Kinda weird GPT4-Chan wasn’t referenced. A guy fine-tuned GPT-J on 4chan, then deployed bots to write posts. I guess it was more of a stunt than academic or scientific, but training on 4chan improved the model’s performance on a truthfulness benchmark.
Kinda depressing that all of big-tech seems to have given up “innovating” (finding applications for publicly-funded research), and have become rent-seeking dinosaurs.
but it turns out all that cash was going toward a workforce of over 700 Indian engineers, rather than an AI.
I doubt much of that cash was going to their workforce. Should have though.
Haven’t used it yet, but venice.ai looks interesting; they have a good privacy policy. Right now, I just use ChatGPT with the “improve models” setting turned off, and use “temporary chat” mode. I don’t really trust OpenAI to be doing the right thing though. I’ve used 14B models locally, but they aren’t as good as 72B+ models.
Privacy policy changes are worrying. They also implemented Privacy Preserving Attribution, which sends anonymized data to advertisers, and enable it by default. I personally like LibreWolf.
I’d guess it’s mostly the AI autocomplete stuff. I.e. you keep on typing until the AI guesses it right then press tab to save keystrokes. LLMs are really bad at making test cases in my experience; they, ironically, can’t do the simple but nuanced computations needed to figure out what the output should be given the inputs, or to recognize and test the edge cases.
They can require ID verification of all users, as is done in some states in the US to access pornography sites.
Used Teams for a bit. Seemed fine, just used it like any other IRC clone. Didn’t use it for video. Windows has a lot of annoyances; death by a thousand cuts. The Windows ecosystem also sucks: to the point where graphic card and mouse driver installers try to install spyware.