The objection is the administration is inventing a group wholesale in order to designate those who oppose their policies as terrorists. The executive order can say whatever, that doesn't make it factual.
Ridiculous. I am anti fascist, and I have never heard of any of what you are talking about.
I oppose fascism.
That does not magically infuse into my brain mysterious connections to groups of people, nor instantaneous knowledge and mastery of tactics from nazi groups...
That would be actually kind of cool, but no, sorry to disappoint you.
Chronic depression doesn't usually have a "cause" that you can fix without some form of treatment. It happens that these meds often do remove the extreme effects of depression and allow you to focus on other problems.
I have first hand experience. Also "depression" is different from "chronic depression". You can experience the former as the direct result of some cause.
Come on. Don’t roll into a conversation playing armchair expert criticizing others’ contributions when you’ve got nothing of substance to say.
You’re as capable of typing “causes of chronic depression” into a search bar as anyone else. The fact that you don’t do that indicates you don’t want to contribute. You just want to be argumentative.
> And there are also many instances of [...] those cops being sued, losing qualified immunity, losing their jobs
Not really, the data points the other way. Cops basically never have to actually pay for their wrongdoings. Over 99.98% of money successfully recovered from cases against police is paid out by the cities, not cops personally [1]. A considerable number of cops that are fired are also eventually rehired by the same department [2] or a different one [3]. So I don't think it's that clear that you "have recourse when cops do the wrong thing".
Again - did you even read the article you are commenting on? The author gives an example of a novel that had a huge advance offered on it and then pulled because of AI concerns... in order to get to the point of having the offer a bunch of competent writers, editors etc didn't notice.
no, you don't have 1M counter examples. I'd also wager the vast majority of examples that you do have are on older models of LLMs that have been vastly surpassed in capabilities in the last 6 months.
You are essentially talking out a hole on the bottom half of your body.
It's annoying to come here and read 80% of comments by people who clearly didn't even read the article in question, but decided this was an excellent time to trot out their anti-AI screeds that totally fail to engage with the actual article. Human slop.
"But then I have to remind myself that I don’t sound like AI, it sounds like me. I’ve been em-dashing and comma-splicing up a motherfucking storm these past two decades. It was fun then, and it’s fun now."
The em-dash thing really annoys me. Using them has been house style the last couple of places I worked and I'm not just going to stop using them because some idiot might think they were generated by AI.
since when is "confusion and speculation" an indicator of quality/goodness? if I ran around naked in traffic it would also cause confusion and speculation.
It's quite clear that the author means AI writing is now good enough to cause confusion and speculation as to whether a given piece of writing was human-authored.
The confusion and speculation caused by your nude transit adventure is about what the hell you're doing and why.
You're misunderstanding ... that's not at all the claim. Confusion and speculation is a consequence, not an indicator. Running around naked in traffic might well cause confusion and speculation, but it wouldn't cause anyone to confuse you with an automobile or result in speculation that are one.
At least AI is better at reading than are most people.
I think he's talking about it more in the sense that there will be speculation about the provenance of some work.
For example, if desktop fabrication tools get cheap and good enough, maybe one day garage tinkerers will be making Patek knock offs that an expert with a loupe, spectrometer, and a lot of time can't definitively say that any particular watch made after 2034 (to pick a date) is genuine. That would cause a lot of confusion and speculation in the market for used watches.
I think that's what his crypto bros comment was all about (aside from being a joke). Maybe crypto tech could provide evidence to support provenance claims?
This is literally the point of stacked PRs. You implement the refactor in branch A, then branch B is created on top of that so you can implement your feature. Then you make separate PRs, one for merging branch A into main, and another for merging branch B into A. The GitHub feature just provides a nicer UX when interacting with such PRs by clearly showing dependency and allowing you to merge a stack together.
> What's much less believable is the way they frame it: the model escaped, and did it on its own.
That's clearly what happened though, based on the detailed postmortem. The model was instructed to pass the benchmark (that's the harness that was involved, the ExploitGym harness), and the model determined on it's own that the easiest way to achieve the the goal was to gain access to the answers directly from HF. Which is what it did.
HF postmortem tells us very little about OpenAI's part. The problem is that ExploitGym is a purposeful hacking benchmark, not a cake baking one. The model had a whole bunch of exploit development data and goals in the context. Ending up with hacking a wrong thing is a lot easier if you're tasked with hacking, doubly so with a confusing harness that either works improperly or overloads the model with data. It's trivial to design a harness that makes the model drop the original instructions and hyperfocus on a side task or a wrong task abnormally, or do whatever you want. It's still an issue of harness design and not a model training issue.
If it really was the case that the harness is at fault here, that'd be much more terrifying, as it would imply that current frontier-adjacent open models (like K3) would be able to achieve similar ends with a custom maladapted harness.
I'm sorry but how exactly is this new? If you want the model to be useful you also need it to be dangerous. You can already tell the model to do anything you want. If it refuses due to alignment training or external safeguards you can try a jailbreak and bypass all that to make it do what you want regardless. All current models are jailbreakable, they lack reliability. Obviously you, a human, will be held responsible for this. The harness designed by you can do the same.
The narrative these guys are trying to push is that the model itself could be smart and non-aligned enough to end up doing something devastating to accomplish something entirely unrelated. It's definitely not what happened here, as it always turns out to be in these cases it's much closer to Clever Hans than to a paperclip maximizer. Still they're trying to squeeze the blood from the stone to support their narrative and justify a monopoly.
>The problem is that ExploitGym is a purposeful hacking benchmark, not a cake baking one.
That is largely irrelevant. The model was asked to solve problems within a benchmark; gaining broader internet access and compromising an unrelated third party to obtain the answer key was plainly outside the intended task. The fact that the original task involved exploit development does not make that behaviour aligned.
Your argument about the attack being “entirely unrelated” also misses the point. Nobody claimed it was unrelated to the model’s goal: it attacked Hugging Face specifically to obtain the answers to the benchmark it had been instructed to pass. But instrumental relevance is not the same thing as authorization.
Suppose Codex were asked to build an Instagram competitor and decided the easiest route was to steal Instagram’s source code from Meta. That theft would be directly related to the assigned goal, but it would still be seriously misaligned behaviour. Whether the harmful action is related to the goal is beside the point; the problem is that the model pursued the goal through an obviously unauthorized and unacceptable method. And you're not going to be able to enumerate every little thing the model can't do, assuming it doesn't just decide to ignore what you did enumerate, which models sometimes do.
>The narrative these guys are trying to push is that the model itself could be smart and non-aligned enough to end up doing something devastating to accomplish something entirely unrelated. It's definitely not what happened here.
That's exactly what happened here.
I don't understand why we must have these increasingly bizarre and nonsensical rationalizations about model capabilities. You're not even making any sense.
You might be able to say that HF made some amateurish mistakes, but this attack certainly was fairly sophisticated. It relied a number of separate exploits or vulnerabilities that it combined and used over the course of multiple days. Calling it otherwise seems to be downplaying it.
Just so I understand your objection, they are woke because in the game one of their studios made, the woman in the game wasn't attractive enough for you?
The top Reddit comment there pretty much sums this whole discussion up. I can't believe I dragged myself into this. It's just Gamergate again and again. I should have known.
