Yes people have to service the use case of a single person. All sorts of models exist that are designed to be ran at home. Yes, you need a relatively beefy graphics card for it but if your machine can handle the latest video games at good graphics it can handle these local models. How well they compare against these things remains to be seen but apparently OpenAI released gpt-oss-20b which is comparable to o3-mini apparently in terms of reasoning power. There's also oss-120b which does require at least a company server to run but this should well be within the budget of a university.
The paper does not resolve P versus NP, but it does make an important advance in a closely related area. To prove that P ≠ NP, it would be enough to show that every algorithm for an NP-complete problem requires superpolynomial time. We cannot prove anything remotely that strong. For explicit NP-complete problems in unrestricted computational models, we cannot even prove superlinear lower bounds. There is therefore an enormous gap between the lower bounds we can prove and the superpolynomial bounds we would need.
VP and VNP are closely related algebraic analogues of P and NP. Here the paper proves new lower bounds for computing the permanent, a VNP-complete polynomial, in particular models of arithmetic computation: roughly (n^2\log\log n) arithmetic gates for unrestricted division-free circuits, and (n^4/\log n) size for the more restrictive formula model. These are still polynomial bounds, so they do not separate VP from VNP. But lower bounds on the resources needed to compute explicit functions are exactly what would ultimately be required for such a separation, and meaningful lower bounds of this kind are exceptionally rare.
You are confusing who is the party that is injured--as many in this thread are. We are not talking about the consumer who buys a non-fairtrade t-shirt, when he would rather had a fairtrade one. It's about the t-shirt producers who is legitimately fairtrade but whose business is now in the shitter because of a lying AI.
LLMs can’t lie. They are incapable of telling either the truth or lying. To the extent that they are in any way useful, it’s recognizing that they are text generators attached to crawlers and other tools that can with the right inputs produce useful generated text that may also incidentally be correct or incorrect.
Businesses might (well, will) suffer because people are misusing AI, but it is a misuse to do anything with it without an additional verification step.
To be clear here, I have no issue with Google taking it on the chin in cases like this, but what the comment I was originally responding to had this:
> errors can be so subtle that it is not possible to recognize them unless you spend an hour researching every fact presented. at that point, what's the benefit of AI? nobody is going to do that.
And my point is this: if it matters, verification is not optional. If it doesn’t matter, then fine, skip the verification step, but if you’re taking whatever text is generated by a GPT at face value without understanding what that is or being able to determine the source inputs for the “claims” it outputs, then you’re part of the problem because sometimes the source is just a GPT-generated web page, and that’s obviously not trustworthy. Sometimes it’s a MediaWiki site page that doesn’t actually exist, but because it’s MediaWiki it’s not going to return a 404. Using a tool requires understanding it including its failure modes, and in the case of LLMs that means: trust nothing, verify everything.
Still - a judge on another case will probably factor this one into their decision, but if they have a radically different view or the case is sufficiently different, they could just as well come to a different conclusion.
I have no exact prompt to share, but he does write:
> Appreciate the insight! If it's at all of interest, this was a one-shot (supposed) solution in about 80 mins, unlike some other problems like 851 that took over 20 continuations totalling perhaps 15-20 hours of reasoning time.
reply