Join Nostr
2026-04-27 11:17:25 UTC

Andrew Anglin on Nostr: lol Yeah, I'm getting a lot of similar feedback on this. People are sending a lot of ...

lol

Yeah, I'm getting a lot of similar feedback on this. People are sending a lot of data on successes of AI job replacement.

Clearly, the maximalist claims were wrong, AGI is a hoax. And there is an unsustainable amount of money in the industry that is not making a return, pushing the “growth company” model beyond anything we’ve ever seen before.

Further, I will continue to maintain that Chinese AI is much better and is actually driving development now, rather than the reverse. Qwen and Deepseek/Huawei (Qwen also integrating Huawei) are the defining AI models already, and the question now is the effectiveness with which US companies will be able to rip them off, particularly with relation to efficiency.

The infinite money glitch between Nvidia and OpenAI/Oracle is unsustainable unless the compute becomes significantly more efficient.

However, having spent a day reviewing this, it seems my premise was slightly outdated, especially in relation to code. The “firing and rehiring because the AI can’t actually do the jobs” is waning, and this latest round of firings appears to be legitimately representative of large scale AI job replacement operation. (Thankfully, most of these layoffs are Indians thus far, who will lose their H1-B status.)

So I concede an informal and non-binding partial correction, which I will blame on having been largely out of it in recent months, due to irresponsible alcohol consumption levels and then PAWS.
It’s gotten extremely good in the last six months, so much so that it feels completely worthless to even attempt to code without it. The criticisms people had previously were valid, lots of hallucinations and slop code, but it’s much better recently.

You also have to take into consideration that the primary goal of the AI labs is a genius programmer in a box. The amount of effort put into making it good at coding is Herculean. It makes sense strategically. If they make a machine god coder then they can just ask it nicely to automate everything else automatically.

You can’t really give a good judgement of how good it is if you haven’t used it to program in the last six months. As far as scientists you follow, Jack Dorsey recently laid off a ton of people because they don’t need human coders anymore. Lots of companies forbid manually coding (definitely not all of them.)

The problem is the cost. The $20 a month subscription costs OpenAI $2000 if you max it out each month. The psyop will come from the other side, if anything, they will psyop people into thinking it sucks and only let fag satanists use it. I expect them to use the uber model and start charging more.

I just enjoy telling you that you are wrong about this because you are generally right about everything else.