Embed Notice

I like making these companies walk that tightrope. For example, LLMs. There are FOSS (de facto) LLMs available, and while I appreciate the merit in them being runnable locally, the quality of the output legitimately just does not compare to GPT-4 or Claude+. I also don’t go out of my way to trigger the janny modules of either of these tools. But if they janny me too much, I will 100% use one of those clunky FOSS LLMs, and have.

Whichever one gives me the best output with the least bullshit is the one I use. And this attitude legitimately does pull these companies in the direction of sanity.