying and doing. They aren't building a utopian future for all mankind, and are saying it openly.
Yet jump on any online forum and there's always this wishy-washy "oh lets see what happens!" mentality.
> They aren't building a utopian future for all mankind, and are saying it openly.
They aren't going to stop even if you pass laws to halt progress. The Pentagon is a direct customer of Nvidia, and they'll continue to develop cutting edge surveillance and war AI as they have for the last 30 years. It's facetious to imply that any type of LLM regulation would legitimately prevent AI (or even LLMs) from being used for any evil purpose.
> there's always this wishy-washy "oh lets see what happens!" mentality.
Why would the US halt commercial AI research if China could continue developing and exporting equivalent capabilities? They already can today. We already live in that "see what happens" world since unaligned finetunes of frontier models are freely available on HuggingFace. It feels like OpenAI and Anthropic only cry foul because neither of them can IPO in the current economic conditions.
I don't have the energy for an argument like whats being instigated here, and what you're saying isn't quite relevant to what I'm saying sorry. (And if that's not the case, then also sorry. Social media has me beat)
What I'm saying is that the owners/ceos of the companies are saying they expect unrest/disruption/collapse from their technology (eg, Sam Altman saying exactly this), and yet the enthusiast base say this isn't the case. If I'm having a house built, and the architect says it will fall over, but other home owners say it won't, I feel I need to still trust what the architect is saying.
Sam has been repeating the idea that society would collapse because of AI long before ChatGPT existed whatsoever[0], it's his pet obsession and is part of why YC invested in him. It is a sci-fi delusion, and equally exhausting to hear people insist that it has to be taken for granted.
He is absolutely not alone in that, and even if you think it is delusional it is a shared opinion between many of the AI figureheads. My point being that the people in charge are telling a different story to the fans.
I think that we are speed-running towards a future that very few people really want, and I find it terrifying that few companies feel entitled to choose this future for the rest of us.
Some of the arguments in this document would call for an immediate, global, pause on frontier AI training: we need time to consider how and to what extent AI should be part of our future.
Personally, I can't picture a scenario where humanity thrives alongside an alien super-intelligence, especially if it cannot be fully controlled. Let aside super-intelligence, I am not even sure that deploying an AGI that replaces (instead of augmenting/assisting) humans in most intellectual tasks would be in the best interest of our species. This conversation has to happen, on a global level and as soon as possible.
By googling "4x4 matrices multiplication 48" I ended up on this discussion on math.stackexchange https://math.stackexchange.com/questions/578342/number-of-el... , where in 2019 someone stated "It is possible to multiply two 4×4 matrix A,B with only 48 multiplications.", with a link to a PhD thesis. This might mean that the result was already known (I still have to check the outline of the algorithm).
One of the authors here. We are aware of the Winograd scheme, but note that it only works over commutative rings, which means that it's not applicable recursively to larger matrices (and doesn't correspond to a rank 48 factorization of the <4,4,4> matrix multiplication tensor). The MathOverflow answer had a mistake corrected in the comments by Benoit Jacob.
More details: the Winograd scheme computes (x1+ y2 )(x2+ y1 ) + (x3+y4)(x4+y3)-Ai-Bj, and relies on y2y1 (that comes from expanding the first brackets) cancelling with y1y2 in Bj=y1y2 + y3y4. This is fine when working with numbers, but if you want to apply the algorithm recursively to large matrices, on the highest level of recursion you're going to work with 4x4 block matrices (where each block is a big matrix itself), and for matrices Y2Y1 != Y1Y2 (for general matrices).
Here is a website that tracks fastest (recursively applicable) matrix multiplication algorithms for different matrix sizes, and it stands at 49: https://fmm.univ-lille.fr/4x4x4.html
From some conversations on Twitter, it seems plausible that the rank-48 decomposition of the 4×4 matrix multiplication tensor really is new; and that perhaps where things have gone awry is attempting to summarise this result in a more lay-friendly manner: the algorithm in that post apparently doesn't constitute or imply a rank-48 tensor decomposition.
As already noted in a post by fdej further down, Waksman's algorithm from 1970, which works over the complex numbers, requires only 46 multiplications (and I guess, divisions by 2, which may or may not be relevant depending on your actual ring).
The answer says "For rings in which division by 2 is permitted". Is there the same constraint for AlphaEvolve's algorithm?
Edit2: Z_2 has characteristics 2.
Edit: AlphaEvolve claims it works over any field with characteristic 0. It appears Waksman's could be an existing work. From the AlphaEvolve paper: "For 56 years, designing an algorithm with fewer than 49 multiplications over any field with characteristic 0 was an open problem. AlphaEvolve is the first method to find an algorithm to multiply two 4 × 4 complex-valued matrices using 48 multiplications."
If you don't want to allow division by 2 then there is Winograd's algorithm from 1967 which works over any commutative ring and uses 48 multiplications for 4 x 4.
So, did LLM (namely Gemini-Flash) helepd with the combinatorial optimization process? I'm sure not all of their discoveries (one on kissing numbers, etc.) have previous solutions in some other form, but yeah these findings looks more like very large combinatorial optimization tasks.
reply