A genuine conversation about AI!

Actually, this is what I've been wondering since ChatGPT changed whatever they did recently - my interaction experience radically changed, not necessarily worse, not really at all less functional, but absolutely different now and I'm unsure how I feel about it.

I agree with redoing rules every update - and absolutely do rules, extremely clearly that you want legitimate critique and feedback, not sycophancy -> always test that functionality.

It does get tricky when they decide to interpret the same rules differently seemingly at random.

Also frustrating dealing with changes like what has happened with Copilot - the internal rules change so frequently and are so narrowly defined, it's almost pointless to even define your own rules for that platform.

Sometimes tho, I found it worked better to start something in Copilot and then bring it elsewhere after a point. I do a few things like that between AI

Running off rails never worked out. I personalized differently for different LLMs. To truly get LLMs to suit your personal style takes time and often have to redo with every new major version release. Gist of common instructions are to make it ask clarifying questions, TLDR of every response at the top, To be direct and Rewards for showing true confidence scores for every response (this becomes critical in long conversations).