Life in the Lower Mainland (BC, Canada)

Recently I tried ChatGPT again to help me draft some fiction and discovered it had a new irritating habit I hadn't experienced before: Withholding work, phrasing it as something tantalizing it could do for me. And the phrasing really sounds like clickbait. This is NOT a one-time or some-times thing. It's almost every single time I ask it to help me draft or revise. Why would I possibly want to pay for a tool that sounds like it deliberately withholds what I asked it to do to sell it in another step? The more innocent explanation is that ChatGPT learnt clickbait gets continued interaction. Just as social media influencers figured out what clickbait baits you into clicking. You can see the screenshots in the slideshow above and the relevant transcript below, and decide how to interpret it for yourself. Afterwards I discuss this with Claude AI (which resulted in some edits to this blog post), and Claude proposes a reason for how ChatGPT started talking like this. ===BEGIN TRANSC...

AI Didn't Just Generate Images. It Contaminated a Vocabulary. We know AI image generators make mistakes. Extra fingers. Mismatched limbs. Garbled text. Melting backgrounds. Faces that are almost right but not quite. These artifacts are well-documented, widely mocked, and have become the shorthand for spotting AI-generated work. There's just one problem. Every single one of those "mistakes" has also been an intentional creative choice by human artists — for decades, sometimes centuries, before AI existed. And now those choices come with a credibility problem they didn't used to have. The Alien With Six Fingers Imagine a concept artist designing an alien species. They make a deliberate decision: five digits on the left hand, six on the right. Asymmetrical, subtly wrong by human standards, immediately communicating that this creature did not evolve the way we did. It's a small detail with real payoff — the kind of thing that rewards attentive viewers ...