Google has streamlined a key feature of its Gemini AI assistant on Android devices, speeding up image sharing and editing, as spotted by Android Authority. The latest Gemini update lets you send images directly from other apps to Gemini instead of the more cumbersome setup that was in place before.
Now, if you have a picture in, for instance, Google Photos, that you want Gemini to look at in conjunction with a text prompt, you can submit it directly via Android's built-in share sheet as you would to send a text with the image attached. That's much easier than starting in the Gemini app, tapping on the upload image button, locating the image you want, and attaching it. And if your image is in the cloud, you would also need to download it to your device. It might not be more than a minute or even less, but if you want Gemini to explain a photo or use one to inform a new AI-generated image, that extra time and friction might put you off the idea.
It's not a total revolution for Gemini, however. Submitting images to the AI is faster, but only images. You can't use the sharing button to send text or a link to Gemini. It also doesn't encompass the Gemini overlay, which lets you use Gemini without switching out of the app you're currently using. While the image gets sent to the AI app, you still actually need to switch to the app to use Gemini's features.
Gemini Speed
Though subtle, the update is part of Google's efforts to smooth the road for intuitive engagement with Gemini. If you often use Gemini for multimedia content, the update could save you time in the long run. Gemini will be able to analyze the image and provide insights, descriptions, or even text content based on what it "sees" more quickly than before. This makes the app more useful for users who need to switch between different types of media in their daily workflows.
Even if Gemini is only an occasional part of your mobile usage, a minute or less can affect whether you decide to skip using Gemini. That's anathema to Google's plans to embed Gemini throughout your mobile device experience and your life in general. It's also another way for users who already rely on Google's ecosystem, such as Photos or Drive, to thread Gemini into how they use those other services. Making Gemini more convenient is clearly a major goal for Google. As ChatGPT and other AI assistants keep upping their multimodal features, Gemini will need this kind of edge to stay ahead, or at least keep even, with its rivals.
Tesla's bizarre-looking Cybertruck has suffered through yet another recall. It's a sign that not only is the truck a bit like an experiment on wheels, but that Tesla is increasingly no longer the standard bearer for our EV aspirations.
"Elon Musk is the reason both cars in my driveway are electric. He’s also the reason neither of them are Teslas," wrote Firebird Music CEO Nathan Hubbard on Threads. The post encapsulates a growing sentiment among EV owners and EV curious: Tesla is not necessarily the electric car company that will own the future of the market it all but created.
Look, I don't want to dive too deeply here into the unpleasant actions and social presence of Tesla CEO and X owner Elon Musk. Let's just say that the once inspiring entrepreneur has become, to put it lightly, a polarizing figure. It's not a good look for someone who wants to sell electric vehicles to everyone and not just those who share his world (and off-world) views.
There is, though, no denying that Musk has put Tesla fans in a difficult position and they are perhaps making different choices. As of Q2 2024, Tesla owned, according to the US Energy Information Administration 48.9% of the US EV market. The number is notable because Tesla no longer owns the "majority" of the market. Granted, what's left is split among numerous competitors, including major US auto manufacturers who, over the last few years, moved aggressively into the EV and Hybrid market. It hasn't all gone smoothly.
Earlier this year, Ford pulled back on its aggressive EV plans. Plus, consumers still appear more interested in half-measures. Hybrids, which combine combustion with one or two electric motors appear, based on recent United States Department of Transportation statistics, to be far out-pacing pure EV ones, which, is likely another cause for concern for Tesla.
United States Dept. of Transportation • Bureau of Transportation Statistics EV growth(Image credit: United States Dept. of Transportation)
The EV market continues to shift around Tesla, with more competitors offering comparable products that are, finally, approaching Tesla on the range front and often beating it on sticker price. The point is we now have choices. I know that when I finally make the EV move, it won't be with a Tesla. That's an important shift for me. I was a stalwart Tesla fan who wrote extensively about early Tesla Model S owners and remember fondly being among the first to experience a Model 3 test drive.
Now, though, I can't stomach the idea of enriching Elon Musk. I know Tesla employs over 140,000 people, and most of them probably don't agree with Musk's often publicly stated views, but he remains a highly visible and vocal leader.
On October 10, Musk is expected to finally launch his Tesla robo taxi. That event will be notable, I think, for its lack of direct consumer appeal. There'll be no Model 3 update or a cheaper Tesla. Affordable is not in Tesla's vocabulary. The company recently stopped selling the cheapest version of the Model 3.
I'd say there's no longer a point in waiting for a truly affordable Tesla. Generally, the automaker's appeal to everyday Americans is, I think, waning, and there may be no better evidence than the Cybertruck.
Considering all that, perhaps these recalls are unsurprising. The latest one from the US National Highway Traffic Safety Administration relates to rearview mirrors that deliver a feed from rear cameras as you back up. The problem, according to the NHTSA (PDF) is that when you start to back up, it takes an extra second for the feed to appear. Fortunately, the fix will arrive in an over-the-air software update.
Recalls are not inherently a disaster. Untold combustion engine vehicles get them. For Tesla, they're certainly not confined to the Cybertruck. The company has had dozens that range across its most popular models. Still, this is already the fifth recall for the relatively new Cybertruck and it often seems like the EV was built not based on practicality (watch it fail as a "truck"), or elegance but on the design whims and demands of Musk, who I think wanted something that looked more at home on the surface of Mars that it does on Earth.
You can wrap it in black but it still doesn't look normal.(Image credit: Lance Ulanoff)
When I see a Cybertruck rolling through my neighborhood, I don't admire it; I gawk. It always looks out of place. I also think I instantly know the driver, not personally, but who this person is. They're clearly someone who's bought into not just the Tesla EV strategy but Musk's perspective on trucks, modern transportation, and maybe his twisted worldview. Why else would you spend over $100,000 to drive something that looks like it's well-equipped to survive a nuclear blast but ill-prepared for a casual run to the grocery store?
Cybertruck screams, "Look at me," but as sentiment about Tesla changes in the EV market, that scream sounds increasingly like a cry of desperation, "Look at me, please, while I try to remain relevant."
The number of phishing emails that masquerade as notifications from Microsoft services is skyrocketing, a new report from Check Point has warned.
In the report, the researchers said that just in September, its service caught more than 5,000 such emails - and to make matters worse, the attackers have gotten extremely good at creating a legitimate-looking email.
The usual suspects - spelling and grammar, color scheme, the email’s outline - all of these things have been brought to perfection: “The language is perfect. The style is familiar. The graphics look impeccable,” the researchers said. “So, what should organizations do?” Furthermore, these emails now come with copy-pasted Microsoft privacy policy statements, or links to Microsoft and Bing, all of which makes spotting the ruse with the naked eye almost impossible.
Training and AI
Ultimately, even the ‘sender’ field in the email looks believable now. Instead of the usual private, or unknown domains, these emails appear to be coming from organizational domains impersonating legitimate administrators.
All of this means there is a higher chance of organizations losing sensitive information, or becoming infected with malware and even ransomware.
In response, organizations need to invest heavily into user awareness training, since employees will no longer be able to hunt for spelling and grammar mistakes in phishing emails, Check Point argues.
Also, they should deploy AI-powered email security, essentially fighting AI with AI, and finally, always keep their software and hardware updated.
Hackers can now steal people’s cryptocurrency wallet seed phrases, even when they are stored as an image file, experts have warned
When a user sets up a new crypto wallet, they get a “seed phrase” - a set of 12 or 24 random words, which can later be used to restore the wallet in a new app or device (in case of loss or theft). Crooks that happen to steal a seed phrase can manage the money found in the wallet however they like.
But when a person saves the seed phrase in an image file (for example, with a screenshot), it makes the criminals’ job that much harder.
A highly potent threat
Enter Rhadamanthys version 0.7.0, recently introduced and carrying new, important bells and whistles. Recorded Future's Insikt Group recently analyzed this new version and released an in-depth report, which states that the infostealer now comes with Artificial Intelligence (AI) capabilities, and allows for optical character recognition (OCR).
Together, these two tools are called "Seed Phrase Image Recognition" which, in the above context, is pretty self explanatory.
"This allows Rhadamanthys to extract cryptocurrency wallet seed phrases from images, making it a highly potent threat for anyone dealing in cryptocurrencies," Recorded Future's Insikt Group said in its analysis. "The malware can recognize seed phrase images on the client side and send them back to the command-and-control (C2) server for further exploitation."
Even before the new features, Rhadamanthys was a potent, and popular infostealer. It was first discovered back in 2022, and has since grown into one of the most formidable pieces of malware. Hackers can subscribe to the service, paying $250 a month for the infostealer (or $550 for 90 days).
The latest version was released in June 2024, and comes as a "complete rewrite of both client-side and server-side frameworks, improving the program's execution stability." Recorded Future concluded.
AMD is pushing the boundaries of integrated GPUs with its upcoming Ryzen AI Max series, based on the Strix Halo architecture, according to recent leaks and reports from various tech forums. Positioned to go beyond the Ryzen AI 9 HX 375, this series aims to offer exceptional performance in integrated graphics and processing power.
According to leaks from David Huang on X (formerly Twitter) and additional reports from Chinese tech forums, the Ryzen AI Max lineup will feature three models. The Ryzen AI Max 385 will have 8 CPU cores and an iGPU with 32 Compute Units (CUs). Above this is the Ryzen AI Max 390 with 12 cores and 40 CUs, while the flagship Ryzen AI Max+ 395 boasts 16 cores and 40 CUs.
These chips will reportedly have enough GPU power to outperform dedicated cards like the Radeon RX 7600 XT, making them ideal for for tasks like gaming, AI and professional workloads.
Zen 5 cores
Built on a chiplet design, Strix Halo will feature Zen 5 cores and powerful integrated graphics based on the RDNA 3.5 architecture.
On X, David Huang shared further details, stating, “Today I talked about gfx1151 with my group friends and found out that the RDNA3.5 on Strix Halo was added with a 192K complete VGPR [Vector General Purpose Registers] description in LLVM td not long ago, and various ROCm libraries have recently secretly added official support for Strix Halo... Combined with the CPU Geekbench 5 running score I saw before, it seems to have a full AVX512 width and twice the bandwidth per CCX. Can laptops and even tablets be used as supercomputers?"
The inclusion of up to 128GB of LPDDR5X memory and a 32MB MALL cache further boosts the potential for high-performance graphics and computing tasks. Strix Halo is also expected to support up to 96GB of memory for video processing, making it suitable for workstation applications and tasks like LLMs. The APUs will also feature a 60 TOPS NPU, enhancing AI processing capabilities.
The new Strix Halo APUs will likely debut early next year with CES 2025 expected to serve as the official launch platform. No doubt more information will appear the closer we get to the event.
The WashG1 is a dedicated wet floor cleaner and Dyson's first attempt to prove that it doesn't just do carpets. It launched in the UK and Australia last month but has just gone on sale in the US. It's currently only available to buy direct from Dyson, and has a list price of $699.99.
Unclutch those pearls; we all knew it was going to be expensive. I do think that some Dyson products justify their eye-watering price tags, but in this case, there are things worth factoring in before you decide to gamble your child's college fund on a wet floor cleaner.
I tested one out and you can get the full low-down in my Dyson WashG1 review, but the gist is that it works fantastically well on perfectly smooth, flat floors like linoleum or polished concrete but is nowhere near as impressive on textured or uneven floors (including tiled floors with grouting gaps).
(Image credit: Future)
This is Dyson's first dedicated wet floor cleaner (I say 'dedicated' because we do have the Dyson V15s Submarine, which is a vacuum cleaner with a wet floorhead that can be swapped in). Significantly for this brand, which has built its reputation on being really good at moving air about, it doesn't use suction. Instead, it employs a combination of agitation, hydration and separation to get your floors gleaming.
Water is expelled through the cleaner head, rollers help loosen the dirt and pick up things like hair and solid particles, and then the inner mechanisms separate liquid and solid spillages. That last part is designed to make maintenance easier.
Should you buy one?
It's very good at certain things. Like today's best Dyson vacuums, it's extremely maneuverable; the floorhead can pivot any which way, and it'll get right up close to baseboards, too. The fact that it can handle liquid and solid waste is really helpful for things like dinnertime messes. I have a small niece and nephew who cannot complete a meal without coating everything in the vicinity with whatever they've been eating, and a once-over with the WashG1 is by far the least disgusting way to deal with it that I've found so far. The base will take care of some of the maintenance by running a self-clean cycle, when you dock it, too.
(Image credit: Future)
However, it's not worth the investment if you have uneven floors. The WashG1 will struggle to clean them evenly, as I discovered when I tested mine on a flagstone floor. Because the rollers don't really 'scrub', it's only really capable of tackling surface dirt.
That includes missing the grouting cracks between tiles. (Apparently, the engineers found that adding more water is a more effective way to tackle stubborn dirt than rubbing at it, and while they might have a bit of a point, I still think there are limitations to this approach.)
Those niggles aside, it still might be a good investment for some shoppers. Because it's brand new, don't expect discounts any time soon – I have my fingers crossed for a price-drop in the Black Friday sales, though.
At its AWS re:Invent 2023 event in November 2023, CEO Adam Selipsky unveiled the company's Graviton4 chips, built on Arm’s “Demeter” Neoverse V2 core. These new processors were claimed to offer up to 30% better compute performance, 50% more cores, and 75% more memory bandwidth than the Graviton3 processors, aimed at boosting memory-optimized and compute-heavy workloads.
Graviton4-powered R8g instances became generally available in July 2024 and when Phoronix benchmarked it, the processor came in ahead of the Intel Xeon instance by about 5% and wasn’t too far behind AMD’s EPYC.
AWS has now stepped things up by introducing new memory-optimized X8g instances.
(Image credit: AWS)
Improved performance
X8g instances are available in ten virtual sizes and two bare-metal configurations, featuring up to 3 TiB of DDR5 memory and 192 vCPUs. These instances are also considered AWS's most energy-efficient EC2 offerings to date. With a 16:1 memory-to-vCPU ratio, the X8g instances are also equipped with full encryption across all high-speed hardware interfaces, and offer additional security features.
The X8g instances provide notable improvements over their predecessors, offering three times more memory and vCPUs than X2gd instances. They also double the EBS and network bandwidth, making them an attractive solution for memory-heavy workloads. Graviton4’s twice the L2 cache per core and 160% higher memory bandwidth contribute to up to 60% better compute performance.
In terms of real-world applications, AWS says X8g instances are already being used for SAP HANA and SAP Data Analytics Cloud, with performance gains of up to 40% for transactional workloads compared to Graviton3.
AWS notes the X8g instances allow users to scale up or out, providing flexibility for memory-bound applications currently running on distinct instances. The new instances are now available in US East (N. Virginia), US West (Oregon), and Europe (Frankfurt) AWS regions, supporting various pricing models such as On-Demand, Spot, and Reserved Instances.
