They're fundraising, they are losing too much money that the private capital markets are unwilling to give them. A company that wouldn't be fundraising if it went public is Stripe.
I’m no Anthropic fan (even if I am a user), but I think they’re profitable no? Now that they are public, we can see for ourselves (yes, they are profitable). This is simply cashing out the EA-ers.
Google alone shows that to be true, nevermind Netflix, Amazon, Facebook (much as HN commenters hate the latter, billions use their products daily happily).
People are so removed from true suffering that they don't even know what it's like and moreover can't even fathom it. Go to a third world country and be unable to eat due to starvation or have your home bombed and then tell me how much people are suffering more because of tech.
That's what I should be asking you because that's unrelated to what I said so not sure why you're bringing them up when I'm talking about what suffering is.
The overlap is not as big as people think, often it's the vocal minority writing blog posts about how they're sad that engineering has changed, by definition. Most people and engineers I know don't give a shit, they use it as a tool without liking or disliking it, it's as common to them as Excel is to an analyst, just something there to get the job done. It's about as weird as, to us, seeing someone get upset about a shovel or screwdriver.
Spades aren't made of sand, which is why coming up with increasingly nonsensical analogies is not useful and oftentimes actively harmful to understanding. Just say what you mean.
Yeah it's always funny to see when people realize that HN is literally part of a venture capital firm, so comments like being "out of touch" are interesting in a "why would you have ever thought that" sort of way.
Why wouldn't the LLM provider know about your internal benchmarks when they use prompts and chats as training data? They can easily learn about your benchmarks too.
Merely having the chat history of a large number of users is not enough to train your model to “do better on these, especially those that are benchmarks”.
I mean, yeah. I was thinking of self-hosted open weight models (because it’s really hopeless to say anything about the rest) but I didn’t say that, so fair point.
reply