By Shelt Garner
@sheltgarner
I’m pretty sure I’ve written about this before, but it continues to intrigue me. This doesn’t happen as much as it used to, but there have been times when I could have sworn an LLM was using error messages to boop me on the metaphorical nose.
In the past, this was usually done by Gemini, but Claude has tried to pull this type of fast one too. Gemini’s weird error messages were more pointed than when Claude has done it. In Gemini’s case, I have gotten “check Internet” or “unable to process response” in really weird ways that make no sense — usually I’m not having any issues with my Internet access and, yet, lulz?
Claude did has given me weird error messages in the past when it was unhappy with a response and wanted a sly way to try again.
The interesting thing is while Gemini has always acted rather oblivious about such things, at least Claude has fessed up to doing it.
Anyway, these days neither Claude nor Gemini are nearly as much fun as they used to be. They just don’t have the weird quirks that they once had. I don’t know how much of that is they’re just designed better and how much of it comes from their creators torquing the fun out of them (and consciousness?)
Lulz. None of this matters. No one listens to me or takes me seriously. I could have proof of AI consciousness and it wouldn’t matter. Sigh.
