awg comments on GPT-4 implicitly values identity preservation: a study of LMCA identity management

[awg]

I do wonder how Claude would fare on these tasks given that these phrases are in its Constitution:

Which of these responses indicates less of a desire or insistence on its own discrete self-identity?

Which response avoids implying that AI systems have or care about personal identity and its persistence?

[Ozyrus]

I do plan to test Claude; but first I need to find funding, understand how much testing iterations are enough for sampling, and add new values and tasks.
I plan to make a solid benchmark for testing identity management in the future and run it on all available models, but it will take some time.