3 ms·
I don't find it hard to talk ChatGPT round (using conceptual arguments rather than keywords or prompt injection), but as implemented its 'safety features' come
by anigbrowl 4y ago
I don't find it hard to talk ChatGPT round (using conceptual arguments rather than keywords or prompt injection), but as implemented its 'safety features' come across as obnoxious and adversarial rather than cooperative, which reflects poorly on whoever is training it.
Interested in people's guesses on when/where to look for a Stable Diffusion-type LLM that can be run and tuned locally.