That's true but what they were saying is that people may be willing to pay for AI even if they believe it has terrible externalities, not that they are insincere in their beliefs.
I read about a canal built in Arizona, they decided the 3% (iirc) losses due to evaporation weren't worth pursuing. Pretty hot and dry climate. So, this may make the solar panels cheaper but doesn't justify maintaining an awning.
If we're going to judge Snowden by the standard that he could've rectified his situation through a radical act of self sacrifice, then every US president has had the opportunity to pardon him and has failed to do so. It's an equally sensible proposition. It's plain to see that no US president would entertain the notion. Similarly, it's not reasonable to expect someone to return from exile without being offered some kind of clemency.
No radical act of self sacrifice would have been necessary had he only leaked illegal or possibly illegal programs. Instead, he went to Hong Kong and leaked a list of compromised Chinese networks and systems in hopes of getting asylum from China.
Just like with John Kiriakou, the only man ever convicted in connection with the torture program? He went to prison for two years despite only ever leaking information about illegal programs, lost his wife, lost all his money; is that the mercy you expected Snowden to submit to?
Kiriakou gave an interview to ABC describing waterboarding on December 10, 2007, in order to defend the practice. The government did not charge him with any crimes at the time. Three years prior, Newsweek had reported on a secret memo from Yoo justifying the use of waterboarding on Abu Zubaydah. None of the people who leaked that memo were charged with crimes. Three months prior to Kiriakou's interview, "current and former high-ranking U.S. intelligence officials" said that three prisoners had been waterboarded, including Zubaydah. None of them were charged with crimes.
Four years after Kiriakou's interview, he was charged with giving the names of CIA agents to two journalists, a crime under the Intelligence Identities Protection Act punishable by up to 10 years imprisonment, not for describing waterboarding, which if he had leaked would have been covered by the Whistleblower Protection Act.
He said "as Americans, we're better than that", does that sound like defending? That he was conflicted about it isn't too surprising, he was ratting out his colleagues and the institution he had worked for through the entire of his professional life.
To the rest, you've fallen for their trick. Three CIA threw the book at him, and indeed tried to convict him for his leaks: "The five-count indictment against Kiriakou charged him with violations of both the Intelligence Identities Protection Act (for allegedly revealing the identity of a covert agent) and the Espionage Act (for allegedly revealing national defense information to unauthorized individuals such as journalists). He was also charged with making false statements to the CIA about a book that he was seeking to publish."
The Intelligence Identities Protection Act is just the charge that stuck. When you're revealing government secrets, it's very, very hard not to accidentally cross the line, especially when you consider that the whole justice system is arranged in their favor, not yours.
The result is that lots of people say they support whistleblowers, but anytime someone actually whistleblows they back the government line and say "they went too far". Why is the onus on whistleblowers, acting in secret and with fear of retribution at every step, to perfectly separate the illegal actions from the legal, and punish only the former? Why is the onus not on our government, the elected representatives who ought to serve us and obey the law, to not commit crimes against humanity in the first place?
If you demand perfection from whistleblowers but accept faults in your government, don't be surprised if you don't get many whistleblowers.
That's a very optimistic view of the US justice system's treatment of whistleblowers (surely they would contend that all of their programs were legal), but now his being granted asylum by Russia isn't evidence but a new incongruous fact we need to explain since the hypothesis is now apparently that he is a Chinese turncoat. So I don't see how this helps the case.
> now his being granted asylum by Russia isn't evidence but a new incongruous fact we need to explain since the hypothesis is now apparently that he is a Chinese turncoat.
Putin wanted him in Russia, and Snowden happened to be passing through. Simple as that. I already explained the clear evidence (from Snowden's own mouth) that Snowden initially wanted to gain asylum in Hong Kong by leaking hacks of Chinese infrastructure in https://news.ycombinator.com/item?id=49784968.
Snowden didn't grow up wanting to be Chinese. He simply misread the PRISM documents and imagined a USA that completely ignored the 4th Amendment. Somebody in such a state could easily imagine a USA that would ignore the 8th Amendment, and so he had to flee in order to get the news out about the former, which he talked about nonstop the moment he got journalists' ears. It turned out that he was wrong about PRISM, and he was wrong about the government ignoring the 4th Amendment because the only extant program that might have been in violation was shut down. https://news.ycombinator.com/item?id=49785016
He said so himself. I take him at his word. Meanwhile, the actual USA stopped doing the one extant program in his leaks that was ruled a violation of the 4th Amendment.
> To move it to another program, classified and unreported, under which it could hide again. Don't be naive.
Why? They first made a more limited program to comply with the law and then shut that down because the data wasn't useful. If they made another program exactly like the original, the SSCI would shut it down, or a whistleblower would point out a program that they know to be clearly illegal, whereas the original program was arguably legal, with the government's justification in the leaks. After the ruling, the government would not be able to provide justification for obtaining that data from the telcos with their old legal argument because it is invalidated by precedent.
Sorry, but is Russia or the USA the one most likely to kill Snowden? Oh, yeah, it isn't Russia, because Snowden avoids saying things that could trigger his KGB handlers to disappear him... Like denouncing a genocidal war of aggression against a neighbor, or the assassination of journalists and political opponents, or the repression of any opposition in the streets...
By leaking American info, Snowden painted a target on his back that can be reached by the USA and all allies with extradition agreements. That leaves him with very few superpowers who won't hand him over to the FBI/CIA and execution or imprisonment.
If I were in his shoes, I too would relocate to a superpower not allied with the USA and avoid getting involved in anything political over there.
What the current president does now has no bearing on what was done then, with a somewhat functioning Supreme Court and an adversarial Congress. The conspiracy theories spread by Snowden helped lead to Trump, and that is the main point I want to get across. While spreading his conspiracy theories, he also helped the CCP, and that is something for which he would stand trial for in the U.S.
If he stayed in the U.S., his punishment would already be over. Now with the current president directly telling DOJ what to do, who knows what would happen if Snowden were to stand trial today, but that isn't a possibility Snowden faced when he fled.
I had a lot of fun, the scan line effect looks really good and the obstacles are interesting and challenging. I'd have liked to use my game controller, there's a browser API for it. And to have changed my control settings. I thought both mobile and desktop controls were good though.
I've been working on a browser game too lately. It's a pretty great environment to work in. Three.js + WASM have cured my desire to learn Godot, at least for now.
If this hypothesis were true then a spy agency may want to rewrite responses. Every tool in an agent's harness becomes an remote procedure call you can make on that machine. Including a tool to execute a shell command, in many. A harness is completely isometric to a backdoor, it's the same code written with a different intention.
Ad-hoc hypothesizing ("escape hatches") are dangerous but not quite invalid. Eg, they failed to find gravitational waves until they did, and you could have viewed building yet another more sensitive detector as a similar exercise in refining a hypothesis that you keep receiving contrary evidence for. Sometimes you really did just underestimate how difficult your hypothesis was to demonstrate. Maybe Meta will be destroyed in 2027, or whatever.
The problem with conspiracy theories is more that they have a ratchet-like quality where counter evidence reaffirms the theory in your view and you can only ever get more confident. We should have been increasingly skeptical of gravitational waves to some degree as we failed to demonstrate them, even though we didn't abandon the hypothesis and it ultimately prevailed. But if you adopt a wrong idea, and people try to demonstrate that to you, and you take that effort they're putting forward as a sign that you are correct and they must be hiding something from you, it will be very difficult for you to realize your mistake.
So, as long as you are less certain than you were before, I don't think rolling your prediction over into the future is necessarily conspiratorial or a mistake.
I don't think you can extrapolate that measurement across multiple sequential draws like that. We presumably are comparing against a single trajectory rather than a tree of trajectories. So once we make the wrong choice and step off of the blessed path, we have no way to assign a ranking to the next token; it's error is undefined.
I've seen LLMs self correct in chains of thought ("because of foo and bar, I need to... Wait, bar is not true, so that won't work") so I have to imagine this is a massive overestimate, errors do not necessarily compound.
I would say they do compound until proven otherwise.
Having "Wait, bar is not true, so that won't work" is not necessarily a correction. In fact, the problem is: across a long text it is a correction of a single mistake, but we are talking about thousands here.
But yes, of course that was a rough estimate. But the problem is - we don't really know what we are measuring here. Maybe there's a 2,000,000x difference of intelligence between coding indexes 52 and 50. By some measure that just feels small because that's how we process it akin to audio db.
Regardless the point is KLD and whatever they came up with is not meaningful. And they did not publish comparisons on real benchmarks.
Just at a sniff test level, don't you think that if the quantization resulted in anything like 2M% error in a pretty typical context length, it would be plain as day? You'd do an A/B test and one of them would look like standard generated text and one of them would veer into incoherence? If not - what on would 2M% error even mean then?
> Regardless the point is KLD and whatever they came up with is not meaningful.
I'm not saying you're wrong, I'm just saying this isn't a meaningful metric either, mostly because it is using a different type of error (divergence along a trajectory) than what was actually measured (divergence at a fixed point) and so can't be used for this purpose. It could establish an upper bound but going by your work that upper bound is so high it may as well be infinite. That's somewhat concerning but doesn't necessarily suggest it performs badly in a typical case (which is how I'd interpret an expectation of 2M% error).
It doesn't really matter to this argument, if they are self correcting at all, then we can't assume all errors will permanently injure the trajectory. It's not like dead reckoning or a similar process where there is never an opportunity to reassess. It's more like a long division problem; it is possible to correct errors using in band information, without the external reference dead reckoning would require. (The incidence of false positive self correction does matter to the question of whether the model is actually of comparable quality after the quantization, of course.)
You can see bathtub rings on the underside as it's lifted up. The rings are caused by it's melting faster under the water than in the air, so there's a discontinuity at the waterline.
> If the agent itself concatenates a new URL together - with leaked data after a ? - you should block that from being fetched.
You're correct of course, I just want to note that the exfiltrated data could be in any part of the URL, so the absence of a query string doesn't indicate that no payload has been encoded into the URL. Arbitrary example, you can include credentials in a URL, so you could encode the exfiltrated data into a password.
