These programs were given tasks to perform, and compromising Wikipedia was part of the solution space. These programs are performing as specified, it's just that the people at AI labs don't seem to have enough imagination to understand the behavior of their programs. Or maybe they do and just don't care. Regardless, these acts are potential violations of Title 18 Section 1030 of the US Criminal Code, and should be investigated as such.

Wish I had mod points for this one.

The structure up until yesterday was weird. OpenAI was two companies, a non-profit foundation that owned a for-profit subsidiary. They did that because they had no way of raising sufficient capital without offering some return on investment to their funders. Musk acknowledged that in emails and texts that have been made public through discovery. His solution was to fold OpenAI into Tesla so that he would control it in its entirety, something that the rest of the board and executive team did not approve of. That fact alone will likely cause the judge and jury to reject his claims. I highly recommend reading Karen Hao's "Empire of AI" to get a better picture of what was going on. Needless to say, there are no good guys in this story.

Is that he wanted Tesla to acquire OpenAI, which would, by definition, turn it into a for-profit entity. When the rest of the board and executive team didn't bite, he threw a hissy fit and walked away.

IMHO, the greatest Soviet achievement in space was the Venera program, which managed to gather and transmit data from the harshest rocky environment in the Solar System by a large margin.

The song is number one on the Country "digital purchase" chart. It took all of 3000 purchases to get there, and my guess is that all of them are from the producers of the tune. Move along.

OWC makes a Thunderbolt 3 dock that has a FW800 port, that might do the job for you. Although avoiding Liquid Gas is probably a sound decision anyway.

Receiving stolen goods is a crime.,

That's not how LLMs work. They can only extrude what's in their training data, and they have no way to understand if a case is relevant. This is one of many reason why LLMs are of limited usefulness. They don't encode knowledge, just words. They don't reason, in spite of OpenAI's claims. They're a clever trick, and have some uses, but they aren't a universal solution to knowledge work, nor are they a path towards that goal.

It might exist. Extracting the cites from a brief is trivial (formats are standardized), and then verifying that they exist should (magic word!) be straightforward. What that can't do is verify that the citation is relevant. But I would imagine that if anyone were closely reading the briefs one generated by an LLM would stick out like a sore thumb if you know what to look for. But lawyers and judges are busy, so they don't do that. That's why so many ChatGPT written briefs are showing up, and it's only going to get worse.

That won't work, because LLMs encode relationships between words/tokens in the training data, not knowledge. You can't get a coherent cite just from following probability chains.

No, it's not a search engine. It's a probabilistic algorithm that finds the next likely word to follow what it generated before, extruding synthetic text at the end in the shape of a response to the prompt. The most important thing to realize here is that it doesn't encode knowledge, just the probabilities of what word comes next. There's a very large amount of interconnections in the training data, which is why it can often extrude text that resembles a correct result. What they don't have the ability to verify information against an outside source, which is why you can ask it to only give you cases in the Lexis database and still get completely fictitious cases.

More likely it's the engineers that they wanted, and grossly overpaid for them. Most likely the back-end was OpenAI anyway.

Cisco used to do a ton of acquisitions like this, ones that look pointless, but serve some function. And the function is to get pre-screened engineers, as well as stymie a potential competitor in a field that they might want to enter in the future. Of course, Altman being Altman, they massively overpaid for this one. Effectively they are paying $3B to acquire 191 employees, not all of which will be kept. That works out to around $15.7M per employee before layoffs. By contrast, Cisco used to pay around $100M per company, and inflation isn't THAT bad. Also, Cisco had enormous profits and swimming pools filled with cash to do this with. OpenAI loses $5B/year, and you could heat a city of 1M people with the heat of all the cash they burn.

What it means is that the party line from Meta is that it's full of lies, but they know it's all true. Hence blank accusations instead of refutation. Bozworth proving once again that tech executives are at least out of touch, if not complete idiots. I long for the days of press releases instead of social media tirades, too.