You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Thanks for replying! I had a look and, if I'm being honest, I found this quite lacking. In the spirit of the project, I had a conversation with Claude-3.5-Sonnet about it: https://poe.com/s/tJtRyGL3KitmecrCle7Y
My main concerns are that the project is currently easy to misuse by bad actors (cf. ChaosGPT) as well as carries significant risk of uncontrolled proliferation (i.e. there's no kill switch). The latter might even be in violation of California bill SB-1047 if it gets passed, though I'm not sure whether it would meet the criteria?
I'm thinking that at least the following recommendations are useful:
develop, or have it develop, a Responsible Scaling Policy (i.e. increasing the bar for safety as the performance increases)
evaluate people before granting access to the full code
include safety benchmarks in the performance evaluation
collaborate with Safety evaluation organisations like Apollo, Haize, and METR
I can recommend the full conversation I had with Claude
I'm concerned about safety & alignment issues. Do you have a safety policy?
The text was updated successfully, but these errors were encountered: