The fact that they think OpenAI will honour them opting out of “share your data to train models” it’s surprisingly naive. I guess setting up private LLMs is going to be part of sysadmins tasks.
Absolutely. Lots of siloing is going on, both custom models/finetunes and lots of on-prem inference. Some things need frontier models, but most things don’t. Companies that aren’t just blindly bought in on the AI hype, and have real / meaningful use of LLMs, have already started doing this.
They say the model solved it but I’m fairly certain they have a team of mathematicians running and working on it for months. It’s not like the bot does it on its own with a single prompt or something.
It’s not really about who gets credit though. Whoever gives the proof first gets credit. It’s more about the fact that OpenAI is training on stuff they shouldn’t, and are probably targeting mathematicians who use their product specifically for data harvesting.
Basically, the fine print says they won’t use “your” data, but they use everything adjacent to it (LLMs reasoning and answers) so its essentially the same thing.