Connect with us

Tech

Google launches new AI model to translates vision, language into robotic actions

Published

on

Google launches new AI model to translates vision, language into robotic actions

Google has introduci a new advancement in robotics in a bid to push closer to a future of helpful robots. Robotics Transformer 2, or RT-2, is a first-of-its-kind vision-language-action (VLA) model.

A Transformer-based model trained on text and images from the web, RT-2 can directly output robotic actions. Just like language models are trained on text from the web to learn general ideas and concepts, RT-2 transfers knowledge from web data to inform robot behavior.

In other words, RT-2 can speak robot.

The pursuit of helpful robots has always been a herculean effort, because a robot capable of doing general tasks in the world needs to be able to handle complex, abstract tasks in highly variable environments — especially ones it’s never seen before.

Advertisement

Unlike chatbots, robots need “grounding” in the real world and their abilities. Their training isn’t just about, say, learning everything there is to know about an apple: how it grows, its physical properties, or even that one purportedly landed on Sir Isaac Newton’s head. A robot needs to be able to recognize an apple in context, distinguish it from a red ball, understand what it looks like, and most importantly, know how to pick it up.

That’s historically required training robots on billions of data points, firsthand, across every single object, environment, task and situation in the physical world — a prospect so time consuming and costly as to make it impractical for innovators. Learning is a challenging endeavor, and even more so for robots.

Recent work has improved robots’ ability to reason, even enabling them to use chain-of-thought prompting, a way to dissect multi-step problems. The introduction of vision models, like PaLM-E, helped robots make better sense of their surroundings. And RT-1 showed that Transformers, known for their ability to generalize information across systems, could even help different types of robots learn from each other.

But until now, robots ran on complex stacks of systems, with high-level reasoning and low-level manipulation systems playing an imperfect game of telephone to operate the robot. Imagine thinking about what you want to do, and then having to tell those actions to the rest of your body to get it to move. RT-2 removes that complexity and enables a single model to not only perform the complex reasoning seen in foundation models, but also output robot actions. Most importantly, it shows that with a small amount of robot training data, the system is able to transfer concepts embedded in its language and vision training data to direct robot actions — even for tasks it’s never been trained to do.

For example, if you wanted previous systems to be able to throw away a piece of trash, you would have to explicitly train them to be able to identify trash, as well as pick it up and throw it away. Because RT-2 is able to transfer knowledge from a large corpus of web data, it already has an idea of what trash is and can identify it without explicit training. It even has an idea of how to throw away the trash, even though it’s never been trained to take that action. And think about the abstract nature of trash — what was a bag of chips or a banana peel becomes trash after you eat them. RT-2 is able to make sense of that from its vision-language training data and do the job.

Advertisement

RT-2’s ability to transfer information to actions shows promise for robots to more rapidly adapt to novel situations and environments. In testing RT-2 models in more than 6,000 robotic trials, the team found that RT-2 functioned as well as our previous model, RT-1, on tasks in its training data, or “seen” tasks. And it almost doubled its performance on novel, unseen scenarios to 62% from RT-1’s 32%.

In other words, with RT-2, robots are able to learn more like we do — transferring learned concepts to new situations.

Not only does RT-2 show how advances in AI are cascading rapidly into robotics, it shows enormous promise for more general-purpose robots. While there is still a tremendous amount of work to be done to enable helpful robots in human-centered environments, RT-2 shows us an exciting future for robotics just within grasp. 

Advertisement

Tech

A former OpenAI leader says safety has ‘taken a backseat to shiny products’ at the AI company

A former OpenAI leader says safety has ‘taken a backseat to shiny products’ at the AI company

Published

on

By

A former OpenAI leader says safety has 'taken a backseat to shiny products' at the AI company

A former OpenAI leader who resigned from the company earlier this week said Friday that safety has “taken a backseat to shiny products” at the influential artificial intelligence company.

Jan Leike, who ran OpenAI’s “Superalignment” team alongside a company co-founder who also resigned this week, wrote in a series of posts on the social media platform X that he joined the San Francisco-based company because he thought it would be the best place to do AI research.

“However, I have been disagreeing with OpenAI leadership about the company’s core priorities for quite some time, until we finally reached a breaking point,” wrote Leike, whose last day was Thursday.

An AI researcher by training, Leike said he believes there should be more focus on preparing for the next generation of AI models, including on things like safety and analyzing the societal impacts of such technologies.

Advertisement

He said building “smarter-than-human machines is an inherently dangerous endeavor” and that the company “is shouldering an enormous responsibility on behalf of all of humanity.”

“OpenAI must become a safety-first AGI company,” wrote Leike, using the abbreviated version of artificial general intelligence, a futuristic vision of machines that are as broadly smart as humans or at least can do many things as well as people can.

Open AI CEO Sam Altman wrote in a reply to Leike’s posts that he was “super appreciative” of Leike’s contributions to the company was “very sad to see him leave.”

Leike is “right we have a lot more to do; we are committed to doing it,” Altman said, pledging to write a longer post on the subject in the coming days.

The company also confirmed Friday that it had disbanded Leike’s Superalignment team, which was launched last year to focus on AI risks, and is integrating the team’s members across its research efforts.

Advertisement

Leike’s resignation came after OpenAI co-founder and chief scientist Ilya Sutskever said Tuesday that he was leaving the company after nearly a decade.

Sutskever was one of four board members last fall who voted to push out Altman — only to quickly reinstate him. It was Sutskever who told Altman last November that he was being fired, but he later said he regretted doing so.

Sutskever said he is working on a new project that’s meaningful to him without offering additional details.

He will be replaced by Jakub Pachocki as chief scientist. Altman called Pachocki “also easily one of the greatest minds of our generation” and said he is “very confident he will lead us to make rapid and safe progress towards our mission of ensuring that AGI benefits everyone.”

On Monday, OpenAI showed off the latest update to its artificial intelligence m

Advertisement

Continue Reading

Tech

US, TikTok seek fast-track schedule, ruling by Dec. 6 on potential ban

US, TikTok seek fast-track schedule, ruling by Dec. 6 on potential ban

Published

on

By

US, TikTok seek fast-track schedule, ruling by Dec. 6 on potential ban

The U.S. Justice Department and TikTok on Friday asked a U.S. appeals court to set a fast-track schedule to consider the legal challenges to a new law requiring China-based ByteDance to divest TikTok’s U.S. assets by Jan. 19 or face a ban.

TikTok, ByteDance and a group of TikTok content creators joined with the Justice Department in asking the U.S. Court of Appeals for the District of Columbia to rule by Dec. 6 to be able to seek review from the Supreme Court if needed before the U.S. deadline. 

On Tuesday, a group of TikTok creators filed suit to block the law that could ban the app used by 170 million Americans, saying it has had “a profound effect on American life.”

Last week, TikTok and parent company ByteDance filed a similar lawsuit, arguing that the law violates the U.S. Constitution on a number of grounds including running afoul of First Amendment free speech protections.

Advertisement

“In light of the large number of users of the TikTok platform, the public at large has a significant interest in the prompt disposition of this matter,” the U.S. Justice Department and TikTok petitioners said.

TikTok said with a fast-track schedule it believes the legal challenge can be resolved without it needing to request
emergency preliminary injunctive relief.

The law, signed by President Joe Biden on April 24, gives ByteDance until Jan. 19 to sell TikTok or face a ban. The White House says it wants to see Chinese-based ownership ended on national security grounds, but not a ban on TikTok.

The parties asked the court to set the case for oral arguments as soon as practical during the September case calendar. The Justice Department said it may file classified material to support the national security justifications in secret with the court.

Earlier this week the Justice Department said the TikTok law “addresses critical national security concerns in a manner that is consistent with the First Amendment and other constitutional limitations.”

Advertisement

The law prohibits app stores like Apple and Alphabet’s Google from offering TikTok and bars internet hosting services from supporting TikTok unless ByteDance divests TikTok.

Driven by worries among U.S. lawmakers that China could access data on Americans or spy on them with the app, the measure was passed overwhelmingly in Congress just weeks after being introduced.

Continue Reading

Tech

Spotify sued over alleged unpaid royalties

Spotify sued over alleged unpaid royalties

Published

on

By

Spotify sued over alleged unpaid royalties

Music streaming giant Spotify has been sued in a US federal court for allegedly underpaying songwriters, composers and publishers by tens of millions of dollars.

The lawsuit against Spotify USA was filed in New York on Thursday by the Mechanical Licensing Collective (MLC), a non-profit that collects and distributes royalties owed from music streaming services.

The suit alleges that Spotify on March 1, without advance notice, reclassified its paid subscription services, resulting in a nearly 50 percent reduction in royalty payments to MLC.

“The financial consequences of Spotify’s failure to meet its statutory obligations are enormous for Songwriters and Music Publishers,” MLC said.

Advertisement

“If unchecked, the impact on Songwriters and Music Publishers of Spotify’s unlawful underreporting could run into the hundreds of millions of dollars.”

According to MLC, Spotify reclassified its Premium Individual, Duo and Family subscription streaming plans as Bundled Subscription Offerings because they now include audiobooks.

Royalties paid on bundled services are significantly less. MLC said Premium subscribers already had access to audiobooks and “nothing has been bundled with it.”

“Premium is exactly the same service that Spotify offered to its subscribers before the launch of Audiobooks Access,” it said. In a statement, Spotify said the lawsuit “concerns terms that publishers and streaming services agreed to and celebrated years ago.”

Spotify said it paid a “record amount” in royalties last year and “is on track to pay out an even larger amount in 2024.” “We look forward to a swift resolution of this matter,” the Swedish company said.

Advertisement

In February, Spotify said it paid $9 billion to musicians and publishers last year, about half of which went to independent artists. 

Continue Reading

Trending

Copyright © GLOBAL TIMES PAKISTAN