Good luck! Not being startupy or American I don't understand. But sounds like a schlep problem (see pg essays).
If you ever want to pivot an idea I am suprised no one does is why don't long term bets e.g. 2028 president pay interest. When you bet on something almost certain in 5 years you always lose due to lost interest. Maybe bets can include interest or even be chucked in SP500 for duration.
Sans contract? Probably like if I take a photo of you holding a copy of a recent book. I own copy right of the photo. The author still has copyright of the book.
Welcome to a big company where pretty much everyone has been working full steam for years, in order to take advantage of having a job at a company during a once-in-a-lifetime moment.
If I were to download those weights I can't run them unless I spend $100k on a cluster, so the privacy advantage is not there yet.
We already have Groq, Celebras and AWS Bedrock and others in the inference of open models space, so the model is usable that way.
Is Claude better than Llama, Gwen etc. Probably. For now.
But for how long? Dissolving means relying on Meta or Deepseek etc. to pick up and carry on tuning. Otherwise it'll be as useful as a GPT2 or Atari ST eventually in a competitive environment.
Also open sourcing the weights is handing it over to DoD (aka DoW).
Complicated question but probably not the best move. Keep going means keep working on safety research.
I prefer the claude code CLI interface for everything anyway. It is actually more convenient. Memory is local files, type one word to use rather than navigate.
reply