Oura’s next-gen ring appears to be an upgrade in every way. It features a sleeker design, longer battery life and smarter sensors to offer deeper insights for wearers. No more squarish edges; it’s a perfect ring this time. Oura says Ring 4 has 18 signal pathways, up from eight in the Gen3, which is paired with its new Smart Sensing algorithm.

New features include automatic heart rate and activity detection for up to 40 activities. It’ll land in six colors: Silver, Brushed Silver, Gold, Rose Gold, Stealth and Black. All bar the Stealth is made of titanium. Oura Ring 4 is available to pre-order today, with shipping expected to begin on October 15, 2024. Prices start at $349.

— Mat Smith

The biggest stories you might have missed

OpenAI rakes in over $6 billion in new funding

Bose QuietComfort Earbuds review: A more affordable ANC ace

Amazon tablets are getting AI tools, like writing assist and website summaries

Through an Instagram stream and publicly available information.

Two Harvard students used facial recognition tech and a large language model to unearth a subject’s name, occupation and other details. Their setup (dubbed I-XRAY) can use that information to pull data like addresses, phone numbers, family member details and partial social security numbers from various sources online.

In a demo video, AnhPhu Nguyen and Caine Ardayfio used the glasses to address people who appear to be strangers by name, discuss their work and bring up a place where they may have met in the past, based on information picked up. The students told 404 Media they developed I-XRAY to make people aware of what’s possible with current technology — they won’t release the code they used.

Continue reading.

The long-awaited SUV is finally here.

Following its first two cars of the same name, the Polestar 3 delivers what many of us were looking for. It’s a mass-market machine to fill the needs and wants of buyers looking for an all-electric SUV with proper dimensions — and yellow seat belts. After all the delays, we’ve got a test drive.

Continue reading.

The new minimum price for a Model 3 is $42,500.

Tesla’s Model 3 Standard Range Rear-Wheel Drive is no longer available in its online configurator. Electrek first reported on the absence of the cheapest option from the electric vehicle brand, with a price tag of $39,000. Now the Model 3 with Long-Range Rear-Wheel Drive takes that title with a retail price of $42,500.

Continue reading.

Researchers from Stanford University‘s Scaling Intelligence Lab introduced a new inference framework that could help large language models (LLMs) go through potential responses faster. 

The framework, Archon, uses an inference-time architecture search (ITAS) algorithm to improve LLMs performance without additional training. It is model agnostic, open-source and designed to be plug-and-play for large and small models. 

Archon could ideally help developers design AI model systems using multiple inference-time techniques to cut down on models to determine responses. The Scaling Intelligence Lab said techniques like Archon would help cut down on costs related to building models and inference. As LLM development turns toward larger parameters or more advanced reasoning, costs could increase despite companies like OpenAI anticipating more affordability. 

According to the researchers, Archon automatically designs architectures that improve task generalization, enabling models to perform tasks beyond those they were initially trained on.

“Our Archon framework and ITAS algorithm draw inspiration from neural architectures and neural architecture search, respectively,” the researchers said in their paper. “Archon is constructed of layers of LLMs, in which models in the same layer run in parallel but each later runs sequentially.” 

These layers perform different inference-time techniques, “either transforming the number of candidate responses through generation and fusion (like linear transformations) or reducing the number of candidate responses to improve quality (like non-linearities).” 

Instant wants to recapture some of the magic of the pre-Google-acquisition Firebase by building a modern, Postgres-based open source version of it.  

Like the original Firebase, San Francisco-based Instant focuses on giving front-end developers easy access to a real-time database with offline capabilities without the need to manage any of the back-end architecture. The service supports applications written in React, React Native, and vanilla JavaScript. In addition to the database, it also offers tools to manage authentication and permissions, as well as real-time features like shared cursors, presence, and more. 

The team recently open sourced the project. 

Instant was founded by former roommates Joe Averbukh (CEO) and Stepan Parunashvili, who previously worked at companies like Wit.ai, Facebook, Airbnb, and others. 

At Facebook, Averbukh told me, the architecture was set up to ensure that as a front-end engineer, he was able to move quickly. Meanwhile, at Airbnb, the company’s intricate microservices architecture meant it took a long time to try a new idea. “Thankfully, we already had Firebase inside Airbnb, and we could use it for notifications. So we would actually just use Firebase to prototype these ideas — and that was a huge win,” he explained. 

Before starting Instant, the co-founders worked on a fitness app, based on Google’s Firebase. While they loved all of the features like optimistic updates and offline mode that Firebase offers out of the box, they missed having access to a relational database since Firebase’s focus is on its NoSQL database. 

“You see apps like Figma, Notion, and Linear — they’re winning the market because of this feature set that they have,” Averbukh said. “And we think that basically, in the same way that apps today are much better than apps before, the expectations are going to rise for what people want in their apps. But the reality is, the way that people build these features today is that they have this whole team that’s building all of these sync services.”

At the core of all of this is Aurora, AWS’s relational database service, which is cost-effective and allows Instant to quickly spin up new databases inside of a multi-tenant architecture that still offers strong security guarantees. Averbukh argues that competitors like Supabase have to spin up a new database for every project (and free user), while Instant essentially lives in one large Postgres instance. He also stressed that if a large enterprise customer would sign up for the service, the team could offer them to spin them up a separate database.

On Wednesday, the company announced that it has raised a $3.4 million seed funding round. It’s backed by Y Combinator and SV Angel, as well as prominent angel investors like former Firebase CEO James Tamplin, Paul Graham, Greg Brockman, and Jeff Dean.

First, it was Spotify, then Verizon, then PlayStation, and now it’s apparently Bank of America’s turn to have an outage this week. Some customers reported problems accessing their bank accounts on Wednesday, with the balances showing $0 or “—-” instead of what should be there.

“Some mobile and online banking clients experienced an issue accessing their accounts and balance information earlier today,” writes Bank of America media relations executive Matt Card in an email to The Verge. “These technology issues have been fully resolved. We apologize for any inconvenience.”

Reports spiked just before 1PM ET on Downdetector, with posts across X and Reddit from people reporting that their account balances don’t show anything (some mentioned that the amounts they owe, however, appear to be displaying accurately).

We verified this afternoon that the bank’s app was showing customers a notification that “Accounts temporarily unavailable” and that “Some accounts and/or balances are temporarily unavailable,” as reported previously by CNN. Some people have reported their accounts are working or intermittently gaining access to their information, and it’s unclear how many people have been affected. In July, Bank of America said 58 million clients use its “digital capabilities to help manage their financial lives” and that they connected a record 23.4 billion times last year.

There’s no word on what might have caused the problem, but let us know if things are working for you now.

Update, October 3rd: Added updated Bank of America statement.