That is not the verbal dunk that you think it is.
The newest version of ChatGPT has to be told
Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user’s query
Maybe GPT 5.5 is literally hallucinating:
Just ran across a YouTube short that I think makes a very salient point. The whole thing is here, but I’ll quote the meat of it:
AI can’t tell truth.
Now, let’s imagine that your toddler sneaks into your French pastry pantry and eats a pain au chocolat. You catch them and ask if they ate a croissant. Now the kid doesn’t know the difference between a pain au chocolat and a croissant, but says no in order to get out of trouble. Even though the kid is technically correct, they’re still lying. They predict the response you want to hear and says it regardless of truth.
What we call AI today is just predicting what we want to hear. It’s always lying. Even if it accidentally tells a truth, it’s all this AI can do.
For those of you who enjoy the insightfulness of LLM’s, note that they are terrific at sentiment analysis of text. In this case, someone dumped a bunch of text from different countries into copilot, and it did a pretty good job of separating out the attitudes and tones of those countries.
I guess the catch here is that he dumped the same text for each country. But we know how those countries really are, am I right?
for those (most of you? ALL of you) who don’t bother to click on a link, here’s a take away from that one
there’s a real risk that people are currently using AI to produce analysis that bears no resemblance to what people actually said
I found another great use today, and I wasn’t even looking. I went to watch some board game videos in German on YouTube and was startled to discover that YouTube had reviewed my location and auto-dubbed the four presenters into English.
Since I speak German this felt very above-and-beyond, and I appreciated the additional humor of all four of them, male and female, dubbed in the same male voice. Hye-larious!
More seriously: I assume that the original was transcribed and then translated, and then the translation matched to timestamps. Which is impressive, putting aside whether the translations was good or not (I couldn’t tell).
That reminds me of dubbed films in Poland and I think Hungary, where the same male voice actor played all the parts.
Someone on X posted a real Monet but said it’s AI, and asked others to critique it. Lots of pretentious people took the bait.
I wonder how Charlie Chaplin would have placed in an AI-generated Charlie Chaplin lookalike contest…
A little while ago, IIRC, someone posted in one of the AI related threats that Chrome was automatically downloading AI related “features” which significantly increased the bloatware. I only briefly looked at it and don’t have the details.
Does anyone remember this? My seven-year-old laptop is suddenly using more memory, and I wanted to investigate if this was one of the issues.
Here’s some more info on what you can do about it.
thank you @zbuzz !
Well, I guess I won’t be using Google anymore.
In a blog post, Google VP of Search Elizabeth Reid writes that the company is “introducing the biggest upgrade to our Search box in over 25 years — now completely reimagined with AI. This intelligent Search box puts our most powerful AI tools right at your fingertips, making it easier to ask your questions.” It appears that the box will get bigger when you ask it certain kinds of questions, pushing you even more toward AI-generated results rather than website links. Reid writes that it will also help users “formulate your question with AI-powered suggestions that go beyond autocomplete.”
I want what PCs, browsers, search engines and websites used to be in 2005.
AI is good for writing short stories good enough to win prizes. Maybe?
Did AI Write This Prizewinning Story? - by Novi Zhukovsky
No, I’m not being sarcastic. I assume the story is pretty good, I haven’t read it.
My favorite moment, though, from Granta’s response:
We showed Claude.ai the story and asked whether it was AI-generated. The response was long, concluding that it was ‘almost certainly not produced unaided by a human’. It ended, however, on a cautious note:
‘The strongest evidence against pure AI authorship is the small number of passages that don’t fit the pattern – particularly the Zoongie morning-greeting and the agricultural paragraph about the acre. Those passages carry the kind of off-shape specificity that models still struggle to produce unprompted. If the story has a human core, it is concentrated there, and the AI has been used to elaborate around it . . .’
“We fed the story into the plagiarism machine and it equivocated…”
Whodathunkit!
I defend AI when it’s worth defending, but this is just abuse. When Google introduced their defective “AI summary” feature in search results, there was widespread interest in ways to suppress it that made their way around the internet. This new “feature” might be unavoidable and if so, and if it’s obtrusive enough, I may have to start using a different search engine.
I definitely think Google search is worse now that the first thing it shows you is an AI summary; it’s just continually wrong in my experience; for example I found a tree that looked like an elder tree in every respect, except that the berries were green instead of dark purple; they were ripe and falling off the tree, so I reasoned this must be a variety or mutation that just lacks the purple pigment in the fruit. I searched for all kinds of combinations of ‘elderberry with green fruits when ripe’ and ‘green fruited elder’ and all it would give me was “the phrase ‘green fruited elder’ probably refers to unripe berries”.
It was as I suspected - there’s a variety called Sambucus nigra var. viridis that has berries that are pale green when fully ripe - there’s a fair bit of information about it on the web, but Google insisted on steering me away from it.
Hey I pine for the days when altavista was still an option.