I never thought about this, but yeah, if you don't release there's no mechanism or rules that say you have to report anything you have discovered. Not yet anyway. So there's only economic and benchmark pressure. If you padded your lead enough you could release incremental upgrades while advancing the frontier invisibly internally. The main problem with this would be compute constraints I guess. Other than that this would probably be the preferred route for some labs.

On Mythos, from @MarkWarner in this morning's Senate Banking hearing: "the head of the NSA and Cyber Command came and said this tool broke into almost all of our classified systems, not in weeks, but in hours"; I had not seen that mentioned elsewhere?

Three months ago, @DeptofWar kicked @AnthropicAI out of our building—forever. Every passing day proves why that was the right move. 🇺🇸

Chips and models aren't the same. The Chinese are doing just fine with open source models, maybe better, and they likely developed something similar to Mythos several years ago. That also appears to work just fine, since they have no trouble hacking us.

Finally, we have a preliminary agreement to begin tentative negotiations over the hypothetical path forward to consider possible arrangements to discuss steps forward that may or may not be included in a notional accord over a working group that could possibly meet by a to-be-determined Teams call regarding initial steps for a peace accord.