- Labor
- There are a lot of improvements in the making. Agents, memory, self-improvement. It’s a young technology.
- Currently AI is not good enough to replace people. It’s good enough to improve productivity and will probably get better at that. This will be the reason why many people loose their jobs - there might be human level AI in the future but that’s hard to predict.
- Build more power plants. This is already happening. A problem but not a impossible one.
- That is just plain wrong. Try different models and get different results. What the future will bring is hard to predict but artificial data or self improving models might be the solution to the data problem. Time will tell.
The answer to question 1, to me, seems to be that it is promising to replace workforce acrossamy fields. This idea makes investors and Capitalists (capital C, the people who have and want to keep the money) drool. AI promises to do jobs without needing a paycheck.
I’m not saying I believe it will deliver. I’m saying it is being promised, or at least implied. Therefore, I agree, there’s a lot of grift happening. Just like crypto and NFTs, grift grift grift.
Yeah, that seems to be the end goal, but Goldman Sachs themselves tried using AI for a task and found it cost six times as much as paying a human.
That cost isn’t going down–by any measure, AI is just getting more expensive as our only strategy for improving it seems to be “throw more compute, data, and electricity at the problem.” Even new AI chips are promising increased performance but with more power draw, and everybody developing AI models seem to be taking the stance of just maximizing performance and damn everything else. So even if AI somehow delivers on its promises and replaces every white collar job, it’s not going to save any actual money for corporations.
Granted, companies may be willing to eat that cost at least temporarily in order to suppress labor costs (read: pay us less), but it’s not a long term solution
As usual a critic of novel tech gets some things right and some things wrong, but overall not bad. Trying to build a critic of LLMs where your understanding is based on a cartoon representation skipping the technical details about what is novel about the approach and only judging based on how commercial products are using it can be an overly narrow lens to what it can be, but isn’t too far off from what it is.
I suspect LLMs or something like them will be a part of something approaching AGI, and the good part is once the tech exists you don’t have to reinvent it and can test it’s boundaries and how it would integrate with other systems, but if that is 1%, 5%, or 80% of an overall solution is unknown.
Also at the same time, and for many of the same reasons, fuck Goldman Sachs.
