Feed aggregator

Anthropic just released a new model called Claude 3.7 Sonnet, and while I'm always interested in the latest AI capabilities, it was the new "extended" mode that really drew my eye. It reminded me of how OpenAI first debuted its o1 model for ChatGPT. It offered a way of accessing o1 without leaving a window using the ChatGPT 4o model. You could type "/reason," and the AI chatbot would use o1 instead. It's superfluous now, though it still works on the app. Regardless, the deeper, more structured reasoning promised by both made me want to see how they would do against one another.

Claude 3.7’s Extended mode is designed to be a hybrid reasoning tool, giving users the option to toggle between quick, conversational responses and in-depth, step-by-step problem-solving. It takes time to analyze your prompt before delivering its answer. That makes it great for math, coding, and logic. You can even fine-tune the balance between speed and depth, giving it a time limit to think about its response. Anthropic positions this as a way to make AI more useful for real-world applications that require layered, methodical problem-solving, as opposed to just surface-level responses.

Accessing Claude 3.7 requires a subscription to Claude Pro, so I decided to use the demonstration in the video below as my test instead. To challenge the Extended thinking mode, Anthropic asked the AI to analyze and explain the popular, vintage probability puzzle known as the Monty Hall Problem. It’s a deceptively tricky question that stumps a lot of people, even those who consider themselves good at math.

The setup is simple: you're on a game show and asked to pick one of three doors. Behind one is a car; behind the others, goats. At a whim, Anthropic decided to go with crabs instead of goats, but the principle is the same. After you make your choice, the host, who knows what’s behind each door, opens one of the remaining two to reveal a goat (or crab). Now you have a choice: stick with your original pick or switch to the last unopened door. Most people assume it doesn’t matter, but counterintuitively, switching actually gives you a 2/3 chance of winning, while sticking with your first choice leaves you with just a 1/3 probability.

Crabby Choices

With Extended Thinking enabled, Claude 3.7 took a measured, almost academic approach to explaining the problem. Instead of just stating the correct answer, it carefully laid out the underlying logic in multiple steps, emphasizing why the probabilities shift after the host reveals a crab. It didn’t just explain in dry math terms, either. Claude ran through hypothetical scenarios, demonstrating how the probabilities played out over repeated trials, making it much easier to grasp why switching is always the better move. The response wasn’t rushed; it felt like having a professor walk me through it in a slow, deliberate manner, ensuring I truly understood why the common intuition was wrong.

ChatGPT o1 offered just much of a break down, and explained the issue well. In fact, it explained it in multiple forms and styles. Along with the basic probability, it also went through game theory, the narrative views, the psychological experience, and even an economic breakdown. If anything, it was a little overwhelming.

Gameplay

That's not all Claude's Extended thinking could do, though. As you can see in the video, Claude was even able to make a version of the Monty Hall Problem into a game you could play right in the window. Attempting the same prompt with ChatGPT o1 didn't do quite the same. Instead, ChatGPT wrote an HTML script for a simulation of the problem that I could save and open in my browser. It worked, as you can see below, but took a few extra steps.

(Image credit: Anthropic)

While there are almost certainly small differences in quality depending on what kind of code or math you're working on, both Claude's Extended thinking and ChatGPT's o1 model offer solid, analytical approaches to logical problems. I can see the advantage of adjusting the time and depth of reasoning that Claude offers. That said, unless you're really in a hurry or demand an unusually heavy bit of analysis, ChatGPT doesn't take up too much time and produces quite a lot of content from its pondering.

The ability to render the problem as a simulation within the chat is much more notable. It makes Claude feel more flexible and powerful, even if the actual simulation likely uses very similar code to the HTML written by ChatGPT.

You might also like

• Claude will eventually start speaking up during your chats
• You can now ask Claude to mimic your writing style
• I tried Perplexity’s Deep Research and it doesn't quite live up to ChatGPT's research potential

Stubbed toes, Lego landmines and pet 'presents' can be easily avoided with smart lights and a little automation magic.

Federal authorities are investigating a near-collision at Chicago's Midway airport between a Southwest 737 and a small business jet. The 737 was landing when the business jet entered the runway.

(Image credit: StreamTime Live)

• Luma Labs’ Dream Machine can now add audio to video clips for free
• You can prompt the audio or let the AI come up with something it decides is appropriate
• Sora and other AI video makers mostly lack even an imperfect AI audio creator to go with their visuals

Luma Labs has added a score to the AI videos produced on its Dream Machine platform. The new feature brings audio to your video, custom-generated to match a written prompt or created by the AI, and is based solely on what's happening in the video. That could mean chirping birds at the sunrise scene, a spaceship’s distant hum for your sci-fi animation, the chaotic clatter of a busy café, or anything else you care to hear.

The new feature is free in beta for all users. After generating a video with Dream Machine, you’ll see a new “Audio” button along the row at the bottom of the video next to the existing "Extend" and "Enhance" buttons. Click it, and you get two choices: let the AI decide the best fitting sounds on its own, or take the wheel and provide a text prompt describing exactly what you want. Maybe you’ve got a dreamy nature scene and want to hear a distant waterfall, or maybe you want to hear how the AI does it; either way, it works.

Sound Idea

This update is big because AI-generated videos, while sometimes visually stunning, have always felt incomplete without sound. It's a lot of work to painstakingly add audio yourself. Even some of the biggest names in AI video don't have audio as an option yet, including OpenAI's Sora.

Of course, AI sound generation on its own isn't unique. There are a lot of AI music makers, even full voice and song producers. But, the production within the platform linked to the video already there makes Dream Machine a real standout. That said, it isn't perfect. You can tell from the way the motion and sound don't quite match with this dog as it swims.

On the other hand, when prompted correctly, this crackling fire and laughter of people around it sounds pretty good.

But, I wouldn't rely on Dream Machine to create sound on its own without any guidance in a prompt. With a blank audio prompt, the AI took the same short clip of people around a fire and came up with something a lot spookier.

You might also like...

• This AI video generator can make a banana typing look realistic – and might challenge Sora
• Dream Machine 1.5 catches Sora and other rival AI video makers napping
• OpenAI Sora officially launches to change AI video – 5 things you need to know

The Copilot Voice and Think Deeper AI features are for verbal conversations and more complicated queries, and are now unlimited under its free tier.

• SanDisk reveals E61 8TB Portable SSD
• New device offers drop protection up to 3 meters, with silicone shell improving durability and enhances tactile grip
• There's also fast read/write speeds for large file transfers

SanDisk has launched an 8TB version of its popular E61 portable SSD, expanding its offerings for users who require extensive storage capacity.

Thr new model aims to cater to professionals such as video editors, photographers, and data analysts, who often handle large files and need reliable storage solutions.

However, despite the excitement around the 8TB capacity, concerns linger about the reliability of SanDisk’s SSDs due to a major data corruption issue that surfaced in 2024.

Bigger - but also better?

The 8TB SanDisk E61 comes with a compact and lightweight form factor measuring 100.8 x 52.55 x 9.6 mm, meaning it is quite portable, and features a silicone shell that offers protection from drops of up to 3 meters, along with an IP45 rating, providing resistance against dust and water, making it ideal for outdoor use or travel.

Its USB-C 3.2 Gen 2 interface boasts read speeds of up to 1500MB/s and write speeds of 1000MB/s, ensuring fast data transfers for large files. Furthermore, it has a plug-and-play feature which means users can start using the SSD immediately without the need for additional drivers or software. It also includes 256-bit AES hardware encryption, ensuring data security for sensitive information, whether for personal or professional use.

While the new 8TB SanDisk E61 offers a compelling set of features, concerns remain about the product’s reliability. In 2023, SanDisk’s portable SSDs, including the SanDisk Extreme and Extreme Pro models, were plagued by a major firmware issue that caused widespread data corruption and drive failures. Users reported losing access to critical data, with the drives suddenly becoming unreadable. A class-action lawsuit was filed, accusing Western Digital (SanDisk's parent company) of failing to address the issue adequately.

In response, Western Digital issued a firmware update to mitigate the problem, but the lawsuit claimed that the core issue remained unresolved. As a result, many users, particularly professionals handling large volumes of important data, continue to worry about the long-term reliability of SanDisk SSDs.

Nevertheless, with this new 8TB version, SanDisk has an opportunity to restore customer confidence by ensuring that the previous data corruption issues have been fully addressed. The high price point of around $714 makes it a premium product, especially for professionals who depend on safe, secure, and reliable data storage.

You might also like

• A GPU or a CPU with 4TB HBM-class memory? Nope, you're not dreaming, Sandisk is working on such a monstrous product
• Samsung HBM roadmap shows Google will be Nvidia's fiercest competitor in AI
• Sandisk quietly puts 2TB SD card on sale - Extreme Pro is as fast as a hard drive

The wrap-around physical keyboard cases are coming for leading Android phones, including the Motorola Razr foldable.

U.S. District Judge Amir H. Ali said officials have provided no evidence of compliance with repeated orders to unfreeze the money.

(Image credit: Jam Sta Rosa/AFP via Getty Images)

• Advanced Voice Mode is coming to all ChatGPT free users
• There will be a daily limit on usage
• Deep Research is being released to Plus, Team, Edu, and Enterprise users

OpenAI has just announced, via X, that it is starting to roll out a “preview” version of Advanced Voice mode for ChatGPT free users while also rolling out its Deep Research agent to all Plus, Team, Edu, and Enterprise users.

Advanced Voice Mode, which is currently only available to ChatGPT Plus users, launched initially in the mobile app versions of ChatGPT and arrived in the desktop app version of ChatGPT in November last year. It is one of the nicest features of ChatGPT; it’s a way to communicate with the chatbot using your voice in a free-flowing, natural conversation. It’s almost like talking to a real person, and you have the ability to interrupt the chatbot if you find its reply is going on too long. There are a variety of different voices to choose from too, so you can customize the experience.

OpenAI has previously experimented with offering 10 minutes of Advanced Voice Mode a month to ChatGPT free users, but the new rollout is going to “give all ChatGPT free users a chance to preview it daily across platforms." The company is also being a bit secretive about what the daily limit is for Advanced Voice Mode for free users, as it clearly wants to retain the ability to adjust it depending on demand. The only detail on usage it offers is that ChatGPT Plus users will get “5x the free limit."

Starting today, we’re rolling out a version of Advanced Voice powered by GPT-4o mini to give all ChatGPT free users a chance to preview it daily across platforms.The natural conversation pace and tone are similar to the GPT-4o version while being more cost effective to serve.February 25, 2025

ChatGPT 4o-mini-powered

The ChatGPT free version of Advanced Voice Mode will be powered by ChatGPT 4o-mini, while Plus users will continue to have access to Advanced Voice Mode powered by ChatGPT 4o. In its statement, OpenAI said: “Starting today, we’re rolling out a version of Advanced Voice powered by GPT-4o mini to give all ChatGPT free users a chance to preview it daily across platforms. Plus users will continue to have access to Advanced Voice powered by 4o with the existing daily rate limit, which is more than 5x the free limit, as well as access to video and screensharing in Advanced Voice.”

Reacting to the news some X users expressed concern that the 4o-mini model might be “dumbed down” and expressed frustration that the daily limit remained in place for ChatGPT Plus subscribers “We’re paying for the best, not a crippled version. Get it together”, said X user Emanuele Dagostino.

Gemini Live, Google's voice mode chatbot, is entirely free for Android users.

Advanced Voice Mode in the ChatGPT Mac app. (Image credit: OpenAI) Deep Research

At the same time, OpenAI is rolling out its Deep Research agent tool to all its paid subscribers, rather than just its Pro subscribers. Built using the o3 model, Deep Research is a tool for carrying out in-depth research using the Internet that drastically reduces the time taken by researchers.

The o3 model is optimized for data analysis and can handle text, images, and PDF files that it can access via the web.

Deep Research can work independently. You simply give it a prompt, and it goes off and analyzes and synthesizes hundreds of online sources for you, reducing a job that would take human researchers many hours to a few minutes.

You might also like

• I’m never going to use voice controls for my tech, sorry - and I don’t care how much better it is now thanks to AI
• I cloned my voice with AI and even my wife can’t tell the difference
• Google Gemini Live's AI voice now comes in ten more styles that take inspiration from the stars

Taurasi leaves her basketball career as the most decorated woman to ever play, with three WNBA titles, three NCAA titles and six Olympic gold medals to her name.

(Image credit: Christian Petersen)

Folk musician Rhiannon Giddens said on social media that she has moved her May concert – originally scheduled for the Kennedy Center – to a different venue in Washington, D.C.

(Image credit: Rob Kim)

• OpenWrt launches "unbrickable" security-focused wireless router
• Dual-band Wi-Fi 6 connectivity ensures fast, reliable internet access
• Open-source firmware guarantees no lock-in or obsolescence risks

OpenWrt has officially launched its first dedicated wireless router that promises unrivalled security protection.

Developed in collaboration with Banana Pi, the OpenWrt One is specifically designed to meet the needs of hackers and security-conscious users.

Available for just $89, the OpenWrt One promises to offer a long-lasting, open, and flexible solution for users who value control and security in their networking hardware.

A router built for the OpenWrt community

The OpenWrt One is powered by the MediaTek MT7981B system-on-chip (SoC), supports dual-band Wi-Fi 6 with the MediaTek MT7976C chipset, and offers 2x2 2.4 GHz and 3x3 5 GHz wireless connectivity. For wired connections, the router includes one 2.5 Gbit WAN port and one 1 Gbit LAN port, ensuring fast, reliable internet access.

In terms of storage and expandability, the OpenWrt One features 128MB of SPI NAND and 16MB of SPI NOR flash memory, with additional storage options through a 2230/2242 NVMe PCIe 2.0 slot. It also includes a USB 2.0 Type-A port for connecting peripherals and a USB Type-C port for power delivery and serial interface, providing ample options for customization.

The router's hacker-friendly design includes a mechanical switch for boot selection between NAND and NOR flash, as well as a mikroBUS expansion slot for additional hardware add-ons. Fully compliant with FCC, EC, and RoHS standards, the OpenWrt One is built with security and open source innovation at its core.

According to OpenWrt, what sets OpenWrt One apart from other routers is its dedication to being open and secure. Software Freedom Conservancy (SFC), a group that worked with OpenWrt on the project claims that it comes with OpenWrt’s open-source firmware which guarantees that it will never be locked and it is designed to be "forever unbrickable." This means users can experiment with different configurations, flash new firmware, and make modifications without the risk of rendering the device unusable.

The OpenWrt One launch comes at a time when many older routers and network devices are becoming obsolete due to unpatched vulnerabilities. Manufacturers often abandon support for these devices, leaving users with few options other than replacing hardware. With OpenWrt, users can continue using their existing devices by flashing the open source firmware, extending the life of their hardware and improving security.

Many users are turning to open source solutions like OpenWrt as manufacturers stop supporting older models. For instance, D-Link recently refused to patch critical security flaws in over 60,000 NAS devices, effectively turning them into e-waste. In contrast, the OpenWrt One is designed to avoid these issues, giving users peace of mind that their router will remain secure and usable for years to come.

You might also like

• 'Revolutionary' Wi-Fi router which can send data up to 10 miles away goes on sale for less than $100
• Check out our pick of the best secure routers on offer
• Wi-Fi routers that can send 250Mbps across 10 miles (yes, 10 miles) have been demoed and I'm very excited

Although there aren't many internet provider options in Springfield, we found the best fast and affordable plans in the area.

Add rich, smoky flavor to your grill game with a great pellet grill smoker.

Here are some hints -- and the answers -- for Strands No. 360 for Feb. 26.

The men told NPR they were kept in the dark about why they were in Guantánamo Bay, and were denied access to an attorney or a phone call with loved ones.

(Image credit: PEDRO MATTEY)

During the Assad regimes, Western nations placed crippling economic sanctions on the country. Some have been lifted, but most are still in place, so how does a country shattered by civil war rebuild? And we follow the return to Damascus of one US Syrian Jewish leader, after decades in exile.

• Nvidia's ConnectX-8 features a GPU-inspired design for networking
• ConnectX-8 delivers an impressive 800Gbps throughput capability
• Requires PCIe Gen6 x16 for optimal performance

Nvidia has unveiled the ConnectX-8 SuperNIC, a new card whichcomes with an 800Gbps capability, doubling the 400Gbps recorded for its predecessor.

The design of this card is a departure from conventional network interface cards (NICs) as it looks more like traditional GPUs, with the ConnectX-8 seemingly set to focus on enhancing airflow and cooling efficiency with its low-profile design, backplate and advanced internal layout.

This device also comes with a large connector on the back, which suggests the potential for multi-host cable connections, possibly linking to additional CPUs or functioning as a PCIe switch output.

Multi-host connections and flexibility

The ConnectX-8 SuperNIC is a single-port card that delivers high bandwidth which necessitates advanced PCIe connectivity.

Specifically, it requires either PCIe Gen6 x16 or two Gen5 x16 links to operate effectively, suggesting the card's performance exceeds what a single CPU can handle. This capability aligns with the need for robust connectivity in Nvidia's Grace platforms, where the ConnectX-8 serves as a critical component for auxiliary connections due to the limitations of Grace CPUs.

By integrating these NICs, Nvidia minimizes reliance on Broadcom products, enhancing system efficiency and flexibility.

Unlike older NIC designs that typically feature bulky heatsinks and a more utilitarian aesthetic, the ConnectX-8 showcases a sleek backplate and overall factor reminiscent of modern GPUs.

The introduction of the ConnectX-8 does not seem to be merely about aesthetics, as it suggests Nvidia has a broader vision for AI infrastructure. By aligning its networking products with GPU-like designs, the company likely aims to streamline its integration within data centers while delivering high-performance capabilities essential for AI workloads.

You may also like

• Intel Xeon 6 processors look to take the fight to Nvidia
• Comino has workstation PCs with 8, yes 8 Nvidia RTX 5090 GPUs, for less than $60,000
• A GPU or a CPU with 4TB HBM-class memory? Nope, you're not dreaming

President Trump's calls for the U.S. to take over Greenland have sparked alarm and outrage.

Denmark, which is responsible for Greenland's security, recently announced that it would further boost its defense spending.

And a recent poll found 85 percent of Greenlanders are opposed to being part of the United States.

Parliamentary candidate Naaja Nathanielsen is one of them.

It isn't clear whether Trump's ambitions for Greenland will take. But some politicians in the territory are taking his calls for acquisition more seriously than ever before.

For sponsor-free episodes of Consider This, sign up for Consider This+ via Apple Podcasts or at plus.npr.org.

Email us at considerthis@npr.org.

With a final vote fast approaching, GOP leaders were still working to wrangle support from inside the party for a sweeping multitrillion plan to address defense, energy, immigration and tax policy.

(Image credit: Andrew Harnik)