You didn't break embargo if you didn't know about it. Somewhere in the process things are broken. You are actually one of the better guys out there. There are people who actually knowingly broke the embargo, and there are even people who willingly developed PoCs just to brag on X. You are fine.
If artificial intelligence exceeds human intelligence, human beings will not do frontier math, or frontier anything for that matter. Why is this so hard to grasp? Monkeys don't understand or use calculus, but that's okay, they just continue to live like monkeys.
Why are all the seemingly smart people on HN so narrow-minded to the point of speaking idiotic nonsense? What do you think their 3.6B MAU are, all bots?
"Catastrophe requires multiple failures – single point failures are not enough."
I disagree with this point. From what I have seen and experienced, there are so many SPOFs in the real-world complex systems that do not have adequate defense or even any defense at all. The only reason catastrophe didn't occur (often) is only because the natural failure rate is extremely low. Like the probability of air traffic controller making a critical mistake is very low, maybe once every few years, but when it happens, the SPOF will almost certainly cause a disaster, and the system (ie. the way air traffic is controlled today) has been around for decades and is so outdated but the cost of changing the entire process around the world is simply way too high so we just take this as "normal".
I tried it and it performed poorly on my private benchmark problems. The overthinking problem is real, it takes 5-10x the reasoning tokens than comparable models. It is a sign of inadequate training of the base model and it is using more reasoning tokens to compensate for that. I also noticed that it is likely to get into somewhat repetitive reasoning and forgetting about some user requirements, suggesting that it could be the side effects of using 3:1 linear attention vs full attention.
Considering the amount of AI slop being generated today and LOCs that no one actually reviewed, I doubt we will run out of vulnerabilities to exploit...
What Dario said is half true. The untrue half is that the ship to stop China from overtaking the US in AI has sailed long ago. While China is busy getting their frontier models trained on smuggled chips using distilled data from millions of shell accounts, the US is busy debating whether to check voter IDs, or secure borders. It is just delusional to think the US is actually capable to stop China from getting the chips they want.
The comments here are ridiculous. Do you know how much ChatGPT has helped families, esp. those without access to the top medical doctors and resources around the world? My own experience with it so far have been nothing but positive. It beat all the doctors in various hospitals that we have consulted so far on breadth and depth and the best thing is you can keep asking follow-up questions on anything and it gives you all the details you need without being rushed.
The fact that it is not 100% reliable shouldn't be basis for restrictions. Even top trained doctors are far from 100% reliable either.
I don't think you even understand the problem. The determinant needs to be a non-zero constant AND you need to prove that particular map is not globally injective, meaning you have to find at least two points mapping to the same value. Of course it looks easy when someone shows you the counterexample.
What a great bullshit detector signal it is if someone's involved in the DAOs and NFT stuff. Don't waste your time on a fake AI-generated one-man shell company. The company had like 3 investments since 2017, first $2.5k from school startup grant in 2017, and two more from incubator programs in 2023 and 2024, for "augmented reality". And now you are the most innovative chip designer in AI. Nowhere on Chenming Hu's Linkedin said anything about being an official advisor for this one-man company. I can take a photo with Messi, does that make me a top soccer player?
Chinese frontier models don't need to catch up in every category. They just need to win in coding and that's exactly where they are going. The gap went from 12+ months to 1-2 months with the latest release of GLM 5.2 and coding is a task that you don't need heroic efforts to find rare and long-tail training data, you can just outsmart your competitor by optimizing algorithms and training recipes. This is something they can do at scale with the money and talent pool.
The familiar Chinese recipe for success: Always copy and imitate first, even if it is inferior, always make it cheaper or even free so that the original innovator will be burdened by brutal price competition and much bigger R&D costs and cannot keep up in the long run. Then the copycat will win in the endgame.
It's all about specific people at the leadership. No structure is immune to corruption because people come and go all the time. The reason why Nintendo is so resistant to slop for so long is because their key leadership people were homegrown all the way from entry-level jobs and are still there.
And why is that a good thing? The average user can't even spell Anthropic. Why do you think they can safely pick a third-party model provider that could harvest the hell out of their conversations? The control of ecosystem is part of the privacy and security. My mom's Android phone has like 100 apps that she had no idea how they were downloaded. For real user choices, the vast majority of users just want a phone that they can trust and don't have to be a techie to avoid being exploited. They can choose to buy a phone that can be built from legos, OR they can choose to buy a phone from someone they trust to get the privacy and security taken care of for them. This is the real user choice.
Seeing how many smart human beings commenting on this post mixing up concepts like intelligence, understanding and consciousness and falling for all kinds of logical fallacies makes me believe humans are no more advanced than a good next token predictor :)
He already scanned the codebase with Codex Security and a whole bunch of other AI tools, and fixed 200-300 bugs and CVEs. On top of that Mythos found 1 more bug and 1 more CVE is already impressive.
If you want to take the risk and install some unsigned software on your machine, go ahead, but don't blame Apple, who is gatekeeping for the entire ecosystem for making the decision to keep the restrictions in place so that the other 2.5 billion users don't fall victims to malware defenselessly. Also, as a rule of thumb in cybersecurity, never underestimate human flaws or overestimate your ability to overcome them. Even the most brilliant experts cannot possibly know everything and make zero mistakes, let alone "the users" you are talking about. It is pure illusion that "the users" know exactly what's running on their machine under the hood. We should be thankful that Apple is willing to hold the lines and go this far to tighten security up when nobody forces them to. It is probably one of the best thing coming out of Jobs' relentless push for privacy and security on the iPhone.
Totally agree. I found the results surprising because a bunch of languages are faster than C++. Then I looked closer. The requirements are self-conflicting, No SIMD, but must be production-ready. No one would use the unoptimized version in production. Also looking at the C++ implementation, they are not optimized at all. This makes this benchmark literally pointless.
Judging from their website, all links eventually point to either the VPN extension download website, or a signup link. I'm not surprised if some nation state supported APT is behind this shit.
When SpaceX came about, they said it was impossible for the rocket to come back from space and get reused. They said it wasn't going to work to combine multiple thrusters to form a big thruster and be reliable enough. When Starlink was introduced, they said it was stupid because the bandwidth is too small to be useful. Where are we now? 10 years ago, AI couldn't even beat a high-rank amateur Go player, let alone the best of pros. Everyone takes the excuse of dimensionality curse. Now what?
People who only look at the past/present and conclude impossibility are never going to be the ones who invent the future. Even math and science evolve, let alone engineering. The problems described in this article don't even remotely feel like the kind of barriers we faced when Go was solved, when protein fold was predicted and when LLM was solving problems with one prompt. If there is a strong NEED for datacenters to be up in the space, there will eventually be datacenters in the space.
1. In Silicon Valley, people are not bounded by non-compete clauses and can come and go at will. So fungibility is a top priority for any tech company. The only way to do that is to make sure expertise is shared across the team and not monopolized by one or a few old-timers.
2. Eng teams that have mostly old-timers tend to get stale and slow in changes. This is bad for products that need rapid evolution or new ideas to break status quo. New engineers have way more incentives to make changes to prove themselves and collect credits, while old-timers tend to play safe and stay on the side of stability.