Connect with us

Tech

One YouTuber’s Bare-Chip Cooling Gamble Pours Water Straight Onto Silicon

Published

on

Pour Water Silicon GPU Cooling
TrashBench decided ordinary water blocks were optional. In a recent experiment the channel stripped everything away and simply pumped water across the exposed dies of a graphics card and a processor. A 3D-printed plastic fitting, some epoxy, and a willingness to risk hardware produced results that surprised even the person running the tests.



Work began by attempting to seal the leaks on a dead RTX 3060 card. Early attempts to use translucent plastic prints proved ineffective, as water seeped straight thru the layers. Adding some nail polish and epoxy helped, but it wasn’t until he switched to black PETG as well as adjusted the print settings that he was able to get the part to maintain pressure. Once he had the chamber fixed and it wasn’t leaking everyplace, he moved on to a live GTX 980. Water flowed directly over the exposed silicon, but to his surprise, it did not destroy the card. It turned out that the chamber on that older GPU was much larger, which helped keep the temperatures lower than he’d expected for such a basic setup.

Pour Water Silicon GPU Cooling
A fair comparison with an RTX 2060 SUPER was finally possible, and the results were rather interesting. The card operated at a high temperature of 70 °C under load with stock air cooling, but when we installed a standard AIO cooler, it decreased to around 36 °C. Then he installed the 3D printed direct water connection and, well, you could say it blew the competition away in both Unigine Heaven and Cyberpunk 2077, resulting in a decisive victory for the homemade block over the commercial one. Of course, there was no metal in the way, so the heat just passed into the water running past it. It’s worth noting that while the water warmed up over time, the CPU remained cooler than the polished copper block throughout testing.

Pour Water Silicon GPU Cooling
Next, he tried sub-zero coolant, and while temperatures did drop, the difference was not as dramatic as switching from air to room-temperature water. It seems that the size of the chamber and the contact area made a far bigger difference than how cold the liquid was. He was able to adapt the same concept to operate with a delidded Intel i5-7600K, but the small die and limited mounting space made it much more difficult to fit into a smaller chamber. Unfortunately, he realized that the majority of the heat was concentrated in a small area, and the water just couldn’t keep up. The GPU’s huge die was not an issue because the heat was uniformly distributed across it, but the CPU’s dense hotspot proved difficult to keep cool.

Pour Water Silicon GPU Cooling
Leaks continued to appear as water escaped around the fittings, and he came close to losing the hardware several times. Putting extra epoxy on the broken parts helped to patch things up, but the risk was evident. It’s evident that direct liquid contact can work; immersion cooling has been available for years, but the large difference in outcomes between GPU and CPU was surprising. The GPU is a perfect choice for this type of cooling due to its larger surface area and lower heat density, but doing the same method on a compact processor die is difficult to keep cold.
[Source]

Source link

Advertisement
Continue Reading
Click to comment

You must be logged in to post a comment Login

Leave a Reply

Tech

Yes, You Can Replace A Stihl Weed Eater Head

Published

on





Keeping your outdoor equipment in good condition can involve all sorts of tasks, one of which might be replacing the head on a Stihl weed eater. Thankfully, it’s a fairly straightforward process. First, turn the weed eater off and cut the power by disconnecting the spark plug, removing the battery, or unplugging the unit, depending on the model. Then, put on some work gloves for safety before you begin.

Of course, you should first have the correct replacement head approved for your specific Stihl weed eater, as not every replacement head is compatible with every model. Using a head that doesn’t fit can affect safe operation and performance. Also, if you’re using the AutoCut C 25-2 or PolyCut C 20-3, you’ll need to lock the drive shaft with a stop pin before removing the old head and installing the new one.

Advertisement

From there, the process depends on the replacement head you’re using. The AutoCut C 25-2 and PolyCut C 20-3 heads thread onto the shaft before you tighten them. The AutoCut C 5-2 uses a different mounting design that requires installing and securing the tap head. Once the new head is attached, check that it’s on tightly before using the trimmer. If you have any questions or run into any issues while replacing the head, consult your owner’s manual.

Advertisement

When and why should you replace a weed eater head?

Knowing when to replace the head on your Stihl weed eater depends on a few different factors. First, you should consider how often you actually use the tool. Homeowners who do yard work once a week may not need to replace the head as soon as a full-time landscaper would. The conditions you’re working in matter as well: thick grass, weeds, and other demanding jobs can cause the head to wear faster. A worn-out head may also lead to a common problem with Stihl weed eaters, where bumping the head doesn’t pull out fresh string.

If you don’t replace a worn or damaged head and continue to use it, whether it’s a Stihl or any other highly-rated weed eater, you could be creating a dangerous situation. A damaged cutting attachment could break apart as the weed eater is running, potentially causing serious injury. At the very least, your weed eater may not perform properly, causing you double the work later on. This is why Stihl recommends regularly inspecting the head before use and replacing any parts that are cracked, damaged, or worn out.

Even a new weed eater head requires a visual inspection before each use to catch any issues before they become too serious. You should perform regular maintenance as well, including keeping the head clean and monitoring for signs of damage during use. Your owner’s manual should provide cleaning and maintenance tips for your specific Stihl weed eater model.

Advertisement



Source link

Advertisement
Continue Reading

Tech

Cloudflare built a browser for AI agents. It uses 7x less memory than Chromium and runs entirely on Workers.

Published

on

TL;DR

Cloudflare launched Kitesurf, an agent-first browser on Workers. Uses 3-7x less CPU/memory than Chromium. Passes 215K+ WPT tests. Built in Rust/Wasm in 12 weeks. Free in beta via Browser Run. Will be open sourced.

Cloudflare built a browser from scratch for AI agents and launched it today as a free beta inside its Browser Run product. Kitesurf runs entirely on Cloudflare Workers, uses Rust compiled to WebAssembly, and strips out everything a human browser needs but an agent does not: tabs, extensions, themes, pixel-perfect rendering, and 60-fps scrolling. What remains is a lightweight engine optimised for the things agents actually care about, structured content extraction, screenshots, and isolation between sessions.

The performance numbers are the headline. On a 14-URL test corpus, Kitesurf used 3.1x less CPU for screenshots and 3.8x less for HTML extraction compared to Chromium. Memory consumption was 4.7x lower for screenshots and 7x lower for HTML extraction. Chromium is still faster on wall-clock time by about 1.7x because its JIT compiler outpaces a cold software renderer, but CPU and memory are what drive infrastructure bills. Lower resource usage means more concurrent agent sessions at lower cost.

The project went from first commit to public beta in 12 weeks. It passes more than 215,000 Web Platform Tests and adds hundreds more each week. Bots now generate most web traffic, with AI agent activity growing nearly 8,000% in the past year. That surge is exactly why Cloudflare decided a Chromium-based headless browser, built for humans and repurposed for machines, was no longer the right tool. Kitesurf treats every page load as untrusted input, every session as stateless, and every component as isolated, a security model that matches how agents actually operate.

Advertisement

The browser works with Puppeteer, Playwright, and any MCP client that speaks CDP. Cloudflare is simultaneously blocking AI crawlers that harvest content without paying, while building the infrastructure for agents that do pay. That is the dual bet: charge the bots that take, and give the bots that work a cheaper, purpose-built tool to work with. Cloudflare plans to open source Kitesurf once it is ready, and eventually let customers deploy their own instances on their own accounts.

Source link

Advertisement
Continue Reading

Tech

iPhone 17 could cost more a month before iPhone 18 Pro launch

Published

on

Apple may raise the prices of the iPhone 17 roster on August 10, bringing the smartphones in line with other products affected by the memory pricing crisis.

After months of being insulated from the rapidly rising cost of memory plaguing the tech industry, Apple increased the prices of most products in late June. A month and a half later, the iPhone could do the same.

According to a Friday post to Weibo, leaker Fixed Focus Digital says there are rumors of price rises for the iPhone 17. The change is also apparently happening very soon, with it potentially taking place on Monday.

The leaker openly admits that there’s some doubt about the rumor, so it may not happen, but there’s still a chance.

Advertisement

In the same post, they also mention that Apple has decided against an apparent increase in production capacity for the current supply chain. Some lines for unnamed products are at 15% and would’ve risen to 30%, but that’s seemingly cancelled now.

Weibo leakers don’t tend to have the best track record when it comes to accuracy, generally due to a need to churn out posts on the social media site with little in the way of checks. That said, Fixed Focus Digital has been correct in the past, such as with the name of the iPhone 16e.

That said, there is certainly a case for a raising of prices here, even with an iPhone update just a month away.

A change to the usual schedule

Apple’s typical way of changing its iPhone product catalog over the last few years involves introducing all of the new models at once. At the same time, it culls most of the previous generation.

Advertisement

In practice, the 2025 update brought in the iPhone 17 Pro, iPhone 17, and iPhone Air as the new models. The incumbent iPhone 17 Pro were cut, but Apple retained the iPhone 16 at a modest discount.

However, fall 2026 won’t be following the same pattern at all due to splitting the launches. Since Apple sells more of the premium models at first and then more of the standard edition later, it can help both its marketing teams and supply chain by staggering the releases.

The iPhone 18 Pro and Pro Max are certainly anticipated to launch in September as usual, as well as the iPhone Fold, but the lower-spec iPhone 18 won’t until spring 2027.

Rather than leave a gap in the mid-section of its catalog, Apple should keep the iPhone 17 in that slot.

Advertisement

In that scenario, Apple has an incentive to raise the price of the iPhone 17, on the basis of it being an actively sold device for another six months. All during the memory crisis.

It is unclear if the leaker intends the iPhone 17 Pro line to go up in price, too. However, as the iPhone 18 Pro will replace them in a month’s time, complete with a $300 price increase, it may not be a good move for Apple to make.

Either way, it’s not going to be long until we know if the leaker is correct this time.

Advertisement

Source link

Continue Reading

Tech

X replaces ‘misaligned’ revenue sharing program with Original Content Rewards

Published

on

X, the social media platform now owned by Elon Musk’s SpaceX, is shaking up how it pays influencers and creators.

In announcing the change, the company said it will be winding down its existing Revenue Sharing program and replacing it with something called Original Content Rewards. X will stop accepting new Revenue Sharing participants, while existing participants will continue earning money through September 7.

Then, starting on September 8, they’ll be able to apply for the new program. Participants will still need to subscribe to one of X’s Premium tiers, and there will be qualifying thresholds for follower count (500 verified followers) and impressions (500,000 Home Timeline impressions from verified users in 90 days), but it sounds like the big change is the emphasis on originality. 

What counts as original content? X said it can include original reporting and analysis, photos and videos created by the poster, or memes and graphics they’ve designed themselves. Commentary also counts, but “if your content regularly incorporates material created by others, you’ll need to contribute meaningful original value for it to qualify under our original content guidelines.”

Advertisement

The company also included examples of posts that won’t count as original, such as those just copied over from another account, downloaded from one account and re-uploaded to your own, or reposting content “without meaningful transformation.”

This announcement follows repeated attempts by X to reform the Revenue Sharing program, for example reducing payments to aggregators and “clickbait” accounts in April. But these efforts have also prompted complaints from popular accounts profiting from the current system; Musk even reversed some of those changes (giving a creator’s local audience more weight when calculating payouts) after a backlash.

In a post about the new changes, X’s Allegra Jacchia wrote that the existing program “had reached a point where its incentives were misaligned.”

“Creators should be focused on bringing net new content to the platform instead of maximizing payouts,” she said. “We could have kept adding more rules and exceptions, but ultimately the better decision was to start fresh and build a program designed from day one to reward originality.”

Advertisement

Jacchia added that X be “continue refining the program, improving our models, and raising the bar over time.”

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

Source link

Advertisement
Continue Reading

Tech

A Proper Ferrari Testarossa Finally Appears, If Only on Screen

Published

on

Khyzyl Saleem Ferrari Testarossa Concept
Photo credit: Khyzyl Saleem
Digital artist Khyzyl Saleem, better known as The Kyza, took Ferrari’s brand-new 849 Testarossa and gave it the 1980s face so many people felt was missing. Ferrari revived the name for the SF90’s replacement, yet the production car’s sharp modern lines left fans wanting stronger ties to the wide, straked icon that defined the decade. Saleem answered that quiet frustration by reskinning nearly every panel while keeping the greenhouse and the rear window that doubles as an engine cover.

Pop-up headlights return up front, a feature that Ferrari’s own designers were infatuated with before homologation laws put an end to the idea. Clean strips carved from the fenders sit on either side of a flat hood with a clean vented mesh. Slim inlets cut across the bumper just above a carbon fiber splitter that’s as prominent as it is aggressive, and the wide, pointed front end settles into a period-appropriate stance.


LEGO Technic Ferrari 488 Pista Car 42235
  • SPORTS CAR BUILDING TOY – Fuel kids’ passion for high-performance vehicles with the LEGO Technic Ferrari 488 PISTA Car (42235) model building kit…
  • TOY CAR WITH REALISTIC DETAILS – Kids can learn new engineering skills as they put together and explore the V8 engine with moving pistons…
  • FORZA HORIZON 6 UNLOCKABLE – Unlock a special LEGO Technic livery in the Forza Horizon 6 video game – please note that a copy of the game is…

Khyzyl Saleem Ferrari Testarossa Concept
The flanks of the concept reveal the trademark side strakes pushed in deeper and more aggressively than the factory version. They stretch the body out in a clear reference to the original, while a single driver’s side mirror, another chosen historical element that Saleem did not bother to disguise, maintains the overall look authentic. Vintage HRE 505 alloys with a distinctive star-spoke pattern complete the look, sitting perfectly beneath the modified fenders.

Khyzyl Saleem Ferrari Testarossa Concept
The vehicle’s strongest point is most likely its rear end. Tiny LED taillights sit inside full-width black slats, only framed by those ultra-wide shoulders. A large but elegant carbon fiber diffuser wraps around a couple of round exhaust outlets, combining modern hybrid hardware with the historic Testarossa design without feeling forced or manufactured.

Khyzyl Saleem Ferrari Testarossa Concept
Under the hood, the concept car has the same configuration as the road car: a twin-turbo 4.0 liter V8 connected to three electric motors for a total of 1,035 horsepower transmitted to all four wheels. With 849 horsepower, the production car ranks among Ferrari’s most powerful series cars. Saleem concluded that the packaging simply need a body that felt more related to the name it conveys.

Khyzyl Saleem Ferrari Testarossa Concept
He has previously stated that he admires the factory car’s aggressive attitude, but he couldn’t resist bringing the design closer to the car’s own history to see what would happen. The tragic death of electronic music producer Kavinsky served as an emotional catalyst for Saleem to launch his endeavor. The images soon spread on Facebook and Instagram, earning high praise from many who had hoped for this type of reinterpretation – and Saleem’s design delivered in spades.
[Source]

Source link

Advertisement
Continue Reading

Tech

South Korean satellite spots SpaceX lunar impact

Published

on

science

Before and after shots of Elon’s ejecta snapped by Danuri

The race for the first before-and-after images of the spent SpaceX rocket on the Moon has been won by the South Korean space agency.

The Korean Aerospace Administration (KASA) posted imagery from its Danuri lunar orbiter showing the impact site on the Moon on August 6.

Advertisement

The impact was predicted to occur on August 5, and marked a landing on the Moon for SpaceX several years ahead of schedule, just without humans onboard and traveling a good deal faster than planned.

The words “crater” and “SpaceX” are not something investors nervously tracking the company’s stock price following the weeks since its IPO will want to think about.

The impact came on the fourth anniversary of Danuri’s launch, on a SpaceX Falcon 9 rocket. It took a few months for the probe to reach the Moon, and its initial one-year mission has since been extended twice – most recently through 2027. The plan is currently for Danuri to intentionally make its own impact crater in March 2028.

The upper stage of the Falcon 9 is expendable, and this one was used to send Firefly Aerospace’s Blue Ghost Mission 1 and ispace’s RESILIENCE to the Moon in 2025. On most Falcon 9 missions, the upper stage burns up in Earth’s atmosphere, but a lunar transfer orbit like this one made that impossible. In a post, SpaceX stated, “Impacts like this are rare, but they can happen with objects in these types of orbits, and we worked with NASA on the optimal disposal solution,” and wrote, “a controlled disposal maneuver is not always possible.”

NASA boss, Jared Isaacman, noted that NASA intentionally crashed spent stages into the Moon in the name of seismic research, and commented, “The reality is that meteoroids strike the lunar surface hourly, with impacts comparable to this upper stage occurring roughly every six days.”

Advertisement

Humanity has not done a great job of managing space junk around the Earth. While the SpaceX Falcon 9 upper stage did not pose any risk to humans, it highlights the need to devise rules for the use of lunar space rather than risk an uncontrolled buildup of debris.

In June, The Register spoke to Neuraspace boss Chiara Manfletti about how things might be done differently as lunar traffic increases. Neuraspace is all about satellite collision avoidance, and Manfletti’s ideas included requirements from stakeholders to designating dumping grounds on the Moon.

NASA’s veteran Lunar Reconnaissance Orbiter was also attempting to capture the impact, and more images are likely to be released over the coming week.

Now to name the impact site. Elon Crater, anyone? ®

Advertisement

Source link

Continue Reading

Tech

How to watch Argentina vs South Africa for FREE: live stream details

Published

on

South Africa begin their preparations for welcoming New Zealand in Buenos Aires as they face Argentina.

The Springboks finished the summer section of the Nations Championship with three wins from three and have now gone almost a year since their last defeat — a 22-38 loss to Australia. But August brings the Greatest Rivalry Tour, and with it, the chance to beat New Zealand on their own turf.

Source link

Advertisement
Continue Reading

Tech

Seven experts join AI Office of Ireland board

Published

on

Members include distinguished AI researcher Alan Smeaton and Shutterstock AI director Alessandra Sala.

Seven experts across technology, academia and public administration have been appointed to oversee the newly established AI Office of Ireland. Led by Mary Doyle, a former higher-up across Government departments, the board will help the new office implement the EU’s landmark legislation on AI.

Alongside Doyle as chairperson, the board also features two former members of Ireland’s AI Advisory Council, Prof Deirdre Ahern and Prof Alan Smeaton.

Last week, former Medical Council of Ireland executive director of education, innovation and AI Paul Byrne joined AI Office of Ireland as its CEO.

Advertisement

Minister for Enterprise, Tourism and Employment Peter Burke, TD said that the appointments marked another milestone in Ireland’s “ambition to be a leader in trustworthy and human-centric AI”.

“The board will provide leadership and oversight to support the Office in its role as Ireland’s central coordinating authority for AI regulation under the EU AI Act,” the minister said.

“Their collective expertise will help ensure that Ireland develops a robust and innovative framework for the governance of AI while also safeguarding fundamental rights and fostering public trust.”

The Office, which began operations earlier this week, will act as the primary point of contact for the European Commission, national authorities and the public around matters regarding the AI Act.

Advertisement

It is expected to develop a consistent regulatory framework to implement the law, facilitate centralised access to technical expertise, and drive AI innovation and adoption across the country.

The board will comprise:

Mary Doyle, chairperson

Doyle is a former assistant secretary general and deputy secretary general in the departments of the Taoiseach, health and children, and education and skills.

She is also a board member at Ireland’s national association of community organisations – The Wheel – and has previously served as a member of the National Economic and Social Council, on the board of Science Foundation Ireland, with the National Statistics Board, and as a forum member of the Economic and Social Research Institute.

Advertisement

“AI presents enormous opportunities to enhance public services, support economic growth and improve people’s lives. Realising those opportunities requires a governance framework that promotes innovation while maintaining public trust and protecting fundamental rights,” Doyle said.

Deirdre Ahern

Ahern is a professor of law at Trinity College Dublin, where she is also the director of the technologies, law and society research group. Her research focuses on corporate law, corporate governance, and the regulatory challenges presented by AI and other technologies. She is also a principal investigator with Research Ireland’s Rinn AI and Rinn Quantum. Ahern was appointed as a member of the Irish Government’s AI Advisory Council in 2024.

Michelle Kearney

Dr Kearney, who works in the Department of Enterprise, Tourism and Employment, led efforts to establish the AI Office of Ireland. She is currently working on the upcoming International AI Summit in October, a key event in Ireland’s rotational EU presidency this year.

Siobhán Noble

Noble is the chief data and AI officer at Carne Group, an Ireland-based financial services company. She also serves as the vice-chair of the Irish Funds Emerging Technologies and Innovation working group.

Advertisement
Donal Rooney

Rooney is a non-executive director with Enterprise Ireland and Oxfam Ireland. He is a chartered accountant with experience managing finance across multiple roles in private equity, public companies and public sector environments.

Alessandra Sala

Dr Sala is senior director of AI and data science at Shutterstock, the global digital media company. She chairs the AI and Multimedia Authenticity Collaboration, which attempts to unify standards between organisations to foster a unified response to deepfake technology. Sala is also the co-chair of the UNESCO Women for Ethical AI platform.

Alan Smeaton

Prof Smeaton was a board member of the Irish Research Council for seven years before being appointed to the AI Advisory Council in 2024. He is also a founding director of the Insight SFI Research Centre for Data Analytics, and is a well-established name in AI and data analytics research.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

Advertisement

Source link

Continue Reading

Tech

How These AI Assistants Differ

Published

on

For many, there’s a clear winner in this battle of the artificial minds.

ChatGPT was probably most people’s first encounter with an LLM, considering how many users it picked up shortly after its launch in 2022. It also enjoyed the privileged position of having no real competition until Anthropic released Claude a few months later in March 2023. Since then, both companies have constantly improved their models and added new features. But because there’s an overlap in features — both platforms have a coding mode and a dedicated workspace mode — it can be difficult to decide which tool is best for you.

Also, there’s little difference between the two during everyday usage: You can use either of them to code, answer a quick query, summarize documents, organize your files, and, if you’re feeling particularly blasphemous, write thought leadership pieces in the style of Herman Melville.

However, anything beyond that, and you’ll need to start thinking about which LLM fits your use case better.

Advertisement

Is Claude more accurate than ChatGPT?

Measuring accuracy in LLMs can be tricky, as there’s no straight answer. The specific model you’re using and the prompt you feed into it play an important role in the quality of the output.

When it comes to flagship models — Claude Fable 5 (Max) and GPT 5.6 Sol (Max) — Claude is marginally more accurate according to the AA-Omniscience Accuracy benchmark. The scores stand at 61 percent and 59 percent, respectively. Because the difference is so marginal, you’ll rarely notice it in day-to-day usage.

Advertisement

But, unless you’re tokenmaxxing, you’ll be using mid-tier models for most tasks. On Claude, this is Sonnet 5, and on ChatGPT, 5.6 Terra. Here, the scales are tipped in ChatGPT’s favor: ChatGPT 5.6 Terra (Max) scores 46 percent, whereas Claude Sonnet 5 (Max) is significantly lower at 38 percent.

Another important factor when measuring accuracy is the tendency of the model to hallucinate. Ideally, if an LLM doesn’t know the answer to something, it should flat out refuse to answer it instead of making stuff up, i.e., hallucinating. The benchmark for this is AA-Omniscience Hallucination Rate, in which Claude has a significant leg up against ChatGPT. A lower score is better in this benchmark, and Claude’s Fable 5 model scores 55 percent compared to ChatGPT 5.6 Sol’s score of 89 percent. The difference is even more stark in the mid-tier models: Claude Sonnet 5 scores 37 percent, whereas ChatGPT 5.6 Terra scores 85 percent.

Advertisement

What can Claude do that ChatGPT cannot?

There’s a difference in what people use Claude and ChatGPT for. According to the Anthropic Economic Index report from March 2026, 42 percent of Claude conversations revolved around personal usage and 45 percent were related to work (the remainder was coursework). A similar report by OpenAI states that 70 percent of ChatGPT usage is non-work-related.

Claude Cowork, which came out in January 2026, can perform knowledge-based tasks on your behalf: I mostly use it to organize my files. You can also create instruction bundles called skills that can be invoked mid-conversation using the forward slash (/). Additionally, Claude Artifacts can instantly render code snippets, single-page HTML websites, interactive React components and diagrams. You can share and publish Artifacts on the web too.

ChatGPT supports skills too, but the implementation differs from Claude’s. For example, ChatGPT skills can only be used by individual users in Codex and the API, not regular chats. Claude’s skills can be used across chat, Claude Cowork and Claude Code. ChatGPT’s Artifacts equivalent, Sites, is targeted towards businesses for internal use and is only available in Codex. Additionally, ChatGPT Sites can’t pull live data, unlike Claude Artifacts, which can do so using connected apps and Model Context Protocol (MCP) connectors.

There are some features that are better implemented in ChatGPT. Most notably, the voice feature: ChatGPT sounds more natural, and you can interrupt it and talk over it without it losing track of what it was saying earlier. For real-time assistance, you can use the live video feature while in Advanced voice mode and point it at the issue you’re facing. ChatGPT also has an image generation feature that can create photorealistic images, whereas Claude can only build diagrams, charts and interactive visuals using HTML and SVG.

Advertisement

Is ChatGPT getting worse in 2026?

ChatGPT’s new models score better than previous ones on almost all AA-Omniscience benchmarks, yet the overall experience of using the app seems to be getting worse. If you feel this way, you’re not wrong, because there are areas where the previous models performed better than the latest ones.

One such area is hallucinations: The same AA-Omniscience Hallucination Rate benchmark I mentioned earlier states that GPT-4o has a hallucination rate of 38 percent compared to GPT-5.6 Sol’s 89 percent. OpenAI also introduced safeguards to align with the industry’s maturation. Tone shifts, for example, may seem jarringly different with each update.

Additionally, OpenAI began showing ads in the free and ChatGPT Go tiers of the app while Anthropic improved Claude’s free tier by adding more features and skipping ads. Considering the company loses money even on paying customers, and OpenAI CEO Sam Altman once called ads a “Last Resort,” it’s a reasonable assumption that the ChatGPT experience may get even worse for free tier users.

Advertisement

There’s another reason why users left ChatGPT and Claude saw a simultaneous surge in downloads in recent months. Back in February 2026, Anthropic’s deal with the US Department of War fell through, and the company was designated a supply chain risk because it refused to let its models be used for mass domestic surveillance and fully autonomous weapons. Hours later, OpenAI announced that it had signed a similar deal with the US government in place of Anthropic, with certain safeguards in place. This decision caused many people to switch over to Claude because Anthropic was perceived as the more “ethical” AI company.

ChatGPT and Claude are priced similarly

Both Claude and ChatGPT’s free plans only offer you access to their mid-tier models: Sonnet 5 for Claude and GPT-5.5 for ChatGPT. The difference is that there are ads in ChatGPT.

Claude’s paid plans include the $20/month ($17/month if billed annually) Pro plan, the $100/month Max 5x plan (five times the usage limits of Pro) and the $200/month Max 20x plan (20 times the usage limits of Pro). The paid plans let you access the flagship models and Claude Code as well. However, Pro users have token-based billing for Fable 5, while users on the Max 5x or Max 20x plan can use 50 percent of their weekly usage limits on Fable 5. If you hit the limit, you can always purchase more usage credits.

Advertisement

ChatGPT’s pricing is similar. There are $20/month, $100/month and $200/month plans that offer an increase in usage limits and access to flagship models. Access to Codex is free, but the low usage limits practically cancel it out. However, unlike Claude, ChatGPT has an $8/month plan that offers increased usage limits but still shows ads.

Source link

Advertisement
Continue Reading

Tech

Apple’s rumored “Ultra” lineup would include a foldable iPhone, camera AirPods, and a touchscreen MacBook

Published

on

Rumor mill: Apple is expected to roll out three new “Ultra” products over the next several months, starting with a foldable iPhone expected to arrive this September, followed by new AirPods and a redesigned MacBook. The reported lineup would expand Apple’s hardware ambitions into foldable displays, camera-equipped earbuds, and a touchscreen laptop with possible cellular connectivity.

The foldable phone, likely to be called the iPhone Ultra, would be Apple’s most expensive iPhone to date. Reports put its US starting price between $1,999 and $2,499. Apple is expected to introduce it alongside the iPhone 18 Pro and iPhone 18 Pro Max.

The device is rumored to feature a 7.7-inch inner screen and a 5.3-inch outer screen. The larger display would give users more room for video, apps, and multitasking when the phone is unfolded. The phone is also expected to feature two rear cameras and a single front-facing camera.

One notable reported change involves authentication. Instead of Face ID, the iPhone Ultra could use a Touch ID sensor built into the power button. That would mark a departure from Apple’s current flagship iPhones, which rely on facial recognition.

Advertisement

The company may also use the Ultra name for a new version of its wireless earbuds. The AirPods Ultra are expected to resemble the AirPods Pro but add tiny infrared cameras. Those cameras could give Siri more information about the user’s surroundings and expand the capabilities of Visual Intelligence.

Visual Intelligence is available on the iPhone 15 Pro and newer models. In iOS 27, users can access the feature through the Camera app using a new Siri mode. AirPods with cameras could allow users to use Visual Intelligence without having to hold up their phone.

A small LED on the earbuds would indicate when the cameras are in use. Apple code found in the second iOS 27 developer beta last month appeared to reference AirPods with cameras, suggesting that the product could be moving closer to launch. AirPods Ultra could arrive as early as this year, potentially putting them on the same release schedule as the iPhone Ultra.

Apple is also said to be working on a MacBook Ultra for release by early 2027 featuring an OLED touchscreen. Apple has never offered a touchscreen MacBook, making the feature one of the more significant aspects of the report.

Advertisement

Other rumored features include M5 Pro and M5 Max chips, a Dynamic Island, and a thinner design than the latest MacBook Pro models. The company could also include an Apple-designed C2 modem, giving the laptop built-in cellular connectivity.

Source link

Advertisement
Continue Reading

Trending

Copyright © 2025