Search results

You, not me, brought up the Trump administration’s denial. I was responding to your nonsequitor, pointing out its irrelevance. And then you call me out for it being irrelevant. Sheesh.

The fact that the trump administration actively encouraged people not to protect themselves from the virus has absolutely nothing to do with its origin! Or how to avoid a future pandemic. And please keep your ad hominems to yourself, kind sir.

Doesn’t matter? Because who cares if we learn from our mistakes? Millions die, shit happens. That’s all in the past. Every safety critical industry in the world works by learning from its mistakes. But in this case we aren’t. The virologists convinced the politicians that “it doesn’t matter if...

lol thank you. This forum used to be so informative. The group think here is really sad.

The primary source documents for these communications can be found in the House Select Subcommittee's March 2023 memorandum and their July 2023 report regarding the Slack messages. And the raw communications had leaked publicly long before the politicians got involved, if you don’t trust them...

Raising the possibility of a lab leak is not at all strange. Occam’s razor would find it extremely unlikely that this deadly virus naturally emerged across town from the one place on the planet known to be performing gain of function research on spike proteins in Covid viruses at the time...

I get that lots of people here hate AI coding or just AI period. Maybe they feel threatened that their hard earned skills are less useful and the kids are leaving them behind. It sucks but it happens. Ask any middle aged athlete. I genuinely don’t understand the downvotes on the assertion that...

The fact that comments like this get downvoted so aggressively makes me fear for the audience in this forum. Somehow Ars readers, whom I used to consider so astute and technical that they’d add significant content to most articles, have become devout AI deniers? Anybody saying that AI is...

So aside from being big and having lots of power, what do these satellites do?

SWE bench involves fixing real bugs in real projects from GitHub. They are simple bugs, all in Python I believe. The agents have a max of 1 hour to get the tests to pass. It’s generally assumed that skilled developers could solve most or all of these within an hour. So high 90’s score for a...

I think what you mean is that they often suggest a much stronger genetic basis for many conditions than rigorous science agrees with. Eg they’ll say that a certain SNP makes you more likely to have a certain condition, when the data behind that assertion is thin. Your literal words say...

It is exactly over-fitting. Over-fitting is the bad outcome we can directly measure. Sharp-minima theory explains the HOW/WHY overfitting is happening by looking at the geometry of the loss surface, specifically for neural networks. The geometric understanding of why something is overfitting...

It’s not at all clear who Google thinks they are fooling here, or even what the threat model is they’re protecting against. My guess is they hope to make it safe against bad actors within Google - but that necessarily means rogue employees not the organization as a whole deciding they DGAF about...

Deliberate typo-joke. Schmidhuber was really ahead of his time. Brilliant, but somewhat sharp elbowed, and always had a chip on his shoulder that Bengio, LeCunn, and Hinton didn't invite him to help steer the ICLR reindeer games when deep learning started to take off. Go to any big AI...

This is solidly interesting research. The sharp-minima theory (which was originally published by Jugen Schdmihuber back in the 1890's but rediscovered in 2015) explains a geometric difference between model weights that generalize well and those which are effectively memorizing the data. After...

The study authors just don’t misunderstand different types of LLMs. They tested IFT instruction fine tuned models, which is understandable because that’s what basically everybody uses. IFT models are literally fine tuned to follow instructions, which to start with makes them less authentic. Also...

There is an appropriate deference to AWS engineering practices. And I guess it’s fair to say that you need more than undergrad training to have the imagination to foresee this kind of failure. I’d guess the enactors typically finished their jobs in a couple minutes. If I was looking at this...

(I’m not an amateur at this stuff, but whatever. We should all be able to reason about what happens, if we are given useful post mortem background.) The problem here with having each enactor GC after itself is that they designed it to have multiple enactors operating simultaneously. This...

Yes the failure required both the enactor to run slowly, and its plan to get deleted. But it’s a real stretch to call those two events when they are 100% reliably correlated. Every single time an enactor runs extremely slowly, its plan will get deleted. By (flawed) design. The plans get deleted...

While I agree with you that SPOF is way too simplistic for what happened here, I don't think it's right to say "multiple things went wrong" either. One DNS enactor ran way too slowly, and that triggered a "latent race condition" a.k.a. bug. There was really only one incorrect assumption /...