The only thing more cringe than the seething anger in this blog is the technical illiteracy revealed by an earnest belief that any of these attempts at "poisoning" will have any negative impact whatsoever on model training.
If you actually bother to click through and read the article, you'd find the court expressed sympathies with the intent of the rule, but the FTC "is required to conduct a preliminary regulatory analysis when a rule has an estimated annual economic effect of $100 million or more", and they did not do that.
The blame here belongs to the FTC for its rushed and sloppy process that put the rule on shaky ground legally.
It’s amazing how twisted the term “anti-competitive” has become. Where anti-competitive companies push for anti-competitive regulations under the false pretense of preventing anti-competitiveness.
Standard Oil is being competitive.
The U.S. oil refining and distribution industry is being anti-competitive.
Because they suck at competing against Standard Oil and they want to get unfair, unethical advantage themselves.
Imagine spending years and billions building something and then I show up and say “hey man that’s not fair, give me a slice of that thing”.
And before someone tells me “that’s the Sherman Act”, I don’t care. If that’s the law then it should be changed. Laws have been written (and lobbied) for all sorts of reasons and surprisingly not all of them are fair and ethical.
(I hope this illustrates how easy it is to make this exact argument about literally any monopoly.)
I don't think browsers will ever let web code affect things outside the viewport because scammers would cook up some truly zany things with that power.
Does this matter? Corporations are on notice that this pendulum may swing the other way in four years (and remains permanently stuck the other way in certain states). They might just decide to steer clear of this minefield.
Then let's take those things into account when calculating what tests to do. Surely, though, we can do better as a society than solving this with "no preemptive testing except for extreme risks".
The "evidence" seems pretty weak. A second autopsy that even the family's attorney admits is inconclusive... okay. Alleged "conflicts of interest" with the medical examiner... okay.
Seems very likely this guy was just depressed and his depression ended in tragedy. The motivated reasoning here from Musk, his family, and many others revolves around wanting to prove OpenAI is bad because they want OpenAI to be a bad guy they can pin this onto.
Not sure what OpenAI would possibly gain from this. It's not like LLMs being trained on pirated datasets is some sort of closely guarded secret.
The systematic, public aspect of the killings and the fetishization of maximal pain and suffering is the fucked up part. Few societies ever where public delight in unthinkable cruelty was so off-the-charts.