cross-posted from: https://discuss.online/post/41098657
http://www.smbc-comics.com/comic/huh-2
Alt text
Once I realized this, all those inept AI laundry-folding videos became hilarious.
Bonus panel
cross-posted from: https://discuss.online/post/41098657
http://www.smbc-comics.com/comic/huh-2
Alt text
Once I realized this, all those inept AI laundry-folding videos became hilarious.
Bonus panel
“Oh, I would probably just mess it up. You’re just so much better at that kind of stuff than I am!” :p
Feels especially fucked up/funny in a “haha humanity is doomed kinda way,” in the context of AI hallucinating, then doubling down/gaslighting users when corrected https://www.cmu.edu/news/stories/archives/2025/july/ai-chatbots-remain-confident-even-when-theyre-wrong
It’s weird but I just remembered for quite a while circa 2024, we seemed to be hearing stories about corrected chatbots gaslighting users pretty frequently, then suddenly the topic just kind of disappeared from the news.
not exactly surprising that we are hearing more stories/examples of AI induced delusions in people, but it really makes me wonder if there was a concerted effort to kill news stories related to AI doubling down when incorrect.
No AI(LLMs) just got better, more accurately they got better at detecting they are incorrect and correcting themselves.
If you want to sort of grasp how it works, just search up the
Chatgpt seahorse emojidebacle, arguably the best example of what happens when deep training data fights with the self-check logic.I’m not tech savvy in the least, so not debating they’ve improved significantly since the original article was written in 2025.
The author of the article wrote an opinion piece about it back in March of this year but its hidden behind a paywall and archive links all seem to be unavailable on my end :(
https://www.post-gazette.com/opinion/guest-columns/2026/03/05/ai-system-metacognition-humans-overconfidence-trust-trent-cash-daniel-oppenheime/stories/202603050012