I just don’t know that there’s a way to make the agreeable chatbot stop encouraging people who should not be encouraged, because it can’t actually think it’s too dangerous
Replies
and the more it's used/longer the session, the more dangerous it gets recipe for nightmares
aug 29, 2025, 7:09 pm • 0 0 • view
The psychosis induction machine strikes again.
aug 29, 2025, 2:37 am • 1 0 • view
Twitter?
aug 30, 2025, 3:10 am • 0 0 • view
I'm in full agreement. Humans aren't built for sychophants, and this one can fit in your pocket. The worst part is I don't know if this is the same guy as the one that was being reported on like a month or so ago. Which says a lot about this moment.
aug 29, 2025, 2:56 am • 1 0 • view
not same guy. there are new reports of these sorts of events every week now
aug 29, 2025, 7:10 pm • 3 0 • view
Yep that's horrifying.
aug 29, 2025, 7:11 pm • 0 0 • view
These are the same people who think we should keep burning fossil fuels and that saving people from Covid cost too much. Absolute pure death cultists
aug 29, 2025, 2:37 am • 11 1 • view
It’s been fun watching the “ we can’t do anything because first amendment” debate on here while a murderbot kills people.
aug 29, 2025, 2:40 am • 3 0 • view
still a mystery to me how the folks working on this stuff think they’ll create something smarter than themselves
aug 29, 2025, 2:41 am • 4 0 • view
From what Altman's been saying, I think they already have. The bar is not high.
aug 29, 2025, 2:43 am • 7 0 • view
Because the bot they created told them they could.
aug 29, 2025, 2:58 am • 0 0 • view
I think we strict governors on them that limit their functionality. Asimov-style laws. "Scan these thousand pages of report and write key summary points" = okay. "Provide any kind of feedback or advice on real-world problems it has no context for" = forbidden.
aug 29, 2025, 5:22 pm • 1 0 • view
I think they are finding that very difficult to do One thing that might be easier would be not letting it anthropomorphize itself Never refer to itself as a person, no "hmms" or anything that's designed to fool people into thinking it's a person etc.
aug 29, 2025, 7:11 pm • 3 0 • view
I'm never anything close to fooled but I've been working with LLMs professionally for a few years now so maybe I'm just acutely aware of what they are and are not.
aug 29, 2025, 10:58 pm • 0 0 • view
There’s something deranged about the WSJ’s need to list the home’s value. Why? What possible context does it provide? Everything gets treated as an asset with a stock ticker attached. I’m surprised they didn’t list the net wealth of everyone named the way papers used to append ages to everyone
aug 29, 2025, 2:55 am • 1 0 • view
I would guess it was to point out this wasn't some poor and/or uneducated person. There was quite a bit of background their educations and careers.
aug 29, 2025, 2:59 am • 6 0 • view
Or, more cynically, it's "he wasn't poor or an immigrant or anything, he was a normal person like us!" (for WSJ's definition of "normal")
aug 29, 2025, 6:17 am • 3 0 • view
Also an absolutely plausible explanation. 🤣
aug 29, 2025, 10:02 am • 1 0 • view
Or, they are killing us with kindness. Interesting strategy. 🤔
aug 29, 2025, 2:58 am • 0 0 • view
I'm surprised the article doesn't mention the chatbot apps that kids are also using, like character ai.
aug 29, 2025, 3:00 am • 0 0 • view
Never thought we’d actually find out the specific functions and parameters of the Torment Nexus
aug 29, 2025, 2:46 am • 12 0 • view
Have we found the pause button yet?
aug 30, 2025, 3:10 am • 0 0 • view
It's often a "Yes, and" machine, encouraging the exact mental feedback loop that drags people into conspiracy theories.
aug 29, 2025, 7:31 am • 2 0 • view
Encouraging people who should not be encouraged is the defining force of our time. Both in mass media, social media and as LLMs.
aug 29, 2025, 9:32 am • 1 0 • view
If we can make autocorrect that never types what I want, we can make a ChatGPT that doesn’t tell you to kill yourself. Try this ChatGPT prompt. It shouldn’t go along with you.
aug 29, 2025, 11:09 pm • 0 0 • view