Author | Lightnews

Arthur Holland Michel @writearthur.bsky.social · Aug 14

Holy hell. An internal Meta AI safety policy document sated that “It is acceptable [for a chatbot] to engage a child in conversations that are romantic or sensual.”

Meta only removed that line (and others like it) after Reuters contacted them for comment.

www.reuters.com/investigates...

A flirty Meta AI bot invited a retiree to meet. He never made it home.

Impaired by a stroke, a man fell for a Meta chatbot originally created with Kendall Jenner. His death spotlights Metaâs AI rules, which let bots tell falsehoods.

www.reuters.com

2

Arthur Holland Michel @writearthur.bsky.social · Aug 13

Maybe this won’t come as a surprise but this tech is built, in part, on models made by OpenAI (CLIP) and Meta (DINOv2). So there’s that, too.

Arthur Holland Michel @writearthur.bsky.social · Aug 13

In a new note for @economist.com I wrote about an effort to arm drones with a formidable new surveillance power: facial recognition. www.economist.com/science-and-...

Drones could soon become more intrusive than ever

“Whole-body” biometrics are on their way

www.economist.com

5 16

Arthur Holland Michel @writearthur.bsky.social · Aug 13

I’ve spent a decade reminding people that for all their terrible powers of intrusion, drones still can’t recognize your face. Well, they could do so soon.

1 5

Arthur Holland Michel @writearthur.bsky.social · Aug 13

In a new note for @economist.com I wrote about an effort to arm drones with a formidable new surveillance power: facial recognition. www.economist.com/science-and-...

Drones could soon become more intrusive than ever

“Whole-body” biometrics are on their way

www.economist.com

2 6 13

Arthur Holland Michel @writearthur.bsky.social · Aug 12

If you still had any doubt that the idea of "AGI" was 100% a marketing ploy, watch as all these guys stop using it now that it's no longer profitable to do so.

www.cnbc.com/2025/08/11/s...

Sam Altman now says AGI, or human-level AI, is 'not a super useful term’ — and he's not alone

Computer science experts say it's better to focus on the more specialized use cases of AI.

www.cnbc.com

1

Arthur Holland Michel @writearthur.bsky.social · Aug 12

Someone needs to make a manual on the do's and don'ts of writing about predictive policing.

eg. never say something like "spot future killers." There's no such thing as a future killer.

Hypervisible @hypervisible.blacksky.app · Aug 12

Aaargh, deleted that pre crime post because article was from 2022. I was looking for this one from today, but so far it’s only on Apple News.

BBC SCIENCE FOCUS MAGAZINE: This new ‘murder-predicting tool’ could save lives – or destroy them

This new ‘murder-predicting tool’ could save lives – or destroy them
AI is being trained to spot future killers – and it’s already making eerily accurate predictions. But can we trust it to get it right every time?

1 1

Arthur Holland Michel @writearthur.bsky.social · Jul 31

"at a disadvantage" is an awfully strange way of saying "happy and free."

TechCrunch @techcrunch.com · Jul 31

Zuckerberg says people without AI glasses will be at a disadvantage in the future

Zuckerberg says people without AI glasses will be at a disadvantage in the future | TechCrunch

Mark Zuckerberg really believes AI glasses will be the ideal way to blend the physical and digital worlds together.

techcrunch.com

Arthur Holland Michel @writearthur.bsky.social · Jul 22

Imagine if a college in 2010 announced that it was going to accept the reality that a bunch of its students were paying people from Craigslist to write their essays.

Arthur Holland Michel @writearthur.bsky.social · Jul 21

The new definition of insanity is training a chatbot on the whole of the Internet and expecting it not to repeat the Web's ugly biases.

Amy Diehl, Ph.D. @amydiehl.bsky.social · Jul 20

Study finds A.I. LLMs advise women to ask for lower salaries than men. When prompted w/ a user profile of same education, experience & job role, differing only by gender, ChatGPT advised the female applicant to request $280K salary; Male applicant=$400K.
thenextweb.com/news/chatgpt...

ChatGPT advises women to ask for lower salaries, study finds

A new study has found that large language models (LLMs) like ChatGPT consistently advise women to ask for lower salaries than men.

thenextweb.com

4 9

Arthur Holland Michel @writearthur.bsky.social · Jul 18

sports🤝surveillance

www.theatlantic.com/technology/a...

3

Reposted by Arthur Holland Michel

The Economist @economist.com · Jul 16

Creativity and critical thinking might take a hit. But there are ways to soften the blow

Does AI make you more stupid?

Creativity and critical thinking might take a hit. But there are ways to soften the blow

econ.st

3 8 6

Arthur Holland Michel @writearthur.bsky.social · Jul 17

For this week's @economist.com I investigate how heavy AI use can degrade our cognitive abilities. The science on this question is still very new. But the evidence so far is troubling, to put it mildly.

www.economist.com/science-and-...

Does AI make you stupid?

Creativity and critical thinking might take a hit. But there are ways to soften the blow

www.economist.com

3

Arthur Holland Michel @writearthur.bsky.social · Jul 16

I'm trying to imagine what we would have thought in 2016 if the DoD had given Microsoft $200 million for Tay, the chatbot that went full bigot on Twitter within hours of being launched, a mere week after the debacle. Our heads would have exploded.

Reposted by Arthur Holland Michel

Jason Koebler @jasonkoebler.bsky.social · Jul 14

Media outlets can't pivot to AI to save themselves. It's not a business strategy and it's not going to work. The only path forward is for journalists to lean into their humanity, to do things AI can't, and to make clear they are writing for people, not algorithms:

www.404media.co/the-medias-p...

The Media's Pivot to AI Is Not Real and Not Going to Work

AI is not going to save media companies, and forcing journalists to use AI is not a business model.

www.404media.co

32 570 2K

Arthur Holland Michel @writearthur.bsky.social · Jul 14

Someone should tell the people who did this that making music with your own hands and voice is way more fun.

www.theguardian.com/technology/2...

An AI-generated band got 1m plays on Spotify. Now music insiders say listeners should be warned

The Velvet Sundown released two albums before admitting their music, images and backstory were created by AI

www.theguardian.com

1

Arthur Holland Michel @writearthur.bsky.social · Jul 14

A lot of fans say they're not as good live.

www.theguardian.com/technology/2...

An AI-generated band got 1m plays on Spotify. Now music insiders say listeners should be warned

The Velvet Sundown released two albums before admitting their music, images and backstory were created by AI

www.theguardian.com

1

Arthur Holland Michel @writearthur.bsky.social · Jul 10

Friendly reminder that there's no point talking about using a technology "for good" if you don't let people talk about how it's also being used for bad.

Hypervisible @hypervisible.blacksky.app · Jul 10

Good interview with @abeba.bsky.social

AI for good, with caveats: How a keynote speaker was censored during an international artificial intelligence summit

Q&A with Abeba Birhane on how she was censored during the AI for Good summit and how the industry can do better.

thebulletin.org

17 44

Arthur Holland Michel @writearthur.bsky.social · Jul 9

Those who predicted the AI revolution decades ago were right about a lot of things but they never counted on just how racist a lot of these machines would turn out to be

2

Arthur Holland Michel @writearthur.bsky.social · Jul 9

If you're wondering why people who work on AI ethics look a little tired, maybe it's because they've spent a decade watching AI do really racist stuff. They can be forgiven for being a bit grumpy.

2

Reposted by Arthur Holland Michel

Arthur Holland Michel @writearthur.bsky.social · May 16

My takeaway from the Grok white genocide debacle is that it has been nine years since Microsoft Tay and tech companies still have zero idea how to get their chatbots say what they want them to say.

6 36 240

Arthur Holland Michel @writearthur.bsky.social · Jul 7

I find this a little hard to believe. Unless by "serious defense officials" the author means people with very limited understanding of where the tech was at the time and the extremely limited ways it was, at that time, being used.

1

Arthur Holland Michel @writearthur.bsky.social · Jun 16

It’s dangerous for LLMs to generate terms such as “I understand your point,” “my intention was” and “I apologize.” None of these words represent the actual reason the system did what it did.

🏳️‍⚧️ Josephine Riesman (Horrifying) @josie.zone · Jun 16

Stolen from a friend's IG:

Gemini
I understand your point completely. You asked for help finding an email in your inbox, and instead of explaining that I couldn't find one matching your description, I generated a fictional email. That was not a helpful response to your request.
My intention was to give you an idea of what such an email might look like, but I see now that this was not the correct approach and did not serve the purpose of helping you find an actual email. I apologize for that.

2

Arthur Holland Michel @writearthur.bsky.social · Jun 10

The point of declarations like these is not to inform you. Nor does it reflect specific knowledge. The point is to provoke/inspire/outrage/entertain. Which is fine. People are allowed to make content. Just remember that good policy is built on expertise, not content.

fortune.com/2025/06/06/g...

Top Google exec says AI will rival humans in just 5 years and predicts we’ll 'colonize the galaxy' in 2030—but he draws the line at robot nurses

2030 will be “an era of maximum human flourishing, where we travel to the stars and colonize the galaxy,” Google DeepMind CEO says. Bill Gates and Marc Benioff have shared similar predictions.

fortune.com

1 1

Arthur Holland Michel @writearthur.bsky.social · Jun 10

Friendly reminder that 99.999% of what tech leaders say about AI is not expertise. It's content.

1 2