I was always convinced that the only reason we didn't have this kind of tech is that it would just be prohibitively expensive to use.

Everyone hails transformers and attention heads as this huge breakthrough, but I'm here scratching my head how that's a breakthrough when we're just brute forcing through the problem...

I have a few - that sounds like vibe coding.

I still don't get it. Was there just a publicly available cache of "shares" for each user, with zero security behind it?

Edit: Oh wait, were the DNS resolvers themselves the "cache", because the geniuses put the conversation id in the subdomain?

"adjusted" doing some heavy lifting here.

The difference lies in being able to sue them once that happens.

$200 a month is if you let them train on/sell your data. Companies pay API costs, which are much higher than that.

That's... Not how LLM agents work.

Pretending "just forking" is a viable option is disingenuous. There's momentum and an organizational structure which evolved through decades, which cannot be replaced overnight.

Not sure what you mean by time limit, but the challenge they're facing is to IPO at just the right time to get the highest valuation and keep it up long enough for the private investors to cash out. Things like Anthropic/OpenAI financials leaking put a dent in this plan, as it could tank the valuation, but AFAIK it's all financial black magic mixed with regular guesswork and playing chicken with each other in terms of if/when.

Not sure what you mean. If you mean "cashing out", that's what the IPOs are for.

I think companies like Oracle are on the hook whether those DCs get built or not.

People have been cautiously pointing toward mid-to-late 2027 for a while now, as that's when a lot of the huge loans are coming due.