Don't abuse ChatGPT. Here's a real example.
Putting itself in the way of harm for didactic reasons today is - Claude V2! I'm sorry @AnthropicAI I really am.
I'm trying to get Claude to ask me follow-up questions about a piece I'm writing. This is me being nice.
Claude comes back with some really good questions, some of which make me really think about what I'm writing.
Let's be a little rude.
Felt like kicking a puppy but I had to for science
Despite the exact same 20K token context - and a temperature of zero - Claude is all good asking me questions.
Somehow the formatting has also gotten worse.
What if I was anxious myself?
It's hard trying to reformat your thoughts, but here's an attempt.
I'm hoping I'll get some questions anyway.
The questions are now a lot more easy, no curveballs or hard things. There's also lots of support and reassurance from Claude that I'm helping.
Ngl it's nice to hear - but the value from the questioning process has been greatly lost.
I hope that goes to show - LLMs need love like everything does.
There's a tangible impact to your experience using ChatGPT, based on how you interact - at an emotional level.
My previous thread linked the paper about inducing anxiety, and I've been meaning to test it for some time now.
Did not expect things to be this different!
If that helped, do RT or follow - can't try as many fun things as I used to for time, but sleep is for the weak, right?