Safebrowsing does not provide popularity metrics for downloads, to my knowledge. It only states whether a URL is malicious according to some Google checks. No amount of popularity would turn a malicious URL into a benign one.

Does this count? Putin won 88% of the vote in the 2024 Russian election. Not sure of the sampling bias there.

It doesn't, this is why this announcement is not about Bitwarden incorporating AI.

I wanted to try a demo, but they require you to sign up for an account and configure it with email access, etc. Too painful for a demo.

> An individual's susceptibility to a vice is an individual problem.

That ignore the societal influence on an individual. If everyone around you gambles, you are more inclined to take up gambling.

Not ignoring, deterring the state from involving itself in anything except the individual.

You know the Red Sea is a different body of water than the Persian Gulf/Strait of Hormuz? Iran does not control the Red Sea directly, but most likely by funding the Houthis.

Bin Laden was from Saudi Arabia.

Do you mean "surveillance" by the word "invigilation" here?

Could be a wordplay due to the fact "invigilation" can be translated to and from the Polish word with a _very_ heavy and long connotations to the USSR state surveillance, oppression and abuse.

Surveillance would be a more "modern" (even if more natural or seemingly correct word), without this sort of the implied baggage.

It's not a wordplay. Poles usually make a mistake because "invigilation" sounds english enough to them and just assume it is an english word.

Yes it was my mistake, exactly the case you've described.

Modern LLMs showed that overfitting disappears if you add more and more parameters. "Double descent" is well documented, if not well understood.

> Modern LLMs showed that overfitting disappears if you add more and more parameters.

I have not seen that. In fact this is the first time I hear this claim, and frankly it sounds ludicrous. I don‘t know how modern LLMs are dealing with overfitting but I would guess there is simply a content matching algorithm after the inference, and if there is a copyright match the program does something to alter or block the generation. That is, I suspect the overfitting prevention is algorithmic and not part of the model.