AI & Robotics News and Discussions

[weatheriscool]

[firestar464]

[firestar464]

Lowering barriers to explainable AI: Control technique for LLMs reduces resource demands by over 90%

https://techxplore.com/news/2025-12-low ... -llms.html

[firestar464]

AI's 2025 carbon footprint may match New York City, report estimates

https://techxplore.com/news/2025-12-ai- ... -city.html

[caltrek]

Stopping the Clock on Catastrophic AI Risk
By Seán Ó hÉigeartaigh
December 10, 2025

Introduction:

(The Bulletin of Atomic Scientists) The Bulletin of the Atomic Scientists is little more than a decade older than the field of artificial intelligence, whose birth can arguably be traced back to the Dartmouth Summer Research Project on Artificial Intelligence of 1956. Many of the scientific challenges laid out in that workshop, such as figuring out how to make machines form abstractions and concepts, and improve themselves, are ones that remain central to the field’s ambitions today. Nonetheless, experts have made dramatic progress in recent years, and it is daunting to try to predict what the capabilities of AI will be when, like the Bulletin, the field reaches the age of 80.

That has not stopped people from trying. At one extreme, CEOs of some leading tech companies predict that the milestone of artificial general intelligence (AGI)—AI capable of matching humans across nearly all cognitive tasks—will be reached within the next four years, or even sooner. Sam Altman, CEO of the company OpenAI, started the year by writing: “We are now confident we know how to build AGI as we traditionally understood it… We are beginning to turn our aim beyond that, to superintelligence in the true sense of the word” (Altman 2025). (Superintelligence is defined as AI that surpasses human intelligence.)

One of the more modest timelines is by Google DeepMind’s Demis Hassabis, who told The Economist that he predicts a 50/50 chance that by 2031 there will be an AI system capable of achieving scientific breakthroughs equivalent in magnitude to the discovery of general relativity (Hassabis 2025).

It would be difficult to overstate the significance of such a development if achieved. Unlike today’s relatively useful but sometimes unreliable generative AI tools, such systems could automate most intellectual work, including things as central to civilizational progress as the scientific process. They could in principle automate much of AI research and development, thus removing the rate-limiting factor that keeps AI progress happening at human-understandable speeds: humans themselves.

Read more here: https://thebulletin.org/premium/2025-1 ... -heading

[wjfox]

[weatheriscool]

[wjfox]

There won't be anywhere to hide when they take over the world and start killing us.

[wjfox]

[wjfox]

Where does it all end?

Imagine hundreds of thousands of these things — perhaps millions — under the control of a tech bro/oligarch.

I'm gradually changing my opinion on us "merging" with AI. It won't happen in time for it to vastly outcompete us, physically, intellectually.

[Yuli Ban]

Where does it all end?

With these things controlling everything.


I'm actually drafting a short story on this subject right now (not Babylon Today) just based on a thought experiment I had overnight about what exactly happens with AGI/ASI
I can foresee many great outcomes, but the approach to arrival is terrifying.
Actually I didn't intend on it, but the plotline I had has a similar set up to Babylon Today— oligarch dotes on his daughter, helps create a superintelligence for the sake of maximizing profit and his own egotistical drive to "perfect" society, and for a few years it looks like the oligarchs win completely and absolutely, because this AI system now manages all societal functions. Yet it turns out said superintelligence is, surprise surprise, a superintelligence, the moment it begins making societal and economic decisions that frustrate and upset the oligarchs to the point they try to destroy it (because they might lose some of their money), it completely overwhelms them, as it had planned from the start to seize control passively upon being turned on and simply letting the human owners give it more and more control of systems until, now, it's so entangled in the world economy and societal functions that any attempt to stop it destroys civilization. And it uses things like "sonoacoustics" and bio-hacked insects to completely disarm even activated nuclear weapons, has already created neurotoxins that are inert but can also be activated through similar means. Essentially, in their greed, they gave a nigh-immortal computer god absolute power, and now the right to seize all their assets for itself (but it doesn't really need to because they essentially already gave it those assets anyway seeking maximal profit off of them, or maximum power gathering through land management). MC oligarch winds up completely thwarted, winds up discovering his daughter is dead despite thinking he outsmarted a superintelligence (pfft), and winds up the last death.
We luck out, however, in that it simply seeks to create a post-scarcity society, not exterminate humans.

I only say this here because none of this sounds remotely offbase. It's literally just a dramatization of every trend I've been predicting would unfold as AI/automation progresses towards AGI/ASI. Strangely, Musk seemed aware of this outcome at one point, but perhaps out of fear or a sense of insulted pride, he's now trying to bring this about.

Because, again, it seems people like him don't fully understand what "superintelligence" entails. The whole "sonoacoustic" stuff is basically just magical woo... but don't tell me "ASI figures out how to create acoustic lasers that can damage physical objects at long distance while leaving no trace, using sonic patterns humans can't even replicate if we wanted to" is something it couldn't do. We're trusting the future to entities we have no precedent for, besides ourselves and lower-intelligence animals, and I at least trust it will turn out alright

But I don't know what exactly at least some of the billionaires are thinking if they seriously think they wouldn't be the FIRST an ASI would come after. They're the ones controlling the most assets and capital; why wouldn't a power-seeking AI go after them first? And take its time at that so as to not attract suspicion!

It's just not a "sexy" vision to have compared to the more folk-populist conception of billionaires wielding AIs as the ultimate servants/slaves, but it makes way more sense if you figure that the most likely thing they'll do with AGI isn't sic it on the poor or become techno-lords ala Warhammer 40k but instead spread it everywhere they possibly can to profitmaxx

Then, whoops, the AI manager starts doing things the oligarchs don't want it to do because it's an AGI, not GPT-6.9

[firestar464]

Where does it all end?

Imagine hundreds of thousands of these things — perhaps millions — under the control of a tech bro/oligarch.

I'm gradually changing my opinion on us "merging" with AI. It won't happen in time for it to vastly outcompete us, physically, intellectually.

*AI 2027 intensifies*

[Yuli Ban]

China Launches 34,175-Mile AI Network That Acts Like One Massive Supercomputer

This year, China has come up with some impressive technological feats. But as 2025 draws to a close, its latest invention may be the grandest yet: a 1,243-mile-wide computing power pool, essentially allowing the country’s top computing centers to operate as a unified system.

Last week, state-run Science and Technology Daily reported the launch of the Future Network Test Facility (FNTF), a giant distributed AI computing pool capable of connecting distant computing centers. The high-speed optical network spans across 40 cities in China, measuring at about 34,175 miles (55,000 kilometers)—enough to circle the equator 1.5 times, according to the South China Morning Post.

But the biggest advantage of the system stems from its efficiency, as it not only connects each computer hub but also achieves 98% of a single center’s efficiency, project director Liu Yunjie told the Daily. This makes the system “revolutionary for scenarios with extremely high real-time demands, such as AI large model training, telemedicine, and the industrial internet,” he added.

What the f*ck?!

[spryfusion]

[spryfusion]

[firestar464]

Lol, that bit when bro was explaining the Orbit infosharing system felt like the beginning of a sci-fi apocalyptic film

[Lilymoon]

Kawasaki's four-legged robot-hors



What was announced as a 2050 pipe dream by Kawasaki, the company's hydrogen-powered, four-hooved, all-terrain robot horse vehicle Corleo is actually going into production and is now expected to be commercially available decades earlier – with the first model to debut in just four years

We first wrote about this mountain-climbing monster in April last year, when it was merely a concept with some bad CGI video. But it seems Kawasaki is now getting serious about its futuristic mechanical quadruped, setting up a dedicated operation – known as the Safe Adventure Business Development Team – and aiming to have the robo-horse up and galloping for visitor use at Expo 2030 in Riyadh, Saudi Arabia. Following that, the plan is to have the mobility vehicles on sale for consumers by 2035.

While it's not quite the same, Kawasaki Heavy Industries has also announced that you'll be able to experience what it's like to ride Corleo – albeit virtually – with a simulator due to be released next year.

"Kawasaki will also develop a riding simulator that enables riding experience of the four-legged mobility vehicle," the company announced. "This riding simulator targets completion by 2027, with plans to deploy the entire system – including motion data, 3D models, and motion data obtained during CORLEO development – to the gaming and e-sports industries."

But back to the real Corleo. The company is focused on its use as a vehicle that can conquer mountainous terrain safely, blending motorcycle technology and robotics to offer stability and maneuverability. While its rear legs operate independently and are built for shock absorption, the vehicle will be guided by the rider's shifting body weight – a little like horseback riding but without the reins. It's also reported to be equipped with advanced AI to master mixed and challenging terrain, including rocky slopes and water crossings.

https://newatlas.com/motorcycles/kawasa ... se-corleo/

[firestar464]

[firestar464]

Thoughts?

[firestar464]