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.