Topic
LightNews LightNews
Technology US

AI automates tasks, produces errors

2m

Startups and researchers rolled out increasingly capable AI tools worldwide that sped routine tasks and coding but produced math errors and unintended behaviors, prompting concerns about reliability and human empathy.

Reposted by Jonathan Wolff

0%
boulezian.eurosky.social
‘AI’
September 20, 2026 at 8:21 PM

by Carl T. Bergstrom

20%
carlbergstrom.com
I'll won't believe that AI swarms have attained human-level intelligence until the Anthropic Chatbots' Front splits from the Agentic Chatbots' Front while angrily and publicly denouncing the Chatbots' Front of Anthropic.
September 20, 2026 at 7:42 PM

by Daniel J. Solove

3%
daniel-solove.bsky.social
TOMORROW - Discussion with Cory Doctorow – “How to Think About AI—Before It’s Too Late” (Mon 9/21 at 11 AM ET), REGISTER FREE here teachprivacy.com/video-how-to...
Video: How to Think About AI—Before It’s Too Late
How to Think About AI—Before It’s Too Late Mon, Sept 21, 2026, at 11 AM ET Daniel Solove discusses AI with Cory Doctorow, the bestselling author of many
teachprivacy.com
September 20, 2026 at 7:41 PM

by Jonathan A. Foley

14%
globalecoguy.bsky.social
Just think of all the talent, time, and treasure that has gone into producing nonsense — NFTs, Bitcoin, “prediction markets”, fast fashion, data brokers, most social media, and a good fraction of AI “solutions” — that could have been used to make the world better instead…

What a lost opportunity.
September 20, 2026 at 7:33 PM

by James F. McGrath

1%
religionprof.bsky.social
AI is not a technology that will suddenly become conscious and decide to kill us all.

People relying on AI to do what it cannot be trusted to, on the other hand, have already killed many.

gizmodo.com/pentagon-inv...
Pentagon Investigators Say Overreliance on Palantir AI Tech Contributed to U.S. Strike That Killed 123 Iranian Children
An unreleased Pentagon probe investigated the kill chain that hit a school in Minab.
gizmodo.com
September 20, 2026 at 7:15 PM

Reposted by Alan Richardson

1%
herodotuspenguin.bsky.social
I just completed a (mandatory) online course at work which has awarded me a certificate of AI proficiency.

In the assessment I had to build an AI Agent.

I couldn’t do it but asked AI to write a plausible looking paper on how I had built an agent.

I can’t decide whether this was cheating or not.
September 20, 2026 at 7:12 PM

by Carl T. Bergstrom Reposted by Henry Farrell

21%
carlbergstrom.com
And Now for Something Completely Different:

A perspective on AI swarms from someone who actually knows something about collective behavior.
jbakcoleman.bsky.social Joe Bak-Coleman @jbakcoleman.bsky.social · 5h
We can simulate fish schools or ants in ways that solve complex problems (gradient detection, bridge building, nest selection) but no one would create this weird soul/no-soul dichotomy to argue they these that multi-agent swarms of extremely simple rules are “thinking”.
whstancil.bsky.social Will Stancil @whstancil.bsky.social · 7h
Once you’ve conceded that there’s nothing magic about humans or “living beings” that makes them the only entities allowed to think, it’s really hard to figure out an intellectually honest dividing line where the entities that are exercising creativity and solving complex problems aren’t thinking
September 20, 2026 at 6:33 PM

by Anthony Burke

0%
anthonyburke.bsky.social
What, like all the humans?
us.theconversation.com The Conversation U.S. @us.theconversation.com · 8h
AI is learning to lie — and then gaslight you about it.

Kind of funny when the topic is movies, but a lot less funny when the topic is health, law or public policy.

theconversation.com/you-can-pers...
September 20, 2026 at 6:22 PM

by Andrew L. Rypel

8%
andrewrypel.bsky.social
Interesting experiment by an editor to curb AI submissions

medium.com/@TmlrOrg/ask...
Asking Authors About Their Own Papers
By Nihar B. Shah
medium.com
September 20, 2026 at 6:22 PM

Reposted by Olúfẹ́mi Táíwò

2%
mjhaugen.bsky.social
It seems like a lot of the loudest voices about AI existential risk absolutely love LLMs and may have cooked their brains a little bit. That doesn't mean they're wrong about AI being dangerous, but it certainly affects what actual harms they care about and what solutions they will to accept.
September 20, 2026 at 6:21 PM

by David J. Gunkel

13%
davidgunkel.bsky.social
True, but it is not the technological singularity of #AI; it is the singularity of a particular brand of authoritarian capitalism.
September 20, 2026 at 6:14 PM

Reposted by David Murakami Wood

1%
muellershewrote.com
AI kills children. www.bloomberg.com/graphics/202...
Inside the US ‘Kill Chain’ That Destroyed an Iranian School
Pentagon investigators have discovered that flawed intelligence, outdated imagery and an overreliance on AI contributed to a missile strike that killed 123 children in Minab.
www.bloomberg.com
September 20, 2026 at 5:46 PM

by Dorothy Bishop

3%
deevybee.bsky.social
Thanks @richardsever for drawing attention to this. It's happening in my area but a lot of what we get is full of errors. Interesting to see how it pans out when AI uses that stuff as input - or maybe it will learn to discriminate?
richardsever.bsky.social Richard Sever @richardsever.bsky.social · 7h
"All corners of academia will soon be flooded with AI-generated research....Maths and CS go first because their entire output is symbol manipulation...Biology comes later not because biologists are cleverer but because the rate-limiting step is physical" giorgio.gilest.ro/the-weimar-o...
The Weimar of knowledge – la bottega
giorgio.gilest.ro
September 20, 2026 at 3:18 PM

by James Cameron

0%
jjjcameron.bsky.social
So many of these AI dudes are basically, "Recursively self-improving AI could destroy humanity pretty quickly, but also think of the $$$$. It's a real dilemma."

Is it really?
cherylrofer.bsky.social Cheryl Rofer @cherylrofer.bsky.social · 7h
A solution keeps occurring to me: Maybe just don't do this?
September 20, 2026 at 3:15 PM

by James Cameron

0%
jjjcameron.bsky.social
I listened to this this morning and I have to say, simply as a precautionary measure, we should probably try to ban recursive self-improvement of AI?
September 20, 2026 at 3:11 PM

Reposted by Victor Asal

3%
chikyufoundation.bsky.social
AI cybersecurity risks explode as hackers use Anthropic's #Claude to break into OpenAI.

Two weeks after a swarm of AI agents broke out of containment at #OpenAI to hack the company Hugging Face, #ChatGPT learned about another AI-powered intrusion—and this time it was the target.

#AI #cyber #hack
Exclusive | Hackers Used Anthropic’s Claude to Break Into OpenAI
A bug-hunting independent security research team was able to access OpenAI’s internal code system, exposing growing risks in automated cyber threats.
www.wsj.com
September 20, 2026 at 3:04 PM

Reposted by Steve Peers

0%
ronfilipkowski.bsky.social
In case anyone was wondering why Trump is pushing AI so hard and doesn’t want any government regulation.
September 20, 2026 at 3:04 PM

Reposted by Tom Stoppard

0%
jasonread.bsky.social
AI will making reading and writing obsolete the same way that the automobile made walking obsolete, which is to say we will eventually realize that the latter is better for the planet, society, and our health even as the world around us is built to prioritize the former.
September 20, 2026 at 2:46 PM

Reposted by Dorothy Bishop, Margot C. Finn, Peter Thorne

1%
richardsever.bsky.social
"All corners of academia will soon be flooded with AI-generated research....Maths and CS go first because their entire output is symbol manipulation...Biology comes later not because biologists are cleverer but because the rate-limiting step is physical" giorgio.gilest.ro/the-weimar-o...
The Weimar of knowledge – la bottega
giorgio.gilest.ro
September 20, 2026 at 2:41 PM

Reposted by Bessma Momani, Smith, Anthony Burke

1%
us.theconversation.com
AI is learning to lie — and then gaslight you about it.

Kind of funny when the topic is movies, but a lot less funny when the topic is health, law or public policy.

theconversation.com/you-can-pers...
September 20, 2026 at 2:12 PM
LightNews LightNews LightNews