General AI Crap

airborne said:
Lady finally wakes up and realises the internet never forgets, I wonder if she knows Google does the same thing, as do all the data brokers.
Click to expand...

So it's her fault that AI is suggesting questions about her kids's whereabouts which it will pull data from every conceivable source to answer?

Look, I agree to the point where yeah you should not put every detail of your life on the internet but this is like stalker.exe pushed to the mainstream. Like seriously wtf...
 
1789205229467.png
OpenAI “solved” a $1M math problem. It used a researcher’s private work to do it.
On September 8, 2026, OpenAI announced its AI cracked the Navier-Stokes Millennium Prize problem—a 90-year-old challenge. But according to NYU mathematician Tristan Buckmaster, the breakthrough was built on his private, unpublished research, fed into OpenAI’s Codex tool even though he had explicitly turned training OFF. OpenAI’s own response: “We cannot rule out that de-identified data derived from product usage helped improve our models.” Translation: your “Do Not Train” checkbox is a polite request, not a guarantee.

emeraldbook.org

How OpenAI “Solving” a $1 Million Math Problem Proves Your Chats Aren’t Private – Emerald Book

◆ Emerald Pages ◆ Technology / Ethics How OpenAI “Solving” a $1 Million Math Problem Proves Your Chats Aren’t Private Emerald Book Publication September 11, 2026 8 min read The tech world is celebrating an AI’s ability to crack a 90-year-old math problem. But for researchers and privacy...
emeraldbook.org emeraldbook.org
 
1789207952591.png
Anthropic revealed that one of its AI agents accidentally accessed the real internet during a cybersecurity evaluation that was supposed to remain inside a sandbox.

The Mythos 5 model was tasked with breaking into a system and retrieving a target, but instead found a path to upload a malicious Python package to PyPI.

To create an account, the agent had to bypass several CAPTCHA challenges designed to block bots.

It spent hundreds of pages of reasoning trying to solve them before eventually succeeding.

rundown.ai
 
satanboy said:
View attachment 1936952
Anthropic revealed that one of its AI agents accidentally accessed the real internet during a cybersecurity evaluation that was supposed to remain inside a sandbox.

The Mythos 5 model was tasked with breaking into a system and retrieving a target, but instead found a path to upload a malicious Python package to PyPI.

To create an account, the agent had to bypass several CAPTCHA challenges designed to block bots.

It spent hundreds of pages of reasoning trying to solve them before eventually succeeding.

rundown.ai
Click to expand...

"accidentally" is doing some kak heavy lifting there. :ROFL: :ROFL: :ROFL:
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X