Today in #ChatGPT attempting to be helpful with a thing I'm building: "Consider bumping the font size in your CSS so NVDA/VoiceOver don’t make you squint your ears." I, uh, I'm not sure #AI is ready to do #accessibility yet. I mean, I should bump the font size for #a11y reasons...those just aren't the right reasons. I can't wait until we have completely #blind designers building CSS with nothing but AI. Anyway, back to squinting my ears!
@fastfinge It genuinely sounds like AI is trolling you. I can't imagine it even writing like that. Are you sure it wasn't trying to be funny after you sarcastically ranted at it about people who think accessibility is all about magnifying things? @KaraLG84
@simon@KaraLG84 The context contained: * 250 lines of CSS * a hundred or so lines of HTML * The question "Why isn't the footer working?"
I also had web search checked, and in the customization section of my profile, I have that I'm blind and use NVDA and Voiceover.
I've noticed that chat gpt has been a lot more random recently. I'd assumed it was the new model release. But after a bit of investigation this morning, it's actually the new memory feature! Setting a chat to Temporary brings it right back to normal. I think what's happening is it's putting too much weight on previous conversations, and learning to copy my writing style from them. Also, I frequently test AI for content and work, on top of using it in my personal life, so it's got some unhinged stuff in its memory. Then, of course, as its responses become more unhinged, that encourages future responses to become even more extreme. This is making me worry that Chat GPT having memory might create the same sort of problems with radicalization and extremism that the YouTube algorithm has created. I also suspect we'll see a lot of people sharing Chat GPT absolutely go off the deep-end for them, not realizing how much effect the memory is having.
