I work in AI.

We've known this about LLM's for many years. One of the reasons they weren't widely used was due to hallucinations, where they'll be coerced into saying something confidently incorrect. OpenAI created a great set of tools that showed true utility for LLM's, and people were able to largely accept that even if it's wrong, it's good for basic tasks like writing a doc outline or filling in boilerplate in scripts.

Sadly, grifters have decided that LLM's were the future, and they've put them into applications where they have no more benefit than other, compositional models. While they're great at orchestration, they're just not suited to search, answering broad questions with limited knowledge, or voice-based search - all areas they'll be launched in. This doesn't even scratch the surface of a LLM being used for critical subjects that require knowledge of health or the law, because those companies that decided that AI will build software for them, or run HR departments are going to be totally fucked when a big mistake happens.

It's an arms race that no one wants, and one that arguably hasn't created anything worthwhile yet, outside of a wildly expensive tool that will save you some time. What's even sadder is that I bet you could go to any of these big tech companies and ask IC's if this is a good use of their time and they'll say no. Tens of thousands of jobs were lost, and many worthwhile projects were scrapped so some billionaire cunts could enter an AI pissing contest.