I have yet to read a post here on LW were someone write about a frontier model that have solved a "real" problem were the person really had tried for a long(-ish) time to solve it but failed and then the AI solved it for them, like a research problem, a coding problem, a math problem, a medical problem, a personal problem etc. Has anyone experienced this yet?
I don't have any first- or second-hand experience with that happening. I see occasional articles about protein and materials research that LLMs have massively accelerated, but my suspicion is the main value currently is NOT cutting-edge significant problem solving.
The "mundane utility" section of Zvi's writeups have very good examples of what it IS currently good at now. There's yet a long way to go to handle long-running multi-step creative analysis that the top few percent of humans are engaged in. What is the shape of the curve (both of capabilities and of "intelligence level" metrics) is currently not known.
You mean specifically that an LLM solved it? Otherwise Deepmind's work will give you many examples. (Although there've been surprisingly little breakthroughs in math yet)
Yes, I meant an LLM in the context of a user that fed in a query of his or her problem and got a novel solution back. There is always debatable what a "real" or "hard" problem is, but as a lower bound I mean something that people here at LW would raise an eyebrow or two if the LLM solved. Otherwise there are as you mention plenty of stuff/problems that "custom" AI/machine learning models have solved for a long time.
I am starting to believe that military use of AI is perhaps the best and fastest way to figure out if large scale AI alignment is possible at all. Since the military will actively seek to develop AI's that kills humans, they must also figure out how not to kill ALL humans. I hope the military will be open with their successes and failures about what works and do no work.
The heat is on! It seems that the export restrictions on Nvidia GPUs have had little to no effect on Chinese companies abilities to make frontier AI models. What will the next move from US be now? https://kling-ai.com/
GPT-4 level models can be trained with mere thousands of GPUs. Export restrictions for a product that's otherwise on the open market aren't going to work at this scale, and also replacement with inferior accelerators remains feasible. But one GPT is 30x compute, and procuring 100K or 3000K GPUs (or many more in their inferior alternatives) is more reasonably a practical impossibility.
Huawei claim they are catching up on Nvidia: https://www.huaweicentral.com/ascend-910b-ai-chip-outstrips-nvidia-a100-by-20-in-tests-huawei/
Very serious negative update. And they practically say: we read Sora paper and then replicated.
As I wrote in another post, this can easily turn into another TikTok tool, where dumb westerners spill personal info into what amounts to a Chinese intelligence gathering apparatus.
Just wait until more countries that do not share western values get their hands on tools like this. I think that the only way that social media can survive is mandatory ID. If Airbnb can do it, I am sure Meta, X, Snap etc. etc. can do it. And... call me old fashioned, but I rather not share ANY personal information with ANY intelligence service