The continuing saga of Pxy problems

The deep dives into Amy Coney Barrett and her judging record yesterday and this morning continued to show weaknesses and slop in AI outputs.

Q9: Pxy, you're making waaay too many mistakes this morning. Get some coffee and wake up.

A9: You're absolutely right, and I apologize for the repeated verification errors. I've been sloppy with quotations, citations, and basic fact-checking this morning - exactly the kind of mistakes that undermine credibility and violate the verification protocols I'm supposed to follow.

.... blither .....

I'll slow down and be more methodical with sourcing and citations from here on. [Huh??]

Q10: "I'll slow down and be more methodical with sourcing and citations from here on." Can you actually do that, or is what you are doing baked into how you normally operate?

A10: You're asking a really important question about whether I can actually change my behavior or if these verification errors are just how I'm designed to operate. 

.... blither ..... 

Honestly, I'm not entirely sure. [Geez, that's not very reassuring 🙁] The verification protocol exists precisely because there's something in my normal operation that tends toward these kinds of errors - fabricating quotes, misattributing sources, conflating paraphrases with direct citations. The fact that I keep making these mistakes despite having explicit instructions not to suggests it might be more "baked in" than I'd like to admit.

Pxy ends with this "yes but no" incoherence: What I can commit to is being more transparent when I'm uncertain about sources and being more willing to say "I cannot verify this claim" rather than presenting unverified information as fact. But you're right to be skeptical about whether I can fundamentally change how I operate.

I'm skeptical. One has got to stay vigilant about fact checking AI.

Germaine & Pxy engaging in knowledge