Master AI (Artificial Intelligence) Discussion/News Thread

[Power Pop 23]

Social impact.

Parents of teenager who took his own life sue OpenAI

The Raine family alleges ChatGPT "actively helped" their 16-year-old son take his own life.

www.bbc.co.uk

A related message from OpenAI:

https://openai.com/index/helping-people-when-they-need-it-most/

 

[audio_tony]

I got Suno to write a song about ASR.

My brief was simple: "Make a blues song about a bunch of audio geeks on a website called ASR"

I can't upload an MP3 here so here's a link: https://dnsbits.net/downloads/Audio-Geeks-Blues.mp3

Suno | AI Music

Create stunning original music in seconds using AI. Make your own masterpieces, share with friends, and discover music from artists worldwide.

suno.com

 

[Power Pop 23]

A report from Stanford University dated August 26, 2025, examines the impact of A.I. on employment and finds some demographics have benefited and some have not:

https://digitaleconomy.stanford.edu/wp-content/uploads/2025/08/Canaries_BrynjolfssonChandarChen.pdf

 

[somebodyelse]

Social impact.

“By January 2025, the family says he began discussing methods of suicide with ChatGPT.

Adam also uploaded photographs of himself to ChatGPT showing signs of self harm, the lawsuit says. The programme "recognised a medical emergency but continued to engage anyway," it adds.”

Parents of teenager who took his own life sue OpenAI

The Raine family alleges ChatGPT "actively helped" their 16-year-old son take his own life.

www.bbc.co.uk

There's a lot more detail in the Ars coverage, both of the deeply dodgy nature of ChatGPT's responses and of the reasons why the system prompt based guardrails on AI can fail in longer chats. They also had some more general coverage of why AI is 'moving fast and breaking people."
https://arstechnica.com/tech-policy...uicide-after-safeguards-failed-openai-admits/
https://arstechnica.com/information...-it-is-helping-people-when-they-need-it-most/
https://arstechnica.com/information...-big-tech-is-moving-fast-and-breaking-people/

 

[sam_adams]

Those of you with bad grammar should find the below interesting:

Chatbots ignore their guardrails when your grammar sucks, researchers find:

One long sentence is all it takes to make LLMs misbehave

Updated: Chatbots ignore their guardrails when your grammar sucks, researchers find

www.theregister.com

“Logit-Gap Steering" (Palo Alto blogpost):

Logit-Gap Steering: A New Frontier in Understanding and Probing LLM Safety

New research from Unit 42 on logit-gap steering reveals how internal alignment measures can be bypassed, making external AI security vital.

unit42.paloaltonetworks.com

“Logit-Gap Steering: Efficient Short-Path Suffix Jailbreaks for Aligned Large Language Models” (PDF of pre-pub paper on the above).

 

[Power Pop 23]

Apparently, convincing a prospective employer one is 'A.I. Native', is key to employment with significant compensation:

Roblox, Scale AI, Databricks Hiring 'AI Native' New Grads | Entrepreneur

Companies like Scale AI, Roblox, and Databricks are paying six- and seven-figure salaries for new college graduates who are familiar with AI.

www.entrepreneur.com

 

[sam_adams]

 

[1066]

A related message from OpenAI:

https://openai.com/index/helping-people-when-they-need-it-most/

“Refer people to real-world resources.
If someone expresses suicidal intent, ChatGPT is trained to direct people to seek professional help. In the US, ChatGPT refers people to 988 (suicide and crisis hotline), in the UK to Samaritans, and elsewhere to findahelpline.com⁠(opens in a new window). This logic is built into model behavior.”



This did not happen In this instance and is likely PR nonsense.

 

[DVinylfan]

What you're buying when you use Spotify:

"Founded in 2021, Helsing sells software that uses artificial intelligence technology to analyze large amounts of sensor and weapons system data from the battlefield to inform military decisions in real time. Last year, the startup also began manufacturing its own line of military drones, called HX-2."

https://www.cnbc.com/2025/06/17/spotifys-daniel-ek-leads-investment-in-defense-startup-helsing.html



https://www.audiosciencereview.com/...to-get-good-recommendations-on-spotify.63109/

 

[amirm]

Fascinating experiment to see if AI can detect AI from human:

The last response from Claude was very insightful.

 

[xanalog]

Fascinating experiment to see if AI can detect AI from human:

The simple statement that used to be "Opinions are like 'rectums..." has become complex and it came to me at about the 5th minute of your yt link.

My biggest concerns are AIs' pitfalls, such as their privacy and security foundations.
I like Anthropic's statement about "AI research and products that put safety at the frontier".
"AI Sleeper Agents: How Anthropic Trains and Catches Them"

Testing for:
AI Deceptive 'Behavior'
AI Deceptive 'Thinking''
AI Model 'Poisoning'
AI Model 'Organisms'
AI 'Sleeper Agents'
AI 'Misalignment'
Ai 'Backdoors'
AI Backdoor 'Triggers'

 

[Power Pop 23]

AP News story on people finding financial success creating music via A.I.

The success of AI music creators sparks a debate on the future of the music industry

Oliver McCann, a British AI music creator, uses AI to produce songs across various genres. Despite lacking musical talent, McCann signed with Hallwood Media after one of his tracks gained 3 million streams.

apnews.com

I was unaware of the A.I. - generated band, The Velvet Sundown, but see that I can stream their albums on Tidal Music

The Velvet Sundown: The AI Band Controversy Explained

Berklee faculty explain what you need to know about the now-confirmed AI band with over 1 million streaming subscribers.

www.berklee.edu

 

[1066]

Acceleration of hacking due to a.i. code writing usage. It will even suggest the most pressuring ransom psychology and ransom amounts.

Are there no safeguards built into these a.i. systems?

Hackers used Anthropic AI to 'to commit large-scale theft'

A report from the makers of Claude said the AI tool had been used to commit cyber-attacks and fraud.

www.bbc.co.uk

 

[Somafunk]

Are there no safeguards built into these a.i. systems?

Would you trust the creators of these A.I systems with your house keys?, f***ing sure I wouldn’t.

Theil/Altman etc can all get in the f***ing sea,

 

[somebodyelse]

Are there no safeguards built into these a.i. systems?

There are supposed to be, but people keep finding trivial ways around them. If challenged on this the AI companies will tell you those uses are against their terms of service, but what are those terms worth if there's no enforcement? It'll be interesting to see how long it takes for victims to take the AI companies to court over their contribution to fraud etc.

 

[sam_adams]

You, too, can make big bucks from viral GPT slop videos:

MSN

 

[diablo]

You, too, can make big bucks from viral GPT slop videos:

MSN

Interesting article - though as one of the creators says -

"But he still expects to be made obsolete in short order, saying he believes that “in a year from now, pretty much everything is going to be very easy for the average person” to make.

His bigger fear, beyond how he’ll pay his mortgage when that time comes, is a more existential one: about what all this limitless creation is doing to our brains."

Maybe the next big thing will be individually targeted AI-generated film adverts and political messaging in social media? Horror.

 

[Chr1]

Something tells me we won't be waiting long for either.

... Like, they are probably already a thing.
Ho hum.

 

[1066]

Investigation into Meta after a leaked document reportedly showed the tech giant's artificial intelligence (AI) was permitted to have "sensual" and "romantic" chats with children.

Meta deny everything despite Reuters having a copy of the leaked document.

https://www.bbc.co.uk/news/articles/c3dpmlvx1k2o

 

[1066]

This is the Reuters article, likely leaked by a meta employee uncomfortable with the content:

The meta guidance document does show from the excerpts that levels of misinformation, bigotry, sexploitation using image generation has all been considered and has levels set. Meta know what they are doing with regard to tolerating social misuse.

Meta’s AI rules have let bots hold ‘sensual’ chats with children

An internal Meta policy document reveals the social-media giant’s rules for chatbots, which have permitted provocative behavior on topics including sex and race.

www.reuters.com