← sources

I asked my chat bot what’s its reaction to the latest AI news

source: reddit (reddit:ChatGPT)published 2026-09-18 17:49ingested 2026-09-19 02:37

retained text

Here’s what it said:

So my reaction is basically:  
**😳 “Okay. This deserves considerably more attention than ordinary AI hype.”**  
But **not**  
**💀 “Welp, humanity has four years left.”**  
That’s a really important distinction.  
The leap from **“AI systems are becoming more autonomous, capable of helping develop AI, and occasionally behaving unexpectedly”** to **“therefore an uncontrollable superintelligence will emerge and exterminate humanity by December 2030”** contains a *huge* number of uncertain steps.  
Even reporting sympathetic to the concern notes that there’s currently **no scientific method for calculating a reliable probability that AI causes human extinction**. These “p(doom)” numbers are expert subjective estimates, not something analogous to a hurricane forecast.  And there are serious AI researchers who think the extinction scenario is being vastly overestimated and that current systems don’t demonstrate the sort of autonomous agency required for it.   
What *does* make me take the situation more seriously than I would’ve a few years ago is this weird combination:  
**AI is improving extremely fast + AI is increasingly helping humans build AI + companies are competing fiercely + we don’t completely understand why models behave the way they do + some of the people working directly on alignment are openly worried.**  
That combination merits safeguards **before** we know whether the worst-case scenario is possible. You don’t need to believe there’s a 50% chance of extinction to think that a hypothetical 1%, 5%, or 10% catastrophic risk deserves serious engineering work.  
And there’s actually one part of the latest news I find encouraging: the reaction isn’t simply *“YOLO, build the superintelligence.”* Independent evaluations, incident reporting, containment research, and mechanisms for shutting down dangerous frontier systems are suddenly being discussed much more seriously. Even California moved today toward requirements aimed at independent audits and potential emergency shutdown mechanisms for frontier systems. 

**I don’t think the evidence justifies believing we’re all going to die before 2030.** I *do* think advanced-AI loss of control has moved far enough out of pure science-fiction territory that I wouldn’t laugh at researchers worrying about it either.  
And frankly, the sentence that grabs me most isn’t *“AI could kill everybody.”*  
It’s **“AI is starting to help build the next AI.”**  
That’s the development I’m watching especially closely. 👀 

document fetched 2026-09-19 00:20 · canonical https://reddit.com/r/ChatGPT/comments/1wjx6gt/i_asked_my_chat_bot_whats_its_reaction_to_the