Musing 109

In another entry in the chronicles of “the world moves fast”, I have seen an explosion of skepticism from even the “here’s my soul, AI” crowd that I didn’t expect to see.

Sure, the anti-AI crowd has been loud, and while I’m certain that societally, LLMs have been a net negative (and more neutral on its technical impact), I’ve still been appreciative of the technology for enabling me to get stuff done in a passable manner when handheld the right way.

And while it’s been interesting to watch what has increasingly felt like the ‘new normal’ of “release new model, nerf it, let people complain, release new model to keep them hooked”, I’m not sure how much longer LLMs can keep up the cycle, as on one hand, there’s little new data to train on, resulting in ever more synthetic data in the training data (and every bit of research done on that topic suggests this is bad) but possibly more immediately, the combination of resource limits and “need to make money” feels like its nearing the natural end. There are IPOs being sought, but with the majority of Americans being opposed to AI, I’m not sure those IPs will actually go anywhere.

The simple reality is that as the AI bills are going up without the corresponding gains, the big players will not be turning to them for the real work. The local models will though. And that means there will be a financial vacuum created.

What tries to take its place, only God knows.