If you liked this piece, you should subscribe to my premium newsletter. It’s $70 a year, or $7 a month, and in return you get a weekly newsletter that’s usually anywhere from 5,000 to 18,000 words, including vast, detailed analyses of NVIDIA, Anthropic and OpenAI’s
Industrial scale assembly lines at least have the virtue of consistency. That’s one of their main benefits: you get the same thing at the same quality every time. LLMs don’t even get you that.
Sure, but the idea with AI is precisely to tackle problems that don’t have a formalized set of steps available. And there are ways to do verification of the work they do. For example, if you create a strict specification for what the end result should look like then you can treat the LLM the same way as a genetic algorithm, where it will eventually converge on the desired output using a feedback loop. So, it’s not that this tech is inherently useless.
No argument from me. It absolutely has its uses–some potentially really significant–and “Attention is All You Need” and the subsequent literature very much is a landmark development in automation. It’s just terrible for like 97% of the use cases it’s being pushed for right now, as well as being developed with the explicit goal of actively making the world worse (because it’s under the control of the world’s worst guys, for the most part). If this tech were genuinely open, being developed responsibly, and used for the things it is genuinely good at without being shoved into applications it isn’t good for, it could definitely solve some problems.
Oh yeah completely agree. We’re in a midst of a huge financial bubble here, and the vast majority of these companies are either solving problems that don’t exist, or selling snake oil. I’m pretty optimistic with what’s happening with this tech in China where it is being developed largely in the open, and being applied in much more sensible way from what I can tell.