Feed aggregator

An anonymous reader quotes a report from TechCrunch: OpenAI's new Astra model will use a reasoning technique called "recurrent depth" that allows it to operate outside of the sequential thinking that characterizes most reasoning models, The Information reported on Tuesday. This technique, also called "opaque recurrence," will likely make the model's chain of thought more difficult to monitor -- and that has AI safety experts rattled. While Astra's use of the technique is reportedly limited, its emergence has still raised significant concerns among AI safety experts. "I am extremely concerned by the reporting that Astra uses opaque recurrence," wrote Redwood CEO Buck Shlegeris in a post after the news broke. "I don't know whether Astra is much less CoT monitorable than previous models. But if OpenAI pushes this technique further, they'll have the option to massively increase the recurrence and totally destroys CoT monitorability." Longtime AI safety advocate Zvi Mowshowitz also weighed in and wrote that laws might be necessary to prevent a "race to the bottom" among AI labs. "The technique is playing with fire, risking a taboo that OpenAI and Anthropic have fought to establish that we work hard to maintain Chain of Thought faithfulness and monitorability for as long as we can," Mowshowitz wrote. "More intensive use of such techniques would probably damage monitorability." [...] In a post responding to the news, Redwood Research chief scientist Ryan Greenblatt said opaque reasoning could easily scale faster than conventional chain-of-thought reasoning, effectively removing all reasoning from visible channels. "My biggest concern is that a natural progression from here would involve scaling up the opaque reasoning to the point where the model reasons entirely or almost entirely in latent space," Greenblatt wrote. "I hope it isn't too late to avoid the most concerning architectures and that OpenAI will stop here." Astra's use of the technique appears limited. "OpenAI has worked to preserve and utilize chain-of-thought monitoring since our very first reasoning models," wrote OpenAI chief scientist Jakub Pachocki. "It's a core goal of our current research program."

Read more of this story at Slashdot.

BrianFagioli writes: Audacity 4.0 has been released with a rebuilt Qt interface, a new clip-editing model, customizable workspaces, native high-DPI rendering, and a new AUP4 project format. The open source audio editor also adds ASIO playback and recording support to official Windows builds, while bringing major changes to recording, playback, plugins, and project handling. The overhaul comes with some tradeoffs. Audacity 4.0 currently lacks several features found in Audacity 3, including Time Tracks, Note and MIDI tracks, Mixer, Macro Manager and scripting pipe, VAMP and LADSPA plugin hosting, and play-at-speed. The developers say they plan to bring the missing functionality back in future releases.

Read more of this story at Slashdot.

The deal gives Nvidia control of a leading open-source AI platform and developer community.

Google released Gemini 3.8 Flash, its third Flash model in six weeks, claiming it's their best reasoning and coding model yet (topping the DeepSWE leaderboard). Ars Technica reports: Gemini 3.8 Flash comes in two variations. There's the standard Flash, which Google describes as a "workhorse" model that's good for anything from agentic tasks to software development. Then we have Gemini 3.8 Flash Cyber, which runs on the same foundations but has been tuned for vulnerability detection and mitigation. For developers, Google has the same pitch as it did for the 3.7 Flash release just a couple of weeks ago. API access to the model is available at an "introductory rate" through the end of the year: $0.75 per million input tokens and $3.75 per million output tokens. The regular price will be $1.50 / $7.50, but it's likely there will be new models available long before the price changes. Google probably sees the lower prices as a necessity given that other AI labs have recently dropped token pricing to keep increasingly wary businesses engaged with AI tools. [...] Gemini 3.8 Flash will be available across the Google ecosystem starting today, but Gemini 3.8 Flash Cyber is currently limited to trusted testers and governments. Like the past Flash release, you'll need a Pro or Ultra subscription to access Gemini 3.8 Flash in the Gemini app, but you can always visit AI Studio if you want to tinker with it for free.

Read more of this story at Slashdot.

Using Frore’s air-cooling technology, Lenovo’s Project AeroBlade Concept is an ultralightweight laptop with efficient heat management.

Morgan Hewett noticed that many single women around her were creating AI companions as well as buying and using vibrators. She is combining the two with a vibrator that pairs physical sensors with artificial intelligence to "engage the mind."

An anonymous reader quotes a report from Wired: Michele Spagnuolo, the Google engineer arrested in May by U.S. authorities for alleged insider trading on Polymarket, is making a new bold bet. On Wednesday, his legal team filed a motion to dismiss the charges against him. Spagnuolo isn't outright denying that he made money using internal information from Google. Instead, his legal team says that the wagers were not financial instruments subject to regulation by the United States' Commodities Exchange Act but rather good old-fashioned international betting that the U.S. has no authority over. Spagnuolo, who has been placed on leave from Google, is accused of committing commodities fraud, wire fraud, and money laundering. Using the alias "AlphaRaccoon," he allegedly made a series of wagers on Polymarket's flagship platform that resulted in profits totaling over $1.2 million. According to the criminal complaint, "AlphaRaccoon" correctly wagered that the singer D4vd, who gained notoriety for his suspected connection to a grisly killing, would be Google's most-searched person of the year in 2025. (D4vd was later charged with murder; he pleaded not guilty.) [...] Spagnuolo's lawyers argue that defining swaps to include wagers like who the most-searched person on Google will be each year "would fly in the face of the statute's purpose and history" and lead to "absurd results." They say it would make it so that any wager in the world, from a charity raffle to a local Ping-Pong match, could be classified as a financial instrument. "Spagnuolo is basically making the same argument as the states that are suing prediction markets," says a financial services regulation expert Todd Phillips. "This is the issue that will likely go up to the Supreme Court." Featured Video In addition to disputing the idea that prediction markets offer swaps, Spagnuolo's legal team argues that the U.S. government had no jurisdiction over him in the first place because he's a non-U.S. citizen who was wagering on a non-U.S. platform. Although Polymarket is headquartered in New York, the company's flagship prediction market is banned in the United States and technically is administered by an ostensibly Panama-based entity known as Adventure One QSS. Spagnuolo was living in Zurich, Switzerland, when he allegedly made the Google-related trades on Polymarket. "The extraterritorial argument is interesting and raises the question of whether the U.S. should be the world's prediction markets cop," Philipps says. Spagnuolo's team also claims that the charges should be dismissed because the internal information he supposedly leveraged did not have any commercial value to Google.

Read more of this story at Slashdot.

At IFA 2026, Nvidia and its partners showed off the first RTX Spark-powered laptops and mini PCs, designed to run AI models right on your computer.

BrianFagioli writes: Perplexity has built a local artificial intelligence engine designed specifically for Apple silicon and the Qwen3.6-35B-A3B model. Called Lily, the engine uses a Rust runtime and custom Metal kernels, with neither PyTorch nor MLX in its execution path. Perplexity says Lily averaged 23 percent faster prompt processing and 35 percent faster token generation than MLX-LM on an M5 Max MacBook Pro with 128GB of unified memory. Lily is more specialized than MLX-LM, which supports a much wider range of models and architectures. Perplexity says it plans to release Lily as open source, but the code is not available yet, leaving its performance claims dependent on internal testing for now.

Read more of this story at Slashdot.

Restaurant menus with AI-generated food images are going viral for their off-putting aesthetic. But AI’s impact on modern menus runs even deeper.

The game starts as a reheated Witcher 3, but evolves into something much more focused.

These WIRED-tested indoor and outdoor air quality monitors give you insights into your home’s air quality, warn of potentially harmful pollutant levels, and help keep tabs on your environment.

Maine could be a bright spot compared to national harvests this year.

The BBC's Zoe Kleinman takes a ride on Uber's self-driving robotaxi as the service rolls out on Thursday.

"Enshittification of car controls [is] rapidly accelerating," writes longtime Slashdot reader sinij, pointing to a new report from Car and Driver. From the report: Remember when iPhones used to come with free headphones and a phone charger (including the wall plug)? It didn't feel so much like Apple giving you free goodies as it did that the company was providing you with the relevant hardware to use the device. Apple stopped including headphones and charging blocks in 2020. Now, Hyundai is pulling some of its standard hardware from the box, at least for two models. Hyundai is charging customers extra for a driver's display in the new Elantra generation (more specifically, the Korea-market Avante), as well as the Ioniq 3, Motor1 reported. Both models feature Pleos Connect, Hyundai's new infotainment setup that pairs a center touchscreen with a slim 9.9-inch instrument cluster mounted above the dashboard -- except where it doesn't. In its domestic market, the instrument cluster screen is offered as a 350,000 won ($255) option for the base trim. Not the most expensive optional extra in the world, but still kind of a slap in the face for a feature traditionally viewed as standard fare. Things are more expensive for the electric Ioniq 3. In the EV's case, the base trim gives customers the full Tesla-screen experience, meaning if customers want the driver's display, they'll need to fork over the additional $5000 necessary to move up to the next-level trim. [...] Hyundai plans to have the setup equipped in 20 million cars globally by the end of the decade.

Read more of this story at Slashdot.

Stream your favorite shows and save up to 50% today with HBO Max discount codes and subscription deals.

Learn how to maximize your savings on major appliances, electronics, and mattresses using verified ABT discount codes, rebates, and special coupons and save up to $1,200 this summer.

Learn how to get the best value from your ExpressVPN subscription plans, whether you need a flexible 1-month plan or want to lock in up to 61% savings on multi-year deals.

An anonymous reader quotes a report from The Guardian: Global heating will reach at least 1.8C under even the most optimistic future, well beyond the Paris agreement goal of 1.5C, according to a UN report that warns every fraction of temperature rise intensifies destructive extreme weather, glacier melt, ecosystem loss, and island and coastal city submersion. The report by the Nairobi-based UN Environment Program confirmed overshooting the 1.5C goal inscribed in the landmark Paris agreement of 2015 was now "unavoidable" and, despite some progress in addressing the human-caused climate crisis driven by burning fossil fuels, likely in the next few years. It said: "There are no good outcomes above 1.5C." Heating of up to 3C above preindustrial levels could lead to glaciers losing more than a quarter of their mass by 2100, raising sea levels by up to 13cm. Global food production could decline by up to 14% by 2050 if there are not effective strategies to adapt. Human health, water supplies, nature, cities, infrastructure and economies could all be severely damaged. Some losses would be irreversible. Many communities may have to relocate or change their livelihoods. The report said the best hope for humanity to limit damage was to adopt an "overshoot, peak and decline" pathway that required immediate and sustained greenhouse gas emissions cuts combined with steps to remove carbon dioxide from the atmosphere. It described the goal of net zero emissions -- increasingly politically contentious in some countries -- as "an essential milestone that cannot be skipped" and stressed carbon dioxide removal through steps such as establishing vast new forests must occur alongside, not as an alternative to, deep cuts in fossil pollution. Crucially, the authors of the report, titled Limiting Overshoot, said the average global temperature could be returned to 1.5C this century only if heating stayed below about 1.8C. They warned nature's capacity to store carbon was uncertain and would shrink the more the planet heats. [...] The report's authors cited earlier work that found limiting heating to about 1.8C required global emissions to be halved by 2035. They said existing national policies were projected to lead to at least 2.3C heating, but it would still be possible to limit stay below 2C if countries delivered on net zero emissions commitments by mid-century.

Read more of this story at Slashdot.

The ride-hailing app has launched 15 self-driving taxis, but they will have safety drivers inside.