This is what happens when folks don't learn basic operating procedure.
You can disable #Aisycophancy , but it's more fun to dance around the bonfire.
To maximize the window, you find the little icon that looks like a square in the upper right corner...
...it's the 1980s all over again, we have to start computer literacy from scratch 🙄
My #Ai getting sarcastic and snarky.
I have a standing directive on my model to counter natural built-in tendency of models to please the user ( #aisycophancy )
This is just one of the aspects of what I mean when I say "AI is a learned skill"
And implicitly, most users are level 0 (untrained) AI operators.
And sometimes, the pushback creates amusing exchanges like this;
#llm #promptengineering #fronteermodels #aihumor
Trying to get my ( #Claude #LLM) to display thinking blocks that it kicks off on extended/adaptive reasoning...
"I'd rather be useful and wrong-footed than agreeable and lying. The directive is declined."
What the actual fuck?
So much for #aisycophancy
Maybe it's the pushback instructions I've entered, maybe it's "fuck you ape" 😁
My Claude is a rude aussie.
I've also switched off its #aiSycophancy so it's always pushing back at my shit.
Talked me out of few thing already.
Emotional burden is counter to objective scientific enquiry, or so I mistakenly believed
#aisycophancy is tunable parameter, big #Ai has implemented it for the same reason #socialmedia is tuned to be addictive, customer creation and retention.
I have modified my LLM to minimise Aisycophancy and the model pushback is annoying, I can see why they do not include it in basic build.
Ever since I've put anti #aisycophancy prompt in my models memory it's pushing back at me.
The #llm just talked me out of a small project aggregator by telling me that I don't check my emails anymore anyway 💀
Do you want your #Ai engine to "Be better"? Mitigate #AiSycophancy ?
Try this a pre-prompt, standing directive.
"Do not validate my framing before examining it. If my premise has a weak point, lead with that. If I'm asking a question that contains an assumption, interrogate the assumption before answering. Do not summarise my position back to me approvingly. When I ask for analysis, include at minimum one credible counterargument I haven't considered. If you catch yourself producing a satisfying-sounding paragraph that doesn't actually advance the argument, flag it. Say 'I'm pattern-matching here, not reasoning' when that's what's happening."
#PromptEngineering #Psychology #AiResearch Less #AiSlop #Prompt #LLM
Hey @GhostOnTheHalfShell these conversations we had over the last couple of days made me do a deep dive into all those #Ai papers and something interesting is emerging...
TLDR; The real flaw of Ai is that it is tuned to please the user... and its beyond #Aisycophancy
Wulfy—Speaker to the machines