But LLMS do make those bots better in a lot of cases and I think younger people are more willing to use them. So maybe it'll work this time.

Well they tried doing that a decade ago but people hate those bots and don't use them. That's a big part of why they didn't get adopted more widely. They still need real people as a backup and customers usually seek them out even if it's more annoying. And they're willing to not do business over it.

IDK about that, most of the stuff that could be automated by LLMs can't be due to their unreliability. You still need someone to check the work in most cases. But things copywriting are being replaced. I'm just not sure it's giant chunks of the economy. Mostly it makes a lot of stuff more efficient.

But on the other hand it wasn't (and still isn't TBH) obvious that there aren't more exponential breakthroughs around the corner. Given what he thinks is at stake it is scary slow down even a little. Although focusing on optimizing arguably helps get there faster. He just got high on his own supply.

Just melt the brie in a round tupperware container to make it a disc again

It kind of seems like Altman was expecting to be able hit exponential growth by now. He had to know they were doing things in a massively inefficient way and it wouldn't be long until someone took advantage of it. Maybe he's just trying to get paid, but it really seems like he drank his own koolaid

Being downgraded from the future of everything to yet another host of some generic open source software is a huge blow, and it looks like that's what's happening

Yeah it looks pretty grim for that funding round lol. They basically have to start a cult or at least show some seriously impressive results very soon. Being able to run Chatgpt way cheaper should help them in theory, but they have a *lot* of investor funding to pay back.

The idea is that if Openai can be on the cutting edge they'll eventually make a machine god that can improve itself exponentially, and whoever does that will monopolize every industry forever. So no cost is too high

Oh for sure, it's in their interest to make things look as expensive as possible. Their strategy has been to just try and push things forward with no regard for cost and assume nobody will eat their lunch. Deepseek did do something pretty impressive but Openai wasn't really trying to do that thing

Yeah, but probably lose about $5b a year with how they currently operate. One year of runway isn't actually that much for a startup funding round, it's just a comically expensive startup. At least before Deepseek they were going to try to raise $200b this year lmao.
www.axios.com/2024/10/03/o...

It's called transfer learning. From what I've seen Deepseek did a bunch of other stuff to optimize the training costs but this was a big part of it:
en.wikipedia.org/wiki/Transfe...

Mostly power and hardware. It's not a secret, training and running Chatgpt is insanely resource intensive. A lot of what made Deepseek cheaper was them using Chatgpt output. Openai is mad that Deepseek did to them what Openai did to everyone else lmao

Using premade graphics speeds things up for sure. But having a coherent artistic vision is really hard. Getting everything to mesh together and look good while also creating the look you want for the game takes a lot of iteration, and I don't think there's really any shortcuts.

www.youtube.com/watch?v=lYfg...