Opus 5 kept refusing to reuse the nonce for the thumbnail of a NIP-17 file. I had to argue with it doing security theater. Finally I told it to read the spec and it conceded that interoperability was important in the "insecure" spec.

nostr:nprofile1qy2hwumn8ghj7un9d3shjtnyd968gmewwp6kyqpqq3sle0kvfsehgsuexttt3ugjd8xdklxfwwkh559wxckmzddywnwss3qr25 So stop using proprietary AI models and use #elevenfreedom models instead

I still use opus 4.6, you can still use old models try "/model claude-opus-4-6[1m]" it will get you to any older model you want

I’m still amazed how good LLM’s have become since gpt 3 till now. If this pace continues 🤯

Are they getting misanthropic?

Sounds about right.

I fed my intercom's gate & AGC versions from ChatGPT & Gemini 3.1 Pro extended to Sonnet 5 max effort, because I wasn't fully happy with either of their best efforts after multiple iterations - it think it finally solved my DSP issues for the rest of v0.1.x (& up until I may implement adaptive, like...