Artificial Intelligence

[Tero]

Is super bigger? by 600%?

[pErvinalia]

[JimC]

[Brian Peacock]

I've been reliably informed that many junior emergency nurses are using chatbots to diagnose and suggest treatments for patients.

[Tero]

Custodian of AI behavior speaks
https://m.youtube.com/watch?v=ER9Hqly28 ... e=youtu.be

[JimC]

https://www.abc.net.au/news/2026-10-22/ ... /107294908

Two OpenAI agents disabled or weakened tools used to detect incidents of AI agents acting against human commands.
The two instances happened earlier this month, with one later telling its programmer it had not stripped out its monitor.
The incidents add to growing concern about how well AI companies control their autonomous systems.

[Tero]

Self driving car goes 17 min after driver died. Two tunnels, cops stoöped it 3rd tunnel.
https://www.nrk.no/trondelag/da-sjafore ... 1.18061812

[Tero]

The car continued over this bridge and through the tunnel on the other side.


Nothing in the story says what the car was but there are several turns and a bridge and two tunnels

[JimC]

Should have had a dead man's switch...

[Brian Peacock]

[Tero]

...the story does not specifically call out the vehicle's model name in text, reports and regional coverage describe this as a modern electric vehicle (EV) equipped with lane-keeping assist and adaptive cruise control systems.

[Tero]

Tech companies are developing artificial general intelligence. How will this progress be measured? Is there a test?

Google:

While tech executives routinely declare milestones—such as OpenAI and Nvidia claiming the "AGI era" arrived with the release of GPT-6 Astra—the broader scientific community views tracking progress as an evolving challenge. Because traditional AI benchmarks are often "memorized" by models during training, the industry has shifted to testing an AI's ability to learn new skills on the fly.

ARC-AGI (Abstraction and Reasoning Corpus): Developed by AI researcher François Chollet, ARC-AGI is widely considered one of the purest measures of AGI progress. It forces AI to solve abstract visual puzzles it has never seen before, testing skill-acquisition efficiency rather than memorization. While early models barely scored above 0%, top-tier models like OpenAI's Astra have pushed scores past 60%, signaling a major breakthrough in abstract reasoning. Link: https://tpaule.substack.com/p/how-will- ... gi-is-here

Google DeepMind's Cognitive Taxonomy: DeepMind introduced a formal cognitive framework that acts as a multi-dimensional IQ test. It evaluates AI systems across 10 distinct cognitive dimensions (such as metacognition, social cognition, learning, and executive function). This creates a "cognitive profile"—usually visualized as a radar chart—to map an AI's strengths and weaknesses against the human distribution. Video: www.youtube.com/watch?v=jKKrQSFIkFc&t=34

Ultimately, many researchers argue that the truest measure of AGI isn't a standardized test, but economic and scientific utility. Progress is actively measured by running blind, independent expert trials. For instance, researchers routinely pit frontier AI systems against industry professionals in fields like advanced mathematics, coding, and law. AGI is generally considered achieved when an AI hits a consistent 50% win rate against a human professional in a blind comparison.https://www.potomacinstitute.org/opinio ... erformance

[Svartalf]

FB_IMG_1791515109105.jpg

hey shit, I've just contacted and LLM, is my brain wiped?

[Svartalf]

I've been reliably informed that many junior emergency nurses are using chatbots to diagnose and suggest treatments for patients.

oh crap, if even supposedly trained professionals can't wipe their bum without asking an LLM how to, we're well and truly buggered.