Raphael Fetzer
René FouquetI'm reading a lot of extreme LLM takes in my timeline these days, from both AI believers, who see the technology as more important than the Industrial Revolution and the invention of the wheel, and from total AI rejectors, who believe every LLM is a collection of copyright infringement, that the technology is actually completely useless, and that LLMs are (figuratively, although from some people it sounds like literally) melting your brain.
I guess I'm somewhere in the middle. I see the risks and inherent problems. Yes, the training data is problematic in almost all cases, the AI company boom raises hardware prices by a ridiculous degree, the energy costs are staggering, and many people are in fact using it thoughtlessly and often dangerously. And yet, it does have legitimate and useful applications. Nobody who wants to be taken seriously can argue that LLMs are useless; plenty of real world examples prove their worth every day.
Furthermore, a total rejection of the technology will soon be hard or impossible for most people, because management is often pushing for its use. If you are in a position where you can decide if you want to use LLMs, go ahead and do what you feel is right. But please realize that this is a privileged position that many people will not be in. I'd argue it's more sensible to try to shape the AI processes at work instead of rejecting them outright and hoping the bubble will burst.