You could've said the exact same thing about the internet in the early 2000s when the bubble was at its peak and you had a bunch of companies trying to make products that made no sense. But when the hype died down, and the bubble popped, we got a lot of very useful tech out of it. The situation with LLMs is exactly the same.

While LLMs are stochastic in nature, that doesn't in any way make them useless. There are plenty of scenarios where they work extremely well. For example, just last week, I wanted to figure out how to decode RAW files from my camera. I have a Nikon, and it uses NEF format which is proprietary and has no open source decoder right now. I threw an LLM at decompiling a binary and tracing it in memory as it was doing the decoding. After a few days, the LLM managed to write the code that decodes the images. This is absolutely not something I would've been able to do on my own. And the fact that there aren't any open source drivers yet, shows that it's a very difficult task to accomplish. That's just one real world example.