The risks may be much higher for women, but in most places the base rate is still very low. "Assaulted by a stranger on the street" is actually fairly low in the risks for almost anyone: the vast majority of assaults, especially sexual assaults, are perpetrated by someone the victim knows.
(This also doesn't necessarily mean that such concerns are irrational, just that the risks are not so high that you should expect every woman going out for a run in the dark is eventually going to be assaulted)
I think the opposite is probably the way to go: strip out some layers of abstraction, try to keep them to only the really valuable ones. In my experience LLMs are better at dealing with more concrete codebases.
It's also a lot cheaper to experiment with different abstractions within a codebase to really get them right, and especially to evolve them as the codebase changes.
It's not uncommon to seek legal costs in a lawsuit. They are usually judged and awarded seperately to the main claim, so it's not always loser pays. It can of course be weaponised to make the potential price of losing larger, though. A claim with unreasonably large costs is less likely to actually be granted, but often the lawyers are aiming to not get that far.
I think it's useful to have, even if just to have a calibration of what priority level is "never get done" (which will probably change over time) and a list of things that have been considered and fall below that level. Obviously rearranging the priority of stuff below that threshold is pointless but that's a seperate problem to having the list.
I think it would be much less of a problem if there was an explicit, clear delineation between the legal text and the art. The HTML version is much worse in this regard as well: the license in question reads like part of the text of the document (though the meaning it has in that context is quite ambiguous).
Doesn't seem very likely. The Romans didn't struggle to project power far beyond the range of a spear.
The definition of a state is a monopoly on the legitimate use of force. It's not a statement about the military capabilities of the government, it's a statement about the relationship of the people with the state (that is, a state exists when people act like it has that right).
Or, more likely, they will make a compliance checklist that is sufficiently difficult to comply with fully they'll have a means to deny most or all of the claims against the policy. I don't think the insurance industry is very efficient.
I don't think so. You can't practically buy electricity from halfway across the planet but you can buy tokens. There's no local infrastructure to monopolise.
None of them are a good 'Is this spam?' heuristic, though. All they do is ensure the sending domain isn't forged, but most spam, even phishing emails, doesn't really rely on that. Spammers are much more motivated than average to jump through the hoops necessary to set it up, as well. Any anti-spam signal needs to be much more expensive for spammers than regular senders.
>I wonder why these bright abd capable people didn't join a company that actually pays reasonably.
Sounds like they do when they get the chance. I've worked for a company that didn't have quite as crazy a hiring process but they had the same basic profile - very good at hiring for potential in juniors, but underpaid for the ability they wanted. As a result they really struggle to hire seniors and are often a stepping stone to a better paid position. They do manage to keep some people until retirement, though (a very bimodal distribution - generally either brilliant but don't care to work anywhere else, or utterly useless and can't get a job anywhere else)
Eh, I get what you mean. AI especially seems to have performance that is much more difficult to reason about than human intelligence (where most aspects of mental capability are surprisingly strongly correlated). But in this context, if you broadly focus on intelligence as the ability to complete problem-solving tasks, models have been improving pretty uniformly at the same time as getting cheaper.
>What is the unit of measure for “intelligence per dollar”, and how do you measure it?
It's going to depend on the task (and this is prone to cheating), but you can get a decent idea by looking at task benchmarks and the cost of completing them.
I was inclined to agree but Anthropic's latest S-1 statement only includes a paper loss due to their value going up. They're objectively making money now, the only question is can they keep it up and can they grow enough to be able to pay the bills they've signed up for in future?
For sure, which is ironic given it was the application GTK was created to support. But even so, I don't think I've ever really seen a GTK-based application I actually liked (the rest of the UI is going to have to make up for the atrocious file picking dialog, for example)
Yup, it can create a situation where you have a huge incentive towards workarounds and hacky fixes because doing it properly incurs huge costs. This is especially bad when a fundamentally quite bad design makes it through certification (most processes don't really prevent this).
If you are counting by bytes and including all sensory input, sure. In terms of amount of text, it's not close in the other direction. It's very difficult to make an apples to apples comparison here.
It does seem like they are much better in the small as opposed to the large. I agree on the judgement being questionable, and every report on making large swarms seems to suggest it's relying on spectacularly diminishing returns on the performance of a single agent, especially a single agent with a human in the loop.
Apart from the general difficulty of organising like this and actually evaluating the marginal revenue from an additional employee (it's hard enough to attribute to your existing employees!), this marginal revenue is likely to have diminishing returns as the company grows and from a risk management point of view you will want to have some margin so that you don't flip into losing money as soon as business slows a little.
The internet is a loose connection of people and organisations that is very difficult to coordinate. A lot of people do block these packets, which is why they're basically vestigial, but it's not something that's really worth spending co-ordination effort on when it's something that's easily implementable by those who want to and there still isn't universal support for things like standards that prevent BGP hijacking.
The facts will depend on each individual datacenter proposal. I'm not saying they all should be built or that none should be built, but that each one should be based on understanding of what each proposal entails that is grounded in reality as opposed to emotive broad-strokes generalisations that are either false in general or not necessarily true for every example, and ideally the tools available in the decision making process empower the former and not the latter.
Systems still take some effort to actually exploit. For example, the vast majority of bugs they found here are local privilege escalations, and most security people have long learned that user separation in the kernel is not a particularly strong security boundary. Network-exploitable vulnerabilities are a lot rarer, and it's noteworthy that there have been a few big ones discovered recently, but it's a much smaller number overall.
The heads of the big AI labs do seem to be pretty catastrophically bad at PR, I do agree. But this is hacker news, I thought we were meant to care about and look at the facts and not get swayed by the PR?
There can be multiple effects for a given action. The main objection I have to this argument is the idea that the only beneficiaries of any corporate decision are the shareholders.
I live near a sewage plant. One or two days a year when I am cycling past the fence on my commute, I might get a whiff of something. People really overestimate the problem.