If you’re in the market for a new tablet, now is the time to take the plunge. Even though it’s past the official day, the Fourth of July sales are still in full force and Amazon Prime Day deals are already starting to heat up. Which means it’s a great time to upgrade your tablet for less. Our CNET shopping experts have been scouring all of the sales going on now to find the deepest discounts on top tech, and one of our favorite offers is on the Samsung Galaxy Tab S10 FE Plus.
Right now the 128GB Samsung Galaxy Tab S10 FE Plus with Wi-Fi is down to $600 on Amazon ahead of Prime Day, which saves you $50 over its usual asking price. Best Buy is also matching this deal. If you need more storage, Amazon is selling the 256GB version of this tablet at $50 off, bringing it down to $700.
The Samsung Galaxy S10 FE Plus tablet has a 13.1-inch screen and a stunning QHD display with 2880 by 1800 pixel resolution. That means you’ll get detailed images, brighter colors and better contrast. Samsung equipped this tablet with a battery that lasts up to 21 hours and made the S10 FE Plus waterproof for more durability. Your purchase comes with an S-Pen calibrated to work precisely so you can draw, take notes and work more easily. You’ll also get a USB-C cable for easy charging as well. All S10 FE Plus tablets can seamlessly sync with other Samsung devices, which is a boon if you’re well-integrated into their ecosystem.
If you’re on a budget and need alternatives, the 128GB Samsung Galaxy S10 FE is now just $450 at Best Buy, saving you $50. Amazon also has this same offer. Looking for a new tablet but not sure if this deal is for you? Check out our list of the best Prime Day tablet deals for more ideas.
Best Prime Day Tablet Deals
Save yourself from scrolling through Prime Day tablet deals with the help of our shopping experts, who have rounded up top savings for you right here.
See Now
TABLET DEALS OF THE WEEK
Deals are selected by the CNET Group commerce team, and may be unrelated to this article.
Why this deal matters
The Samsung Galaxy S10 FE Plus offers a long-lasting battery, QHD resolution and plenty of memory. This new discount of $50 at both Amazon and Best Buy offers the perfect opportunity to score one of these tablets without paying for full asking price, and S10 FE options are also available for those who want a slightly cheaper option.
Valve has released the latest Deadlock patch notes, as it continues on the road to the full launch. The Half-Life and Team Fortress 2 studio has made a plethora of changes with this update, as a global reduction to gun and ability damage take centre stage. So whether you’re still active or fancy diving back in, these changes to Deadlock make now the ideal time to get playing.
The first change Valve makes to Deadlock in this newest patch is to bullet cycle time, which has been increased by 5% for every single hero. Developer ‘Yoshi’ explains that this will effectively reduce all gun damage per second by 5%. This should put more weight into how and when you use skills, changing the flow of battle ever so slightly and impacting the fight for minion kills.
Yoshi also says “Ability base damage and AP bonus damage [are] reduced by ~8% (spirit power growth unaffected),” which means everyone should also find their abilities to be slightly weaker moving forward. Combined with the changes to DPS, it looks like Valve wants to reduce lethality across the board in the game, which should give you more chance to turn fights around. While Deadlock is already one of our picks for the best MOBAs on PC right now, it’s worth noting that Valve is still deep in development, so a lot of this could change again.
Your overall Spirit gains from leveling have also been reduced from 1.25 to 1.1, while your Spirit bonus based on souls spent in the tree is increased by a whopping 25%, changing the way we power up throughout a match. Speaking of which, the team is also increasing Ultimate cooldowns by 10%, which it says “affects base and upgrades, so the total CD with AP is 10% longer.”
While we don’t know when the next Deadlock patch could be, I’d expect some big stuff on the horizon. Earlier this year the studio completely reworked the shop, introduced new items, and made some major visual changes, all of which almost made Deadlock feel like an entirely new game. With the Deadlock release date still on the horizon, I can’t wait to see what comes next.
Valve has now released the Friday July 4 Deadlock update. You can go through all the changes right here.
As you dive into all of these changes, make sure you check out our tier list of all the Deadlock characters. We’ve also got a very useful look at Deadlock crosshairs, which will help your shooting.
You can follow us on Google News for daily PC games news, reviews, and guides. We’ve also got a vibrant community Discord server, where you can chat about this story with members of the team and fellow readers.
We spend hours testing every product or service we review, so you can be sure you’re buying the best. Find out more about how we test.
Street Fighter 6 Years 1-2 Edition marks the fighting game’s availability on Nintendo Switch 2 – as a launch title for the console no less. Back in 2023, I scored Street Fighter 6five stars in my review, praising everything from its world class visuals to its intense one-on-one battles – backed up by the incredible Drive gauge system that allows for a high skill ceiling.
Review information
Platform reviewed: Nintendo Switch 2 Available on: Nintendo Switch 2, PS5, Xbox Series X|S, PC Release date: June 6, 2025 (originally released on June 2, 2023)
On Nintendo Switch 2, that high-quality experience has been replicated for the most part. Seemingly gone are the days of heavily compromised fighting game ports (looking at you, Mortal Kombat 1), as Capcom’s highly scalable RE Engine shows that even the best looking fighting games can still run and play great on a handheld machine.
Now, okay, you’re not quite getting the crystalline image quality of Street Fighter 6 on PlayStation 5, Xbox Series X and Series S or indeed PC. The Switch 2 version does look slightly worse, with noticeable graining and overall lower resolution – exacerbated when blown up on a 4K display. But honestly, it’s not a million miles away from those other versions, and the game is still perfectly readable and maintains a solid 60fps performance where it matters: in those intense online matches as well as most offline modes.
Perhaps obviously, it’s the single-player World Tour mode that suffers the most. Granted, it’s far from perfect on other systems, but you’ll really feel the compromises on Switch 2, with noticeable animation frame limiting on distant non-player characters (NPCs) and a hard 30fps lock during battles – far from ideal.
Still, everything else in the package is comparable. And with full crossplay support and a complete character roster at the time of writing, you can hop into one of the best fighting games on Nintendo Switch 2 and start ranking up just as well as you can on other consoles. Well, so long as you’re using a Nintendo Switch 2 Pro Controller for Classic motion inputs, of course. But more on that later.
The gang’s all here
(Image credit: Capcom)
So first off, what’s included in Street Fighter 6 Years 1-2 Edition on Switch 2? As mentioned, you’re getting the entire roster of fighters. That’s 18 from the launch version, in addition to the four DLC characters from Year 1 (Rashid, A.K.I., Ed, and Akuma) and four from Year 2 (M. Bison, Terry Bogard, Mai Shiranui, and Elena). You’re also getting four DLC stages and a decent amount of Drive Tickets to spend in the game’s shop. Sadly, additional character outfits aren’t included in this package – you’ll have to buy those separately.
Otherwise, this is a feature-complete version of Street Fighter 6. All three main content hubs are here, including the single-player World Tour mode. You’ll also find the Battle Hub for hanging out with player-created avatars and engaging in casual online matches, as well as Fighting Ground where you have access to the bulk of the game’s modes. That includes ranked and unranked online matches, Arcade mode, training, and combo trials for every character.
There are also a couple of modes exclusive to the Switch 2 version of Street Fighter 6, but these are unfortunately novelties that wore off after a couple of tries. Gyro Battle uses the Joy-Con 2’s motion controls for special move inputs, while Calorie Contest also applies those motion inputs to a gimmicky fitness-based mode. Both are inoffensive inclusions, but you likely won’t look at them twice unless you want to get your grandma into Street Fighter, or something.
The Joy (Con) of fighting
(Image credit: Capcom)
How I feel about Street Fighter 6 now hasn’t changed much over the last couple of years. It’s still one of the best fighting games available today, with plenty of content for both casual and hardcore players to sink their teeth into.
That being said, there may be some growing pains for those jumping over to the Switch 2 version, or playing the game for the first time on the new hardware. While the Nintendo Switch 2 Pro Controller is a fine fit for Street Fighter 6’s Classic control scheme (which grants full access to characters’ moves and full motion inputs), the Joy-Con 2 certainly isn’t.
That’s because it’s almost impossible to reliably input directional special moves – especially the likes of the Dragon Punch motion and Zangief’s full-circle Spinning Pile Driver – on the Joy-Con 2’s button-styled d-pad.
However, if you prefer the Modern control scheme, which simplifies special moves to a single direction, you’ll have a much easier time if all you can play with is the Joy-Con 2. And it still helps that Modern is an incredibly well-designed control scheme that’s even preferred by no shortage of professional Street Fighter 6 players, so you absolutely can hop online and learn the game efficiently in this way.
Small world
(Image credit: Capcom)
As for World Tour, Street Fighter 6’s marquee single-player mode, it’s certainly something I’ve still got a soft spot for – but it’s definitely not at its best on Switch 2. In case you’re unaware, World Tour has your player-created avatar exploring 3D environments, getting into fights, and meeting legendary Street Fighter characters in order to level up and learn new moves – all in aid of winning tournaments and becoming the strongest fighter in the world. It’s not without its faults, with plenty of incredibly tanky enemies who love employing cheap tactics like lobbing wrenches from across the screen, but it is a fun diversion from the chaos of the online suites.
While it was never a top performer even on PS5, Xbox, and PC, World Tour on Switch 2 does suffer from some compromises. Draw distance is quite poor, and there’s plenty of choppy animation on distant NPCs. Overworld exploration targets 60fps and does get there frequently, but actual battles are strictly locked to 30fps, which simply feels horrible in a fighting game – even in a more casual-friendly environment.
I’d still say World Tour is worth a playthrough if you haven’t tried it before, as the experience here is far from unplayable. Just be prepared for a massive disconnect in performance between it and the rest of the package, which manages to be incredibly stable on Switch 2.
Should I play Street Fighter 6 Years 1-2 Edition on Switch 2?
Play it if…
Don’t play it if…
Accessibility
Street Fighter 6’s accessibility suite is just as impressive on Switch 2 as it is on other consoles and PC. For one, the Modern control scheme is an excellent onboarding tool for players intimidated by the complicated Classic layout. Modern doesn’t give access to a character’s full move list, but it does greatly simplify special move inputs and provides easy combo routes, allowing you to hop online and see success almost right away.
In terms of other accessibility options, there are plenty. Audio options are particularly impressive, with the ability to enable a ‘distance to opponent’ sound to aid players with vision impairments. You can also enable unique sound effects for high, mid, and low attacks, as well as ones for remaining health and Super Art gauge levels.
Sadly, there are no colorblind settings to speak of, and on Switch 2, there are fewer graphical and visual options in general than on other systems.
How I reviewed Street Fighter 6 Years 1-2 Edition
I played Street Fighter 6 Years 1-2 Edition on Nintendo Switch 2 for around 28 hours for this review. That included significant progress in World Tour mode, as well as several rounds of Arcade mode, combo trials, and online matches, both ranked and casual.
In docked mode, playing on an LG CX OLED TV, my controller of choice was the Nintendo Switch 2 Pro Controller. I also tested the game in handheld mode using the Joy-Con 2 controllers.
As a huge fighting game fan, I also reviewed the original Street Fighter 6 release on PS5. I’ve also provided reviews for Granblue Fantasy Versus: Risingand Fatal Fury: City of the Wolves, so I feel confident in delivering a review of the Switch 2 port of Street Fighter 6 that’s fair and coming from a standpoint of whether or not it’s worth your time and money.
Google Wallet is more than Google’s tap-to-pay app. Over the years, Wallet has expanded into a sort of all-in-one digital document manager for anything and everything you might keep in a real wallet — transit passes, event tickets, store loyalty cards, and more.
Last August, Wallet got support to create custom passes for pretty much anything, and more recently, Google rolled out the ability to edit those passes without having to completely recreate them from scratch. If you haven’t interacted with this Wallet’s custom pass functionality much, there’s quite a bit to wrap your head around. Here’s everything you need to know.
What is a custom pass?
Google Wallet’s got bespoke onboarding for payment cards, transit passes, loyalty cards, gift cards, and IDs. For anything else, you can use an Everything else option to create a custom pass.
When you add a pass through this method, Wallet will do its best to automatically detect the type of document you’re uploading to give the resulting digital pass relevant artwork. It can detect event tickets, insurance cards of multiple types, loyalty cards, student IDs, business cards, and more. Any document that doesn’t fall into one of Wallet’s predefined buckets will be added as “Other.”
Related
12 things to add to Google Wallet that aren’t your credit card
Use Google Wallet for more than payments
How to create a custom pass
Creating a custom pass only takes a few steps, and Google Wallet does most of the work for you. The main thing you have to do manually is scan the document using your phone’s camera. Here’s how it works.
Open the Google Wallet app.
Tap the Add to Wallet button in the lower right corner.
Tap Everything else.
Line your document up in the viewfinder and tap the capture button.
On the next screen, you can crop, rotate, or retake the photo. If what you got looks good, tap Done.
Wallet will take a moment to process, then show the resulting pass.
Verify the info on your custom pass is correct. If it’s right, tap Add. If it’s incomplete or inaccurate, tap Edit.
Additional information fields added in the Edit view in step 7 above aren’t displayed on the custom pass graphic. To view any information added manually, tap the three-dot icon in the top right corner when viewing your pass.
How to edit a custom pass
Google just recently introduced the option to edit custom passes after adding them to Wallet, saving you the hassle of reuploading a pass if you spot an error or omission. Here’s how to do that.
Open the Google Wallet app.
Tap the pass you want to edit.
Tap the three-dot icon in the upper right corner.
Scroll down and tap Edit.
Make your changes, then tap Save.
You’re also able to add additional information fields in the edit view, in case the initial upload missed a detail like your vehicle identification number on a car insurance card.
Related
Google Wallet vs. Samsung Wallet: Which is better?
Which digital wallet should you be using?
An easy way to digitize all your docs
Google Wallet’s custom pass functionality makes the app infinitely more useful, and Google’s recent decision to allow users to edit documents after uploading them makes it even more convenient still. It’s a great way to keep documents you might occasionally need at hand without actually needing to carry around all that paper.
WhatsApp developing ‘search draft messages’ feature for iPhone users – Daily Times
We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.OkPrivacy policy
This study is the first to examine the psychometric properties and factor structure of the sexual empowerment sub-scale of the WGE-SRH Index across ten culturally diverse settings. Overall, results suggest the scale is a reliable tool for understanding how women frame and act on sexual decisions across diverse cultures in sub-Saharan Africa. After dropping two items, psychometric properties were strong in most sites, except Rajasthan, India, where item fit remained slightly below desired thresholds.
Findings confirm a two-factor structure of sexual empowerment (existence of choice and exercise of choice), similar to that originally identified by Moreau and colleagues [17]. Notably, however, some items performed differently across contexts. In Rajasthan, India, in particular, full models showed that the item,“If I show my husband/partner that I want to have sex, he may consider me promiscuous” loaded strongly onto the existence of choice domain, while the item, “If I do not want to have sex, I am capable of avoiding it with my husband/partner” loaded strongly onto exercise of choice. In contrast, factor loadings for these two items showed considerable variability in sub-Saharan African settings, with the promiscuity item underperforming in the majority of contexts. Similarly, the “avoid” item did not meet desired thresholds in several sites. As such, both items were dropped from final models to ensure consistency across sub-Saharan African sites, however, these items may still have relevance and be considered as critical aspects of sexual empowerment within the context of research in Asia, where commonality of extramarital relationships and implications of promiscuity may differ [32,33]. Combining the sexual existence of choice and sexual exercise of choice domains into a single measure resulted in a multidimensional index of sexual empowerment with moderate to good internal reliability (alpha range 0.59 in Kongo Central, DRC to 0.69 in Niger), along with evidence of goodness of fit across sites, withstanding Rajasthan, India.
Understanding cross-site differences in existence vs. exercise of choice is relevant for the implementation of interventions and programs seeking to counteract harmful gender norms. Marked variations were seen across sites in women’s perceptions of their partners’ responses to refusing sex, as well as their own confidence in voicing when and when not to have sex. Niger, specifically, stood out as a site with overall low exercise of choice; however, this setting was comparable to other sites for existence of choice items. Such contrast between existence and exercise of choice may be indicative of where a population lies on the empowerment pathway [15]. Of note, Niger has the highest prevalence of child marriage globally [27]. While many contexts have overcome gender and power barriers related to existence of choice surrounding sex, many women still may not feel comfortable exercising this choice. In such contexts, programs and interventions should focus on supporting women’s ability to communicate and negotiate their sexual choices, rather than normative factors prohibiting women’s sexual needs in the first place. Community dialogues, such as those implemented for intimate partner violence, may be useful to support communication and negotiation, while concurrently working with men and the broader community [34,35,36]. As opposed to contraceptive empowerment, where women can exercise their choice to use contraception via covert contraceptive use without first having the existence of choice [30,37], sex requires an interaction between sexual partners and, therefore, both existence and exercise of choice must co-exist for women to achieve volitional sex. These results point to the importance of disentangling the domains of existence vs. exercise of choice for sexual empowerment.
Identifying within-site variation is also necessary to determine where empowerment interventions are needed or may have fallen short. Some items were particularly polarizing within contexts—specifically, the “promiscuous” item had the majority responses concentrating in strongly agree or strongly disagree categories, with few women indicating neutral empowerment (i.e., neither agreeing nor disagreeing). These polarized responses persisted for exercise of choice items across contexts and were particularly pronounced for items focusing on women’s desire to have sex rather than not to have sex. Such results highlight the sexual double standards between men and women reported in previous studies [15,38,39,40] and speak to the undervaluing of women’s sexual pleasure as a key component of their sexual health [41].
These cross-cultural findings further elucidate that sexual empowerment is largely constrained—when examining our overall sexual empowerment scores, less than half of the sites reported the majority of women with high or highest empowerment scores (Lagos, Nigeria: 74.5%, Kenya: 58.4%, Kinshasa, DRC: 57.7%, and Rajasthan, India: 54.7%). In sites where existence and exercise of choice were in opposition (i.e., Niger, Kano, Nigeria, and Kongo Central, DRC), overall empowerment concentrated in the neutral category. Notably, these lower empowerment sites likely represent different stages of gender norms programs (i.e., focus on existence of choice rather than exercise of choice) and should continue to be examined.
This study has several strengths, including distinguishing between sexual existence and exercise of choice and its use of ten population-based samples to validate a sexual empowerment sub-scale and compare sexual empowerment levels across sites. Nonetheless, findings should be considered in light of some limitations. Specifically, item wording pertains to a woman’s current husband/partner, precluding understanding of how these items apply within concurrent partner relationships. Women may feel empowered sexually with some partners, but not with others. Within-site variation in sexual empowerment was not examined to afford fuller exploration of sexual empowerment across populations, however, given cultural heterogeneity in practices within many study sites, this should be examined in future studies. Reasons for sexual empowerment within and across sites, as well as an understanding of who is empowered, as additionally needed. Lastly, it is noted that only one small site within Asia was included in this study (Rajasthan, India) and is not generalizable to the whole of Asia nor the whole of India.
The sixth generation of the Fairphone repairable mobile was launched at the end of June. Now spunger-flingers iFixit have got their hands on it, and liked the result.
The pleasure of an easy charge port swap (click to enlarge) – image courtesy iFixit
The iFixit site is a top resource for most of us trying to repair or upgrade a computer – especially if it’s a Mac. Fewer dare to try to fix their own phones, but iFixit reckons you’ll be able to if you spring for a Fairphone, and the newest model gets a perfect 10.
Remove two screws to slide the back off. You will need a screwdriver to swap the battery, though — no more tool-free battery swaps. It’s now held in by four screws and a cable. (It is, however, glued onto a metal backing strip, replaced with it.). Eight more screws hold the screen in place. In part because it’s not glued together, it only has an IP55 rating for dust and water resistance, not the IP68 some devices boast, but that’s still not bad. A nifty touch is a modular USB-C connector, which you can replace on its own if damaged. In total, 12 components are replaceable.
Away from the tech specs, iFixit gives it still more credit: the warranty is an impressive five years, with updates and spares available until 2033, including a promised seven Android version upgrades. That’s hard to beat. The theory is that your new Fairphone should last much longer than most other phones. If you keep using it for five years, that should lower its carbon footprint by nearly a third.
Fairphone’s Generation 6 handset launched on 25 June. GSMArena has full specs. We don’t have one to play with, but it looks like a decent enough mobe: 6.31 inch screen, 8 GB of RAM and 256 MB of storage plus a microSD slot, and a 50 MP rear camera. It’s €599, which translates to $870 or £535.
Headache-free screen swap (image courtesy iFixit)
The manufacturer says it’s Fairphone’s most sustainable ever. Half of it is made from recycled or ethically-sourced materials, which genuinely does matter. As The Register pointed out way back in 2001, many of the important materials in phones come from ecologically disastrous sources. In 2010, El Reg looked at the problems of “coltan” and the “3Ts plus gold” from the Eastern Congo region. Avoiding that, recycling the substances wherever and whenever possible, is seriously important, both to people and to mountain gorillas, as well as less charismatic wildlife.
All that means that it costs more, though. It’s a perfectly capable phone, but you don’t get an amazing spec for the money. It has a slightly smaller screen than its predecessor, the Fairphone 5, which incidentally also got a perfect 10 score a couple of years ago. There’s no headphone socket, and only one physical SIM slot, although it can support an eSIM as well. The battery is slightly more capacious at 4415 mAh, and with a 30 W charger it can get to half full in half an hour. As with most phones now, this happens through a USB-C port – but if you hook it up to your PC, this only operates at USB 2 speeds.
There is a big lime-green colored switch on the side that enables what the company calls Fairphone Moments mode which strips the UI down to five apps, reduces digital distractions, mutes the on-screen colors and so on. There is a built-in Google Gemini LLM-bot “assistant”, but you can turn it off.
If you, like the Reg FOSS desk, are powerfully averse to such gimmicks, the same phone is also available with Murena /e/ OS instead. We looked at version 3 last month, and it does allay a lot of Google-related privacy concerns, at the price of some features and convenience. With the Fairphone, there’s a less metaphorical price, too: it costs €150 more than the base model. ®
You know your software is a success when its name becomes a verb: You’ll now commonly hear about images being photoshopped, even if the editing wasn’t done with the Adobe image editor. Adobe might not like it, but the usage shows how dominant its flagship product has become.
On mobile though, Photoshop hasn’t achieved the same kind of ubiquity or brand recognition. We’ve had official Photoshop apps of various types down the years, but none of them have really translated the power and feature set of the desktop application over to smaller devices in a way that fully works.
Now, Adobe is trying again: The all-new Photoshop app (available now for iOS, and just released in beta for Android), is Photoshop “reimagined” for a smartphone. The app attempts to bring over as many of the tools and features of the full software as it can, while optimizing them for use on a mobile touchscreen.
While it’s impossible to port everything over—in the same way macOS wouldn’t work on an iPhone—there’s a lot to explore in the new Photoshop app. You can download it for free on iOS, though some features require a premium subscription (either $7 a month for the new Mobile and Web plan, or an existing Photoshop plan). Android users can download the beta version and use it entirely for free for a limited time. Regular pricing will kick in when the app becomes generally available worldwide.
Quick Actions
By default, when you open up a new image in the iOS version, you’ll be invited to apply a “quick action” to it. (Android users, this is coming soon.) If you’d rather just go straight to the full editing interface every time, check the Don’t show this again box, and tap Go to editor. However, if you are wanting to apply a basic effect in as little time as possible, these quick actions can be helpful.
You’ll see they all involve backgrounds. Tap Hide background, and Photoshop tries to identify the main subject of the image, and cut out the rest—you can then drop in any kind of replacement background you like. Alternatively, there’s Solid color background, which drops in a color you can edit, or Black & white background, which keeps the subject in color but turns everything else grayscale.
Whichever one you pick, Photoshop will take care of creating a new layer for you, so you can work on the foreground and background independently. You’ll see the layers are available via the thumbnail down in the lower right corner of the image. With the solid color background, for example, tap the layer thumbnail, then the left arrow, then Edit color to switch to a different shade.
Layers and Masks
Choosing an adjustment layer.
Courtesy of Michael Calore / WIRED
Layers and masks are essentially what separates basic image editors from advanced ones, letting you stack different elements in an image on top of each other (layers) and make edits to them (masks). If you’ve used a quick action you’ll already have a layer created for you, but you can create new ones by tapping the + (plus) button just above the bottom toolbar on the right.
Large language models (LLMs) are becoming less “intelligent” in each new version as they oversimplify and, in some cases, misrepresent important scientific and medical findings, a new study has found.
Scientists discovered that versions of ChatGPT, Llama and DeepSeek were five times more likely to oversimplify scientific findings than human experts in an analysis of 4,900 summaries of research papers.
When given a prompt for accuracy, chatbots were twice as likely to overgeneralize findings than when prompted for a simple summary. The testing also revealed an increase in overgeneralizations among newer chatbot versions compared to previous generations.
The researchers published their findings in a new study April 30 in the journal Royal Society Open Science.
“I think one of the biggest challenges is that generalization can seem benign, or even helpful, until you realize it’s changed the meaning of the original research,” study author Uwe Peters, a postdoctoral researcher at the University of Bonn in Germany, wrote in an email to Live Science. “What we add here is a systematic method for detecting when models generalize beyond what’s warranted in the original text.”
It’s like a photocopier with a broken lens that makes the subsequent copies bigger and bolder than the original. LLMs filter information through a series of computational layers. Along the way, some information can be lost or change meaning in subtle ways. This is especially true with scientific studies, since scientists must frequently include qualifications, context and limitations in their research results. Providing a simple yet accurate summary of findings becomes quite difficult.
“Earlier LLMs were more likely to avoid answering difficult questions, whereas newer, larger, and more instructible models, instead of refusing to answer, often produced misleadingly authoritative yet flawed responses,” the researchers wrote.
Get the world’s most fascinating discoveries delivered straight to your inbox.
Related: AI is just as overconfident and biased as humans can be, study shows
In one example from the study, DeepSeek produced a medical recommendation in one summary by changing the phrase “was safe and could be performed successfully” to “is a safe and effective treatment option.”
Another test in the study showed Llama broadened the scope of effectiveness for a drug treating type 2 diabetes in young people by eliminating information about the dosage, frequency, and effects of the medication.
If published, this chatbot-generated summary could cause medical professionals to prescribe drugs outside of their effective parameters.
Unsafe treatment options
In the new study, researchers worked to answer three questions about 10 of the most popular LLMs (four versions of ChatGPT, three versions of Claude, two versions of Llama, and one of DeepSeek).
They wanted to see if, when presented with a human summary of an academic journal article and prompted to summarize it, the LLM would overgeneralize the summary and, if so, whether asking it for a more accurate answer would yield a better result. The team also aimed to find whether the LLMs would overgeneralize more than humans do.
The findings revealed that LLMs — with the exception of Claude, which performed well on all testing criteria — that were given a prompt for accuracy were twice as likely to produce overgeneralized results. LLM summaries were nearly five times more likely than human-generated summaries to render generalized conclusions.
The researchers also noted that LLMs transitioning quantified data into generic information were the most common overgeneralizations and the most likely to create unsafe treatment options.
These transitions and overgeneralizations have led to biases, according to experts at the intersection of AI and healthcare.
“This study highlights that biases can also take more subtle forms — like the quiet inflation of a claim’s scope,” Max Rollwage, vice president of AI and research at Limbic, a clinical mental health AI technology company, told Live Science in an email. “In domains like medicine, LLM summarization is already a routine part of workflows. That makes it even more important to examine how these systems perform and whether their outputs can be trusted to represent the original evidence faithfully.”
Such discoveries should prompt developers to create workflow guardrails that identify oversimplifications and omissions of critical information before putting findings into the hands of public or professional groups, Rollwage said.
While comprehensive, the study had limitations; future studies would benefit from extending the testing to other scientific tasks and non-English texts, as well as from testing which types of scientific claims are more subject to overgeneralization, said Patricia Thaine, co-founder and CEO of Private AI — an AI development company.
Rollwage also noted that “a deeper prompt engineering analysis might have improved or clarified results,” while Peters sees larger risks on the horizon as our dependence on chatbots grows.
“Tools like ChatGPT, Claude and DeepSeek are increasingly part of how people understand scientific findings,” he wrote. “As their usage continues to grow, this poses a real risk of large-scale misinterpretation of science at a moment when public trust and scientific literacy are already under pressure.”
For other experts in the field, the challenge we face lies in ignoring specialized knowledge and protections.
“Models are trained on simplified science journalism rather than, or in addition to, primary sources, inheriting those oversimplifications,” Thaine wrote to Live Science.
“But, importantly, we’re applying general-purpose models to specialized domains without appropriate expert oversight, which is a fundamental misuse of the technology which often requires more task-specific training.”
“KTC defines the MegPad A32Q7Pro as a portable, smart, touchscreen monitor,” KTC’s spokesperson told me. “It combines key traits of a smart display and a large-screen tablet. While it shares some features with smart TVs, tablets, and monitors, it doesn’t fully belong to any single traditional category. It’s a hybrid device designed to bridge those use cases.”
Android tablets on wheels
Many devices like the MegPad represent a push for more Android-powered, non-Google devices that has been buoyed by a program that Google launched in 2022, the Enterprise Devices Licensing Agreement (EDLA).
As explained by partners like BenQ, EDLA is a way for third parties to incorporate Google Mobile Services (GMS), which are Google’s most commonly used apps and APIs bundled for use across different types of devices. GMS apps include popular software like Google Drive, Gmail, the Google Play Store, and YouTube.
“Previously, GMS was only officially available for smartphones, tablets, TVs, and wearables. Under the new EDLA, the list of devices eligible for GMS certification has now been expanded to include enterprise solutions such as smart boards,” a blog from BenQ, which has EDLA-certified smart displays, reads.
Since 2022, (the year LG’s StanbyME launched), there has been an uptick in non-Google devices with this EDLA certification. One of the categories taking advantage of the newer program is tablets on wheels, like the MegPad and similar options from Kefeya, Apolosign, Innocn, and DuraPro.
Demonstrating the marketing value of EDLA certification, the MegPad’s product page reads: “Google EDLA certification provides secure, direct access to Google services and the Google Play Store with regular updates, offering greater stability and data protection than open app ecosystems with unverified apps.”
Most EDLA-certified devices seem to be interactive displays used for education. With EDLA certification, devices like the MegPad may also draw the attention of educators or even businesses. Meanwhile, Google is happy to hand out EDLA certifications, as they can drive Android adoption, giving Google more data and access to customers outside of the typical Android devices, such as phones. Products like the MegPad can also be easier to shop with (Google loves when people use its offerings to shop) than Android devices with smaller screens.