> To be clear, we have not yet determined whether Claude Opus 4 has definitively passed the Capabilities Threshold that requires ASL-3 protections. Rather, due to continued improvements in CBRN-related knowledge and capabilities, we have determined that clearly ruling out ASL-3 risks is not possible for Claude Opus 4 in the way it was for every previous model, and more detailed study is required to conclusively assess the model’s level of risk. (We have ruled out that Claude Opus 4 needs the ASL-4 Standard, as required by our RSP, and, similarly, we have ruled out that Claude Sonnet 4 needs the ASL-3 Standard.)
If Claude’s Constitutional AI research is legit, what prevents China to pull a few strings through SAE and nudge the model to be more aligned with China’s political economy ideal? It’s not like the web doesn’t have enough left-leaning training dataset.
Growing up in China (and left 15 years ago), I don’t feel the “Grand Strategy” kind of person tends to survive in the harsh dog-eat-dog hand combat nature of political life in China.
Lastly, there was a popular lie in China around 1840/First Opium War: the British empire eats so much meat that they need our tea. Surely they can’t attack us if they don’t want to get digestive problems.
I think it also strengthens the neural pathway so that <speculation>when the next time you face the many options, the weight would be just slightly higher</>.
(I am assuming human brain works similar to how neural net works. I can be wrong here. )
Yes. There are families on the fringe that tries to make it work. But it’s an exception. Not a norm.