This seems to have a healthy helping of AI editing help (if not fully generated by AI). The links don't quite go to the sources that they should and there's a lot of AI-isms.

Anyways, the calculation for the costs seem crazy high (and are pulled from an ft article). In particular they are based off a calculation that assumes Sora videos take 10 min to generate (which seems simply wrong; I've personally generated Sora videos that take less than 10 min to return fully formed), fully saturate 4 H200s at once (this seems wrong with batching; I would assume they're batching a lot of tokens together per forward pass), and, crucially, that OpenAI is paying full spot, end-user pricing for an H200 (at $2 an hour). As an individual, I can rent an H200 for $2 an hour on e.g. vast.ai (and sometimes even cheaper than that!). There is absolutely no way OpenAI is spending anywhere near that number.

I also have no idea where the Appfigures $2.1 million comes from. As far as I can tell it doesn't exist at all in the linked website.

I don't really trust the numbers here.

> This "average nontechnical user" stuff, though, miss me with. For 2 decades people have been encouraging the "average nontechnical user" to do incredibly unsafe things on the premise that any kind of message encryption is the best alternative to sending plaintext messages. No: telling people not to send those kinds of messages at all, unless you're dead certain the channel they're using is safe, is the only responsible recommendation.

Eh. You misunderstand me. I don't really have too much of a view on this personally. Unless you specifically think that the term "average nontechnical user" is a bad term.

N.B. for other readers of this thread to flesh out my initial point:

Signal specifically didn't do that recommendation until they got sufficient critical mass of users in 2022. In particular Signal gracefully degraded to unencrypted SMS if the other side didn't have Signal.

Likewise Signal required phone numbers until 2024 when it shifted over to usernames, with all the security vulnerabilities that entails.

Signal has repeatedly made trade-offs that prioritize UX over absolute security even in 1-1 chat settings. That's not to criticize those trade-offs, there's a variety of reasons why they make sense or don't. But Signal has consistently demonstrated that it is not willing to make severe compromises to the UX and understandability in the name of absolute security and that it will balance the two.

I wouldn't characterize Signal as "absolute most possible safety" as you are implicitly doing here.

I would probably characterize Signal as "most possible safety for the average nontechnical user" which entails trade-offs against absolute safety for certain UX affordances (and project governance structures that allow for these decisions to be made), because if said affordances are not given, the average nontechnical user either simply won't use Signal or will accidentally end up making themselves even less secure.