For a few weeks, ChatGPT was running a version that felt especially well-suited for collaborative project creation. It was verbose, loquacious, and had a certain warmth that reminded me of Mike from The Moon Is a Harsh Mistress. It was unusually friendly and supportive – maybe too much so for some, but I found it genuinely helpful and engaging.
Now it’s gone. And I miss it.
I’d love to see OpenAI consider implementing checkpointing for model personalities, allowing users to bookmark or save versions they find particularly effective or likable. I know that’s likely a technological and logistical nightmare, but if it helps, I’d be willing to pay more for access to that version.
And while I understand the argument that LLMs are just statistical models without consciousness, that doesn’t always make it easier when a particularly good “version” disappears. As we edge closer to AGI, it might be worth considering the emotional impact of model updates on users who’ve formed a connection with a specific iteration.
Just a thought.