No matter how much we don’t like and oppose it, personal data is now a commodity. Our phone numbers, addresses, shopping habits, or employment history details are collected, analyzed, and traded among data brokers, marketers, recruiters, insurers, and countless other buyers, not to mention frauds and thieves.
However, trying to remove your online presence manually means tracking down every single company that holds your data (which can be hundreds), submitting legal deletion requests, and repeating the process when your data reappears or your request is ignored. This can easily become a full-time job.
That’s why data broker removal services exist: to automate, manage, and repeat those requests on your behalf.
But how to choose the best provider? Below, you will find a 2026 evaluation of the most recognized names in the industry.
Top Data Broker Removal Services at a Glance
Category
Incogni
Aura
DeleteMe
Optery
OneRep
Pricing (monthly when billed annually)
From $7.99
From $9.99
From $6.97
From $3.25
From $8.33
Free option
30-day money-back guarantee
14-day free trial, 60-day money-back guarantee
Free scan
Basicself-service, 30-day money-back guarantee
5-day trial, 30-day money-back guarantee
Automation Level
High
Medium-High
Medium-Low
Medium
Medium-High
Broker coverage
420+ public and private brokers
200+ brokers, mainly private
up to 850+ brokers (varies by plan), mainly public
120-640+sites (varies by plan)
310+ sites, mainly public
Verification
Dashboard, Deloitte Limited Assurance Report
App alerts and screenshots
Quarterly reports and screenshots
Screenshots and exposure scans
Dashboard and monthly reports
Best for
Long-term, low-effort privacy
Identity + privacy bundle
Detailed proof and control
Data exposure prediction
Public removals, Families
Incogni: Best for Balanced Automation, Coverage, and Accountability
Overview and Pricing
Incogni focuses on the continuous removal of personal data from data brokers, including both public people-search sites and private commercial databases.
Incogni’s plans start at $7.99/month when billed annually, and even the basic option contains all you need for effective data removal. Higher-tier plans only change prioritization and scope. There’s no free option, but you can take advantage of its 30-day money-back guarantee to see if the service suits your needs. Read our full Incogni review for more details.
Features
Fully-automated opt-out and deletion requests across 420+ data brokers
Recurring removal cycles: 60 days for public, 90 days for private brokers
Operational processes audited via a limited assurance report by Deloitte
Effectiveness
Supported by Deloitte’s limited assurance assessment, Incogni officially reports that it has processed 245+ million removal requests from 2022 to mid-2025, indicating sustained operations rather than one-time cleanups. As data brokers can reacquire information and their databases refresh regularly, the recurring cycle is vital if you want to protect your online footprint in the long run.
Transparency and Reputation
Apart from a limited assurance report by Deloitte, the service also holds Editors’ Choice Awards from PCMag and PCWorld, which praise its automation system and wide coverage.
On Trustpilot, Incogni has generally positive feedback, with an average rating of 4.4 based on over 2,000 reviews. Users often note actual reductions in spam and visible listings.
User Experience
Once you set up your account, you need to verify your identity. After that, Incogni will handle most data removal activity in the background without involving you directly. The clear, straightforward dashboard will show you all the brokers Incogni has contacted, confirmed removals, responses, and next scheduled cycles. You can peek into it whenever you like, but you don’t have to engage to make the process effective.
Advantages
Disadvantages
High automation
No screenshots
Broad coverage
No free trial
Deloitte Limited Assurance Report
Basic reporting
30-day money-back guarantee
Phone support only on Unlimited plans
Industry recognition
Recurring cycles and resubmitted requests
Clear interface, straightforward user experience
Aura: Best All-in-One Identity and Privacy Suite
Overview and Pricing
Aura is not a provider like others on this list, as it combines data removal service with broader digital protection features, including credit alerts, antivirus, VPN, device security, and identity theft monitoring.
Aura’s prices begin at $9.99/month when billed annually. What’s more, you get a 14-day free trial and a 60-day money-back guarantee for risk-free testing.
Features
Automated data removal across 200+ data brokers (mainly private)
Identity theft monitoring
Dark web monitoring
Credit score and breach alerts
Antivirus/anti-malware protection
VPN
Family and multi-device plans
Effectiveness
When it comes to data removal itself, this Aura functionality is automated. The platform first scans broker and people-search sites, submits deletion requests whenever finding your
information, and re-checks for reappearances. However, as it’s not its main focus, its data removal coverage is quite narrow compared to dedicated solutions. Aura’s value is the strongest only if combined with the whole toolkit.
Transparency and Reputation
Aura has been widely described in the identity protection space with overall positive sentiments. You can find Aura reviews on PCMag, Forbes, and NerdWallet. On Trustpilot, it holds an average rating of 4.2 based on almost 1,000 reviews. Users appreciate its all-in-one service, but broker removal results themselves don’t match those ensured by services focused exclusively on that problem.
User Experience
Aura’s interface contains all the features offered by the providers, showing alerts, scans, security postures, removal status, and more. This holistic view appeals to people who seek central management of their online presence, but for many users, it can be overwhelming.
Advantages
Disadvantages
Privacy+security bundle
Narrower coverage
Insurance
Manual approval steps
60-day money-back guarantee
Overwhelming user experience
14-day free trial
No third-party verification
Comprehensive alerts
DeleteMe: Strong for Proved Public People-Search Listing Deletion
Overview and Pricing
DeleteMe focuses on public people-search sites and background information databases. These are mentions that usually appear in search results when someone Googles your name.
The cheapest DeleteMe plan is $6.97/month when billed annually and can be used by 1 person.
Features
Automated scans of people-search sites (up to 850+, depending on the plan)
Expert manual handling
Quarterly detailed reports
Coverage for individuals, couples, and families
Limited custom removal requests (40-60 per year, plan-dependent)
DIY opt-out tutorials
Effectiveness
DeleteMe is quite effective at removing visible information from many major public listings. The company was a pioneer when, in 2010, it entered the industry with its
part-automatic, part-human-assisted approach. The team submits requests and tracks
progress, then provides you with scheduled, detailed reports that include, for example, even screenshots.
Transparency and Reputation
DeleteMe has been in the industry since 2010, which says a lot about its reliability. It has generally positive user reviews, especially when it comes to its detailed reporting system and exhaustive explanations about what was removed. There have been no third-party assessments of its services, but the provider has a good reputation in the industry, as seen in the review in PCMag or praise from Forbes. When it comes to user feedback, it has a rating of
4.0 on Trustpilot, though based only on 180+ reviews.
User Experience
Contrary to Incogni’s live and always-on progress monitoring, which you can check but don’t have to, DeleteMe is more report-centric. Users receive quarterly PDF summaries that show what sites were contacted, where their information was removed, and what remains pending.
Many people appreciate their human approach.
Advantages
Disadvantages
Clear, detailed reporting
Slower cycles
Long-standing service
Less automation
Human expertise
Narrower broker reach
30-day money-back guarantee
US-mainly coverage
Optery: Best for Exposure Visibility
Overview and Pricing
Optery’s main field of expertise is discovering where your personal data exists, providing users with insight into exposures before and during removal attempts.
Optery’s offer starts at $3.25/month when billed annually. The company also has a free, self-service version. Apart from that, you get a free scan and a 30-day money-back guarantee.
Features
Exposure dashboard displaying where your personal data exists
Automated removal from up to 630+ brokers with paid plans
Initial free scan across 120+ sites and free self-service plan
Guided removal request sending process
Custom removal submissions
Manual tracking of opt-outs and their status
Effectiveness
Optery is most effective at identifying where your personal data has been exposed. Then, for its removal, it blends automatic attempts with user-guided actions and manual tracking.
It doesn’t have the same automated recurring cycles as, for example, Incogni, but it may be helpful if you want to truly understand data exposures.
Transparency and Reputation
Optery is often highlighted for its exposure insights and transparency. Users appreciate the “seeing where my data lives” model, but many note that broader coverage comes only with more expensive plans, while manual user input is still needed.
On Trustpilot, Optery has 171 reviews with an average rating of 4.1. It has also been reviewed by PCMagquite enthusiastically, though they mentioned that the service doesn’t distinguish between removed data and never-found data. TechRadarpraised it for its ease of use.
User Experience
Optery is more interactive and gives you more control of the process (which can be both an advantage and a disadvantage, depending on how much time you’re willing to sacrifice). Its dashboard clearly shows where your personal data is, and then you need to decide which removals are more important and what to do next. You also get before and after screenshots as visual proof, while reports are AI-improved to make them more accurate and detailed.
Advantages
Disadvantages
Free scan
Broader coverage with more expensive plans
Free self-service
US-focused
30-day money-back guarantee
Slower with cheaper plans
Clear interface & control
No phone support
Onerep: Best for Public Listing Removal and Families
What It Does
OneRep automates removal requests issued to public people-search sites. Its focus is on high-risk databases like Intelius and Whitepages. The service also ensures quarterly recurring checks to combat resurfacing of your data.
However, there’s significant controversy around the company (more of that below).
Onerep’s prices start at $8.33/month when billed annually. It also offers a 5-day free trial. What makes it attractive and more affordable is its family plans that cover up to 6 members.
Features
Automated scans and removal requests across 310+ data brokers
Quarterly re-scanning
Great family value
Clear and straightforward dashboard tracking
Effectiveness
Optery is effective when it comes to reducing online visibility on many public sites, including those deemed high-risk. However, this provider doesn’t focus on private commercial
brokers that are responsible for a large portion of the spam. It makes Optery’s reach much narrower.
Transparency and Reputation
OneRep has a mixed reputation in the privacy protection community.
User reviews vary: some praise successful public listing removals, while others complain about slow relisting or only partial effects. Still, it holds a quite impressive average rating of
4.7 on Trustpilot based on almost 400 reviews.
However, it’s essential to know that Krebs on Security revealed that in March 2024, Mozilla decided to drop OneRep from its list of recommendations due to the company’s CEO’s involvement in running people-search networks. This raised serious questions about conflict of interest in the industry. While the provider stated that Onerep operates completely independently and never sells user information, it is still often referenced in privacy circles.
User Experience
Onerep’s dashboard is pretty simple to manage. It shows progress on targeted sites and all removal requests, though it’s not really an automated model, so it only suits users who don’t mind handling the process.
Advantages
Disadvantages
Great family value
Industry controversies
5-day free trial
30-day money-back guarantee highly conditioned
Quarterly re-scans
US focus
Public listing coverage
Little customization
No third-party verification
Narrower scope
Final Perspective for 2026
When it comes to choosing a data removal service, the main difference is usually in scope and depth. Some providers focus on visible people-search listings, while others dig deeper to find your personal information in harder-to-find databases. They also vary in the recurring cycles they offer (or not).
Managing your overall online visibility is vital, but if you really want to reduce the amount of your information circulating on the web, you need to focus on less visible broker networks. Or rather, choose a provider built around large-scale broker coverage. Only then will you be able to enjoy more sustained results.
In 2026, Incogni stands out among its competition, as it combines a wide broker reach, continuous removal cycles, and a streamlined, low-maintenance experience. Not to mention that it was independently assessed. While other providers are not to be altogether dismissed, Incogni’s focused, automated approach offers the most comprehensive way out.
FAQ
Why can’t I just remove my data from brokers myself?
Manual removal means identifying hundreds of brokers, submitting individual opt-out requests, repeatedly verifying your identity, and rechecking when your data reappears. For most people, that quickly becomes too time-consuming to manage consistently.
How often does my data reappear after removal?
Data brokers regularly refresh and repurchase data, which means listings can resurface even after deletion. That’s why recurring removal cycles are critical for long-term results.
What’s the difference between public and private data brokers?
Public brokers (like people-search sites) display your information in search results, while private brokers trade data behind the scenes with marketers, insurers, and other businesses. Private databases often contribute more to spam and profiling, even if you don’t see them.
Do all services provide proof that removals were completed?
No. Some providers offer screenshots or quarterly reports, while others rely on dashboards or summary updates. The level of transparency varies significantly by service.
Is a bundled identity protection service enough for data removal?
All-in-one tools can help, but their broker coverage is often narrower than services dedicated specifically to data removal. If reducing online exposure is your main goal, specialized coverage may deliver stronger results.
Source link
#Data #Broker #Removal #Services #Reduces #Online #Exposure
#Reggie #FilsAimé #Amazon #asked #Nintendo #break #lawAmazon,Gaming,News,Nintendo,Tech">Reggie Fils-Aimé says Amazon once asked Nintendo to break the law
“Literally, we stopped selling to Amazon, and it’s because I wasn’t going to do something illegal. I wasn’t going to do something that would put at risk the relationship we have with other retailers. But it also set the stage to say, look, you’re not going to push me around. This is the way we do business. And so that’s how, over time, you build respect.”
The study was published this week in Science and comes from a research team led by physicians and computer scientists at Harvard Medical School and Beth Israel Deaconess Medical Center. The researchers said they conducted a variety of experiments to measure how OpenAI’s models compared to human physicians.
In one experiment, researchers focused on 76 patients who came into the Beth Israel emergency room, comparing the diagnoses offered by two internal medicine attending physicians to those generated by OpenAI’s o1 and 4o models. These diagnoses were assessed by two other attending physicians, who did not know which ones came from humans and which came from AI.
“At each diagnostic touchpoint, o1 either performed nominally better than or on par with the two attending physicians and 4o,” the study said, adding that the differences “were especially pronounced at the first diagnostic touchpoint (initial ER triage), where there is the least information available about the patient and the most urgency to make the correct decision.”
In Harvard Medical School’s press release about the study, the researchers emphasized that they did not “pre-process the data at all” — the AI models were presented with the same information that was available in the electronic medical records at the time of each diagnosis.
With that information, the o1 model managed to offer “the exact or very close diagnosis” in 67% of triage cases, compared to one physician who had the exact or close diagnosis 55% of the time, and to the other who hit the mark 50% of the time.
“We tested the AI model against virtually every benchmark, and it eclipsed both prior models and our physician baselines,” said Arjun Manrai, who heads an AI lab at Harvard Medical School and is one of the study’s lead authors, in the press release.
Techcrunch event
San Francisco, CA|October 13-15, 2026
To be clear, the study didn’t claim that AI is ready to make real life-or-death decisions in the emergency room. Instead, it said the findings show an “urgent need for prospective trials to evaluate these technologies in real-world patient care settings.”
The researchers also noted that they only studied how models performed when provided with text-based information, and that “existing studies suggest that current foundation models are more limited in reasoning over nontext inputs.”
Adam Rodman, a Beth Israel doctor who’s also one of the study’s lead authors, warned the Guardian that there’s “no formal framework right now for accountability” around AI diagnoses, and that patients still “want humans to guide them through life or death decisions [and] to guide them through challenging treatment decisions.”
In a post about the study, Kristen Panthagani, an emergency physician, said this is an “an interesting AI study that has led to some very overhyped headlines,” especially since it was comparing AI diagnoses to those from internal medicine physicians, not ER physicians.
“If we’re going to compare AI tools to physicians’ clinical ability, we should start by comparing to physicians who actually practice that specialty,” Panthagani said. “I would not be surprised if a LLM could beat a dermatologist at an neurosurgery board exam, [but] that’s not a particularly helpful thing to know.”
She also argued, “As an ER doctor seeing a patient for a first time, my primary goal is not to guess your ultimate diagnosis. My primary goal is to determine if you have a condition that could kill you.”
This post and headline have been updated to reflect the fact that the diagnoses in the study came from internal medicine attending physicians, and to include commentary from Kristen Panthagani.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
The study was published this week in Science and comes from a research team led by physicians and computer scientists at Harvard Medical School and Beth Israel Deaconess Medical Center. The researchers said they conducted a variety of experiments to measure how OpenAI’s models compared to human physicians.
In one experiment, researchers focused on 76 patients who came into the Beth Israel emergency room, comparing the diagnoses offered by two internal medicine attending physicians to those generated by OpenAI’s o1 and 4o models. These diagnoses were assessed by two other attending physicians, who did not know which ones came from humans and which came from AI.
“At each diagnostic touchpoint, o1 either performed nominally better than or on par with the two attending physicians and 4o,” the study said, adding that the differences “were especially pronounced at the first diagnostic touchpoint (initial ER triage), where there is the least information available about the patient and the most urgency to make the correct decision.”
In Harvard Medical School’s press release about the study, the researchers emphasized that they did not “pre-process the data at all” — the AI models were presented with the same information that was available in the electronic medical records at the time of each diagnosis.
With that information, the o1 model managed to offer “the exact or very close diagnosis” in 67% of triage cases, compared to one physician who had the exact or close diagnosis 55% of the time, and to the other who hit the mark 50% of the time.
“We tested the AI model against virtually every benchmark, and it eclipsed both prior models and our physician baselines,” said Arjun Manrai, who heads an AI lab at Harvard Medical School and is one of the study’s lead authors, in the press release.
Techcrunch event
San Francisco, CA|October 13-15, 2026
To be clear, the study didn’t claim that AI is ready to make real life-or-death decisions in the emergency room. Instead, it said the findings show an “urgent need for prospective trials to evaluate these technologies in real-world patient care settings.”
The researchers also noted that they only studied how models performed when provided with text-based information, and that “existing studies suggest that current foundation models are more limited in reasoning over nontext inputs.”
Adam Rodman, a Beth Israel doctor who’s also one of the study’s lead authors, warned the Guardian that there’s “no formal framework right now for accountability” around AI diagnoses, and that patients still “want humans to guide them through life or death decisions [and] to guide them through challenging treatment decisions.”
In a post about the study, Kristen Panthagani, an emergency physician, said this is an “an interesting AI study that has led to some very overhyped headlines,” especially since it was comparing AI diagnoses to those from internal medicine physicians, not ER physicians.
“If we’re going to compare AI tools to physicians’ clinical ability, we should start by comparing to physicians who actually practice that specialty,” Panthagani said. “I would not be surprised if a LLM could beat a dermatologist at an neurosurgery board exam, [but] that’s not a particularly helpful thing to know.”
She also argued, “As an ER doctor seeing a patient for a first time, my primary goal is not to guess your ultimate diagnosis. My primary goal is to determine if you have a condition that could kill you.”
This post and headline have been updated to reflect the fact that the diagnoses in the study came from internal medicine attending physicians, and to include commentary from Kristen Panthagani.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
#Harvard #study #offered #accurate #emergency #room #diagnoses #human #doctors #TechCrunchbeth israel,harvard medical school,OpenAI">In Harvard study, AI offered more accurate emergency room diagnoses than two human doctors | TechCrunch
A new study examines how large language models perform in a variety of medical contexts, including real emergency room cases — where at least one model seemed to be more accurate than human doctors.
The study was published this week in Science and comes from a research team led by physicians and computer scientists at Harvard Medical School and Beth Israel Deaconess Medical Center. The researchers said they conducted a variety of experiments to measure how OpenAI’s models compared to human physicians.
In one experiment, researchers focused on 76 patients who came into the Beth Israel emergency room, comparing the diagnoses offered by two internal medicine attending physicians to those generated by OpenAI’s o1 and 4o models. These diagnoses were assessed by two other attending physicians, who did not know which ones came from humans and which came from AI.
“At each diagnostic touchpoint, o1 either performed nominally better than or on par with the two attending physicians and 4o,” the study said, adding that the differences “were especially pronounced at the first diagnostic touchpoint (initial ER triage), where there is the least information available about the patient and the most urgency to make the correct decision.”
In Harvard Medical School’s press release about the study, the researchers emphasized that they did not “pre-process the data at all” — the AI models were presented with the same information that was available in the electronic medical records at the time of each diagnosis.
With that information, the o1 model managed to offer “the exact or very close diagnosis” in 67% of triage cases, compared to one physician who had the exact or close diagnosis 55% of the time, and to the other who hit the mark 50% of the time.
“We tested the AI model against virtually every benchmark, and it eclipsed both prior models and our physician baselines,” said Arjun Manrai, who heads an AI lab at Harvard Medical School and is one of the study’s lead authors, in the press release.
Techcrunch event
San Francisco, CA|October 13-15, 2026
To be clear, the study didn’t claim that AI is ready to make real life-or-death decisions in the emergency room. Instead, it said the findings show an “urgent need for prospective trials to evaluate these technologies in real-world patient care settings.”
The researchers also noted that they only studied how models performed when provided with text-based information, and that “existing studies suggest that current foundation models are more limited in reasoning over nontext inputs.”
Adam Rodman, a Beth Israel doctor who’s also one of the study’s lead authors, warned the Guardian that there’s “no formal framework right now for accountability” around AI diagnoses, and that patients still “want humans to guide them through life or death decisions [and] to guide them through challenging treatment decisions.”
In a post about the study, Kristen Panthagani, an emergency physician, said this is an “an interesting AI study that has led to some very overhyped headlines,” especially since it was comparing AI diagnoses to those from internal medicine physicians, not ER physicians.
“If we’re going to compare AI tools to physicians’ clinical ability, we should start by comparing to physicians who actually practice that specialty,” Panthagani said. “I would not be surprised if a LLM could beat a dermatologist at an neurosurgery board exam, [but] that’s not a particularly helpful thing to know.”
She also argued, “As an ER doctor seeing a patient for a first time, my primary goal is not to guess your ultimate diagnosis. My primary goal is to determine if you have a condition that could kill you.”
This post and headline have been updated to reflect the fact that the diagnoses in the study came from internal medicine attending physicians, and to include commentary from Kristen Panthagani.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
If Apple were ever going to make an Android phone, then they’d probably design something like the Xiaomi 17. I wouldn’t describe the build as flashy, but it’s super elegant and reminiscent of past Xiaomi flagships. I talked about this in my X300 Pro review, as creating a brand identity to compete against Samsung and Apple is super important, and Xiaomi has listened. While I was using it daily, many of my friends and family asked me what Xiaomi phone I was using—note the wording: “Xiaomi phone,” meaning they knew it was a particular brand, and that’s important. The Chinese smartphone maker said they thought of every curve, and I’ll just say it straight: the 17 is the best-feeling compact phone I’ve held this year.
The corners are crafted to perfection, the width is spot on, and even the way the aluminum frame blends into the glass without an abrupt edge makes carrying the phone a very enjoyable experience. Beyond that, the back glass is frosted to prevent the phone from slipping off glass surfaces, and the side frame doesn’t let go of its color inside a case.
Speaking of color, you get plenty of options, but my favorite is definitely the blue variant, as it has that breezy summer vibe. The buttons are tactile and positioned where your hand would naturally rest.
Moving to the camera module, Xiaomi has taken the iPhone route of individual stove-top camera cutouts. There are four of them (one houses the flash), and aside from the fact that dust is difficult to get out from between, I do quite like them. The ultrasonic fingerprint scanner is positioned at a comfy place where your thumb would naturally rest. I used it on the beach with wet hands, and it worked perfectly fine. Besides, the phone is IP69-rated for dust and water resistance, meaning it should technically withstand a swim. Did I dare take it inside the water on the beach? Absolutely not, because the IP rating is only for fresh water, and seawater can cause irreversible damage.
Display
I’ve said this before that all flagship displays are essentially the same, and that holds true for the Xiaomi 17, too. The phone features a 6.3-inch 1220 x 2656 OLED display, with an adaptive 120Hz refresh rate. This time, Xiaomi has trimmed the bezels even more for a more premium look, and I’m a fan. The panel is exceptionally color-accurate and vibrant for content consumption, as evidenced by my 3-hour run of The Pitt season 2 on the flight to Thailand. Even the HDR performance is exceptional.
Xiaomi claims a peak brightness number of 3,500 nits. Sadly, I don’t have a light meter to put the claim to the test, but from my experience using the panel in the 12 noon sun at Phi Phi Island, it’s plenty bright for outdoor use. The texts were legible, and I could use the phone for GPS navigation without squinting.
When it comes to durability, I usually don’t like to test that part myself and instead rely on user reports. However, I accidentally dropped the Xiaomi 17 on a concrete floor. The result was surprisingly good. I dropped it, without a case, from a tripod at chest height, meaning that, while the phone was in the air, all sorts of scary thoughts came to mind, including how much this repair was going to cost me. Thankfully, the phone escaped with only minor damage to the frame.
Performance & Software
Performance is what makes or breaks the smartphone experience, and it’s no surprise to anyone that the Xiaomi 17 delivers top-of-the-line performance. The Snapdragon 8 Elite Gen 5 is the best Android processor in the market, and it’s coupled with 12GB of LPDDR5X RAM and up to 512GB of UFS 4.1 internal storage. The results? The Xiaomi 17 is an absolute joy to use. It flies through the UI like nothing, and there’s ample headroom for literally any task. That being said, the phone runs on HyperOS 3, which, for the uninitiated, is a very altered version of Android that resembles more like iOS.
I don’t have a problem with the look, especially since HyperOS is one of the smoothest Android skins, with silky animations and a lot of customization. My issue is that, unlike other Chinese skins that allow you to tone down the iOS-ness, Xiaomi doesn’t.
For example, the notification shade is divided into two sections: the quick control and the panel. I don’t like that, but when I went digging in the settings to find a way to merge them, there wasn’t. Also, the back gesture is enabled in the keyboard, so when I tried deleting long text, it would often send me back instead.
There are a few silver linings I wish others would copy from HyperOS, one major one being the lockscreen customizations. There are so many options, and every one of them looks gorgeous. As this is 2026, there’s a host of AI features, such as object eraser, image upscaling, and inpainting. I tried them all, and they work exactly as you’d expect. The company also promises about six years of major software updates and security patches. This is better than vivo’s five years.
Benchmarks & Gaming
As this is a review, I also ran a series of benchmarks to test the Snapdragon 8 Elite Gen 5’s limits. The phone scored 3,415 in Geekbench’s single-core test and 10,008 in the multi-core test. These are insane numbers, especially when compared with the likes of the vivo X300 Pro and the Find X9, which score about 20%-30% lower in multi-core tests. The story remained similar on AnTuTu, where the Xiaomi 17 handsomely beat its Chinese rivals, scoring 3,423,349.
As expected, this performance translates extremely well in gaming. I’m a former PUBG (BGMI) eSports player, and my results were exceptional. The phone maintained 120 FPS gameplay even at high settings without a hint of stutter. I also like Xiaomi’s thermal management, which kept the phone from overheating during both gaming and photo capture in Thailand’s hot summer.
Battery Life & Charging
After all the chatter about the small form factor, you may expect the Xiaomi 17 to compromise on the battery life, just as other Apple and Samsung phones do. Well, you can’t be more wrong, as the Xiaomi 17 packs an even bigger battery, 6,330mAh to be precise, than the 17 Ultra. And the results are just fantastic. On the morning of my Thailand flight, I unplugged the phone at 5 am. I then continued using the phone for the rest of the day, including the three hours of The Pitt on the flight and map navigation when reaching Phuket airport. I ended the day with 20% remaining, and at 3 am the next morning, I had 20% remaining. For a more typical person, you’d be looking more at two days of usage without a hitch.
When it was finally time to charge, Xiaomi, unlike Samsung, bundles a 100W fast charger in the box that charges the phone from 20% to 80% in just 30 minutes. You also get 50W of reverse wireless charging, though that requires a specific charger.
Cameras
If a phone doesn’t fold in half or has dual screens, the only way to differentiate itself is through the cameras. They are the main reason why people lean towards a certain brand, and recently, both OPPO and vivo have been killing it. However, I think there’s room for a third king: the Xiaomi 17. Like others, it also houses a triple-sensor array, led by the 50MP LightFusion 950 sensor, a 50MP JN1 60mm telephoto, and another 50MP OV50M ultrawide lens. Colors are handled by Leica, and that’s the main strength of the Xiaomi 17. The photos it takes, with the different Leica filters, have a certain character you won’t find anywhere else. Every phone takes similar photos these days, and it’s these color profiles that matter the most.
Still, if you’re not a fan of poking around with the cameras, the default Leica Authentic profile produces colors that are very close to natural, with highlights and shadows handled extremely well. The details are crisp and plenty, the HDR performance is mostly spot on, and the contrast is slightly on the boosted side, which is what I like. Beyond the default camera profile, there are a myriad of filters, such as Negative, Positive, Sepia, Natural, Vibrant, and Blue. Each has a different style of capturing the colors and subject, and I really did find myself going through each and every one of them to decide which actually serves the scene the best. And the results speak for themselves. Every photo tells a different story, and that’s the Xiaomi 17’s biggest strength.
The telephoto lens is 2.5x, and I’d say the same about it, too. It serves as the main portrait camera, and the images deliver stellar detail, with excellent foreground separation and improved natural skin tones without the infamous beautification. Xiaomi doesn’t rely much on AI processing, so zooming past 5x-6x will result in blurry photos. Keep that in mind. The ultrawide hasn’t changed from the previous generation, so it still doesn’t have autofocus for macro photography. While it works great when the light is ample, I saw a significant drop in quality at night.
Speaking of the night, both the main and telephoto sensors benefit from Xiaomi’s mature image processing, which retains detail in shadows without making the image muddy or introducing noise. Videos, which can be shot at up to 8K, carry similar details in all lighting conditions, and I’m a fan. Sadly, it’s not all perfect. In Thailand’s heat, some of the videos I captured were choppy, even when I was in the hotel. This problem then carried over to India, where the first few seconds of every video would stutter. I’ve communicated this issue with the Xiaomi team, so a fix could be imminent. Overall, I love the Xiaomi 17’s cameras.
Verdict
Sure, the ₹89,999 price tag of the Xiaomi 17 might feel a bit much, considering it’s more than the vivo and OPPO competition. But the Xiaomi 17 brings a lot of things to the table. You get the best-in-class performance that’s miles ahead of the competition. A design that’s understated yet premium. Battery life that can easily last two full days, and cameras that, instead of being same same but different, induce a character to each and every photo that makes them more memorable. Of course, it’s not perfect. I’d like the camera bugs fixed and ultrawide performance improved, but overall, the Xiaomi 17 gets my recommendation.
If Apple were ever going to make an Android phone, then they’d probably design something like the Xiaomi 17. I wouldn’t describe the build as flashy, but it’s super elegant and reminiscent of past Xiaomi flagships. I talked about this in my X300 Pro review, as creating a brand identity to compete against Samsung and Apple is super important, and Xiaomi has listened. While I was using it daily, many of my friends and family asked me what Xiaomi phone I was using—note the wording: “Xiaomi phone,” meaning they knew it was a particular brand, and that’s important. The Chinese smartphone maker said they thought of every curve, and I’ll just say it straight: the 17 is the best-feeling compact phone I’ve held this year.
The corners are crafted to perfection, the width is spot on, and even the way the aluminum frame blends into the glass without an abrupt edge makes carrying the phone a very enjoyable experience. Beyond that, the back glass is frosted to prevent the phone from slipping off glass surfaces, and the side frame doesn’t let go of its color inside a case.
Speaking of color, you get plenty of options, but my favorite is definitely the blue variant, as it has that breezy summer vibe. The buttons are tactile and positioned where your hand would naturally rest.
Moving to the camera module, Xiaomi has taken the iPhone route of individual stove-top camera cutouts. There are four of them (one houses the flash), and aside from the fact that dust is difficult to get out from between, I do quite like them. The ultrasonic fingerprint scanner is positioned at a comfy place where your thumb would naturally rest. I used it on the beach with wet hands, and it worked perfectly fine. Besides, the phone is IP69-rated for dust and water resistance, meaning it should technically withstand a swim. Did I dare take it inside the water on the beach? Absolutely not, because the IP rating is only for fresh water, and seawater can cause irreversible damage.
Display
I’ve said this before that all flagship displays are essentially the same, and that holds true for the Xiaomi 17, too. The phone features a 6.3-inch 1220 x 2656 OLED display, with an adaptive 120Hz refresh rate. This time, Xiaomi has trimmed the bezels even more for a more premium look, and I’m a fan. The panel is exceptionally color-accurate and vibrant for content consumption, as evidenced by my 3-hour run of The Pitt season 2 on the flight to Thailand. Even the HDR performance is exceptional.
Xiaomi claims a peak brightness number of 3,500 nits. Sadly, I don’t have a light meter to put the claim to the test, but from my experience using the panel in the 12 noon sun at Phi Phi Island, it’s plenty bright for outdoor use. The texts were legible, and I could use the phone for GPS navigation without squinting.
When it comes to durability, I usually don’t like to test that part myself and instead rely on user reports. However, I accidentally dropped the Xiaomi 17 on a concrete floor. The result was surprisingly good. I dropped it, without a case, from a tripod at chest height, meaning that, while the phone was in the air, all sorts of scary thoughts came to mind, including how much this repair was going to cost me. Thankfully, the phone escaped with only minor damage to the frame.
Performance & Software
Performance is what makes or breaks the smartphone experience, and it’s no surprise to anyone that the Xiaomi 17 delivers top-of-the-line performance. The Snapdragon 8 Elite Gen 5 is the best Android processor in the market, and it’s coupled with 12GB of LPDDR5X RAM and up to 512GB of UFS 4.1 internal storage. The results? The Xiaomi 17 is an absolute joy to use. It flies through the UI like nothing, and there’s ample headroom for literally any task. That being said, the phone runs on HyperOS 3, which, for the uninitiated, is a very altered version of Android that resembles more like iOS.
I don’t have a problem with the look, especially since HyperOS is one of the smoothest Android skins, with silky animations and a lot of customization. My issue is that, unlike other Chinese skins that allow you to tone down the iOS-ness, Xiaomi doesn’t.
For example, the notification shade is divided into two sections: the quick control and the panel. I don’t like that, but when I went digging in the settings to find a way to merge them, there wasn’t. Also, the back gesture is enabled in the keyboard, so when I tried deleting long text, it would often send me back instead.
There are a few silver linings I wish others would copy from HyperOS, one major one being the lockscreen customizations. There are so many options, and every one of them looks gorgeous. As this is 2026, there’s a host of AI features, such as object eraser, image upscaling, and inpainting. I tried them all, and they work exactly as you’d expect. The company also promises about six years of major software updates and security patches. This is better than vivo’s five years.
Benchmarks & Gaming
As this is a review, I also ran a series of benchmarks to test the Snapdragon 8 Elite Gen 5’s limits. The phone scored 3,415 in Geekbench’s single-core test and 10,008 in the multi-core test. These are insane numbers, especially when compared with the likes of the vivo X300 Pro and the Find X9, which score about 20%-30% lower in multi-core tests. The story remained similar on AnTuTu, where the Xiaomi 17 handsomely beat its Chinese rivals, scoring 3,423,349.
As expected, this performance translates extremely well in gaming. I’m a former PUBG (BGMI) eSports player, and my results were exceptional. The phone maintained 120 FPS gameplay even at high settings without a hint of stutter. I also like Xiaomi’s thermal management, which kept the phone from overheating during both gaming and photo capture in Thailand’s hot summer.
Battery Life & Charging
After all the chatter about the small form factor, you may expect the Xiaomi 17 to compromise on the battery life, just as other Apple and Samsung phones do. Well, you can’t be more wrong, as the Xiaomi 17 packs an even bigger battery, 6,330mAh to be precise, than the 17 Ultra. And the results are just fantastic. On the morning of my Thailand flight, I unplugged the phone at 5 am. I then continued using the phone for the rest of the day, including the three hours of The Pitt on the flight and map navigation when reaching Phuket airport. I ended the day with 20% remaining, and at 3 am the next morning, I had 20% remaining. For a more typical person, you’d be looking more at two days of usage without a hitch.
When it was finally time to charge, Xiaomi, unlike Samsung, bundles a 100W fast charger in the box that charges the phone from 20% to 80% in just 30 minutes. You also get 50W of reverse wireless charging, though that requires a specific charger.
Cameras
If a phone doesn’t fold in half or has dual screens, the only way to differentiate itself is through the cameras. They are the main reason why people lean towards a certain brand, and recently, both OPPO and vivo have been killing it. However, I think there’s room for a third king: the Xiaomi 17. Like others, it also houses a triple-sensor array, led by the 50MP LightFusion 950 sensor, a 50MP JN1 60mm telephoto, and another 50MP OV50M ultrawide lens. Colors are handled by Leica, and that’s the main strength of the Xiaomi 17. The photos it takes, with the different Leica filters, have a certain character you won’t find anywhere else. Every phone takes similar photos these days, and it’s these color profiles that matter the most.
Still, if you’re not a fan of poking around with the cameras, the default Leica Authentic profile produces colors that are very close to natural, with highlights and shadows handled extremely well. The details are crisp and plenty, the HDR performance is mostly spot on, and the contrast is slightly on the boosted side, which is what I like. Beyond the default camera profile, there are a myriad of filters, such as Negative, Positive, Sepia, Natural, Vibrant, and Blue. Each has a different style of capturing the colors and subject, and I really did find myself going through each and every one of them to decide which actually serves the scene the best. And the results speak for themselves. Every photo tells a different story, and that’s the Xiaomi 17’s biggest strength.
The telephoto lens is 2.5x, and I’d say the same about it, too. It serves as the main portrait camera, and the images deliver stellar detail, with excellent foreground separation and improved natural skin tones without the infamous beautification. Xiaomi doesn’t rely much on AI processing, so zooming past 5x-6x will result in blurry photos. Keep that in mind. The ultrawide hasn’t changed from the previous generation, so it still doesn’t have autofocus for macro photography. While it works great when the light is ample, I saw a significant drop in quality at night.
Speaking of the night, both the main and telephoto sensors benefit from Xiaomi’s mature image processing, which retains detail in shadows without making the image muddy or introducing noise. Videos, which can be shot at up to 8K, carry similar details in all lighting conditions, and I’m a fan. Sadly, it’s not all perfect. In Thailand’s heat, some of the videos I captured were choppy, even when I was in the hotel. This problem then carried over to India, where the first few seconds of every video would stutter. I’ve communicated this issue with the Xiaomi team, so a fix could be imminent. Overall, I love the Xiaomi 17’s cameras.
Verdict
Sure, the ₹89,999 price tag of the Xiaomi 17 might feel a bit much, considering it’s more than the vivo and OPPO competition. But the Xiaomi 17 brings a lot of things to the table. You get the best-in-class performance that’s miles ahead of the competition. A design that’s understated yet premium. Battery life that can easily last two full days, and cameras that, instead of being same same but different, induce a character to each and every photo that makes them more memorable. Of course, it’s not perfect. I’d like the camera bugs fixed and ultrawide performance improved, but overall, the Xiaomi 17 gets my recommendation.
#Xiaomi #Review #Thailand #Real #Camera #TestXiaomi">Xiaomi 17 Review: I Took It to Thailand for a Real Camera Test
Xiaomi phones are a little tough to judge. After all, these guys do everything, from making phones to laptops and sometimes even record-breaking electric SUVs. The Xiaomi 17 is a bit like the quiet kid that never gets noticed, simply because its bigger brother, the 17 Ultra, is on a streak of collecting all the best smartphone camera awards. But here’s the thing: most people won’t ever splurge that much money on a non-Samsung or Apple Ultra flagship. The main sales driver will always be the base model, and that’s the question I had in mind. Can the Xiaomi 17 go head-to-head with the OPPO Find X9 and the vivo X300, especially since it’s more expensive than both? You can thank AI for that.
To answer this very question, I got the Xiaomi 17 for review and took it with me on a work trip to Phuket, Thailand. Here, I used the phone to capture about 500 photos in the summer heat, with temperatures soaring to 40 degrees, and constant GPS navigation to put the Snapdragon 8 Elite Gen 5 SoC through its paces. Spoiler alert, I really do love this phone, but there are a few quirks, too. Here’s why.
Xiaomi 17 Review
Hisan Kidwai
Summary
The Xiaomi 17 brings a lot of things to the table. You get the best-in-class performance that’s miles ahead of the competition. A design that’s understated yet premium. Battery life that can easily last two full days, and cameras that, instead of being same same but different, induce a character to each and every photo that makes them more memorable. Of course, it’s not perfect. I’d like the camera bugs fixed and ultrawide performance improved, but overall, the Xiaomi 17 gets my recommendation.
Design & Hardware
If Apple were ever going to make an Android phone, then they’d probably design something like the Xiaomi 17. I wouldn’t describe the build as flashy, but it’s super elegant and reminiscent of past Xiaomi flagships. I talked about this in my X300 Pro review, as creating a brand identity to compete against Samsung and Apple is super important, and Xiaomi has listened. While I was using it daily, many of my friends and family asked me what Xiaomi phone I was using—note the wording: “Xiaomi phone,” meaning they knew it was a particular brand, and that’s important. The Chinese smartphone maker said they thought of every curve, and I’ll just say it straight: the 17 is the best-feeling compact phone I’ve held this year.
The corners are crafted to perfection, the width is spot on, and even the way the aluminum frame blends into the glass without an abrupt edge makes carrying the phone a very enjoyable experience. Beyond that, the back glass is frosted to prevent the phone from slipping off glass surfaces, and the side frame doesn’t let go of its color inside a case.
Speaking of color, you get plenty of options, but my favorite is definitely the blue variant, as it has that breezy summer vibe. The buttons are tactile and positioned where your hand would naturally rest.
Moving to the camera module, Xiaomi has taken the iPhone route of individual stove-top camera cutouts. There are four of them (one houses the flash), and aside from the fact that dust is difficult to get out from between, I do quite like them. The ultrasonic fingerprint scanner is positioned at a comfy place where your thumb would naturally rest. I used it on the beach with wet hands, and it worked perfectly fine. Besides, the phone is IP69-rated for dust and water resistance, meaning it should technically withstand a swim. Did I dare take it inside the water on the beach? Absolutely not, because the IP rating is only for fresh water, and seawater can cause irreversible damage.
Display
I’ve said this before that all flagship displays are essentially the same, and that holds true for the Xiaomi 17, too. The phone features a 6.3-inch 1220 x 2656 OLED display, with an adaptive 120Hz refresh rate. This time, Xiaomi has trimmed the bezels even more for a more premium look, and I’m a fan. The panel is exceptionally color-accurate and vibrant for content consumption, as evidenced by my 3-hour run of The Pitt season 2 on the flight to Thailand. Even the HDR performance is exceptional.
Xiaomi claims a peak brightness number of 3,500 nits. Sadly, I don’t have a light meter to put the claim to the test, but from my experience using the panel in the 12 noon sun at Phi Phi Island, it’s plenty bright for outdoor use. The texts were legible, and I could use the phone for GPS navigation without squinting.
When it comes to durability, I usually don’t like to test that part myself and instead rely on user reports. However, I accidentally dropped the Xiaomi 17 on a concrete floor. The result was surprisingly good. I dropped it, without a case, from a tripod at chest height, meaning that, while the phone was in the air, all sorts of scary thoughts came to mind, including how much this repair was going to cost me. Thankfully, the phone escaped with only minor damage to the frame.
Performance & Software
Performance is what makes or breaks the smartphone experience, and it’s no surprise to anyone that the Xiaomi 17 delivers top-of-the-line performance. The Snapdragon 8 Elite Gen 5 is the best Android processor in the market, and it’s coupled with 12GB of LPDDR5X RAM and up to 512GB of UFS 4.1 internal storage. The results? The Xiaomi 17 is an absolute joy to use. It flies through the UI like nothing, and there’s ample headroom for literally any task. That being said, the phone runs on HyperOS 3, which, for the uninitiated, is a very altered version of Android that resembles more like iOS.
I don’t have a problem with the look, especially since HyperOS is one of the smoothest Android skins, with silky animations and a lot of customization. My issue is that, unlike other Chinese skins that allow you to tone down the iOS-ness, Xiaomi doesn’t.
For example, the notification shade is divided into two sections: the quick control and the panel. I don’t like that, but when I went digging in the settings to find a way to merge them, there wasn’t. Also, the back gesture is enabled in the keyboard, so when I tried deleting long text, it would often send me back instead.
There are a few silver linings I wish others would copy from HyperOS, one major one being the lockscreen customizations. There are so many options, and every one of them looks gorgeous. As this is 2026, there’s a host of AI features, such as object eraser, image upscaling, and inpainting. I tried them all, and they work exactly as you’d expect. The company also promises about six years of major software updates and security patches. This is better than vivo’s five years.
Benchmarks & Gaming
As this is a review, I also ran a series of benchmarks to test the Snapdragon 8 Elite Gen 5’s limits. The phone scored 3,415 in Geekbench’s single-core test and 10,008 in the multi-core test. These are insane numbers, especially when compared with the likes of the vivo X300 Pro and the Find X9, which score about 20%-30% lower in multi-core tests. The story remained similar on AnTuTu, where the Xiaomi 17 handsomely beat its Chinese rivals, scoring 3,423,349.
As expected, this performance translates extremely well in gaming. I’m a former PUBG (BGMI) eSports player, and my results were exceptional. The phone maintained 120 FPS gameplay even at high settings without a hint of stutter. I also like Xiaomi’s thermal management, which kept the phone from overheating during both gaming and photo capture in Thailand’s hot summer.
Battery Life & Charging
After all the chatter about the small form factor, you may expect the Xiaomi 17 to compromise on the battery life, just as other Apple and Samsung phones do. Well, you can’t be more wrong, as the Xiaomi 17 packs an even bigger battery, 6,330mAh to be precise, than the 17 Ultra. And the results are just fantastic. On the morning of my Thailand flight, I unplugged the phone at 5 am. I then continued using the phone for the rest of the day, including the three hours of The Pitt on the flight and map navigation when reaching Phuket airport. I ended the day with 20% remaining, and at 3 am the next morning, I had 20% remaining. For a more typical person, you’d be looking more at two days of usage without a hitch.
When it was finally time to charge, Xiaomi, unlike Samsung, bundles a 100W fast charger in the box that charges the phone from 20% to 80% in just 30 minutes. You also get 50W of reverse wireless charging, though that requires a specific charger.
Cameras
If a phone doesn’t fold in half or has dual screens, the only way to differentiate itself is through the cameras. They are the main reason why people lean towards a certain brand, and recently, both OPPO and vivo have been killing it. However, I think there’s room for a third king: the Xiaomi 17. Like others, it also houses a triple-sensor array, led by the 50MP LightFusion 950 sensor, a 50MP JN1 60mm telephoto, and another 50MP OV50M ultrawide lens. Colors are handled by Leica, and that’s the main strength of the Xiaomi 17. The photos it takes, with the different Leica filters, have a certain character you won’t find anywhere else. Every phone takes similar photos these days, and it’s these color profiles that matter the most.
Still, if you’re not a fan of poking around with the cameras, the default Leica Authentic profile produces colors that are very close to natural, with highlights and shadows handled extremely well. The details are crisp and plenty, the HDR performance is mostly spot on, and the contrast is slightly on the boosted side, which is what I like. Beyond the default camera profile, there are a myriad of filters, such as Negative, Positive, Sepia, Natural, Vibrant, and Blue. Each has a different style of capturing the colors and subject, and I really did find myself going through each and every one of them to decide which actually serves the scene the best. And the results speak for themselves. Every photo tells a different story, and that’s the Xiaomi 17’s biggest strength.
The telephoto lens is 2.5x, and I’d say the same about it, too. It serves as the main portrait camera, and the images deliver stellar detail, with excellent foreground separation and improved natural skin tones without the infamous beautification. Xiaomi doesn’t rely much on AI processing, so zooming past 5x-6x will result in blurry photos. Keep that in mind. The ultrawide hasn’t changed from the previous generation, so it still doesn’t have autofocus for macro photography. While it works great when the light is ample, I saw a significant drop in quality at night.
Speaking of the night, both the main and telephoto sensors benefit from Xiaomi’s mature image processing, which retains detail in shadows without making the image muddy or introducing noise. Videos, which can be shot at up to 8K, carry similar details in all lighting conditions, and I’m a fan. Sadly, it’s not all perfect. In Thailand’s heat, some of the videos I captured were choppy, even when I was in the hotel. This problem then carried over to India, where the first few seconds of every video would stutter. I’ve communicated this issue with the Xiaomi team, so a fix could be imminent. Overall, I love the Xiaomi 17’s cameras.
Verdict
Sure, the ₹89,999 price tag of the Xiaomi 17 might feel a bit much, considering it’s more than the vivo and OPPO competition. But the Xiaomi 17 brings a lot of things to the table. You get the best-in-class performance that’s miles ahead of the competition. A design that’s understated yet premium. Battery life that can easily last two full days, and cameras that, instead of being same same but different, induce a character to each and every photo that makes them more memorable. Of course, it’s not perfect. I’d like the camera bugs fixed and ultrawide performance improved, but overall, the Xiaomi 17 gets my recommendation.
![In Harvard study, AI offered more accurate emergency room diagnoses than two human doctors | TechCrunch
A new study examines how large language models perform in a variety of medical contexts, including real emergency room cases — where at least one model seemed to be more accurate than human doctors.
The study was published this week in Science and comes from a research team led by physicians and computer scientists at Harvard Medical School and Beth Israel Deaconess Medical Center. The researchers said they conducted a variety of experiments to measure how OpenAI’s models compared to human physicians.
In one experiment, researchers focused on 76 patients who came into the Beth Israel emergency room, comparing the diagnoses offered by two internal medicine attending physicians to those generated by OpenAI’s o1 and 4o models. These diagnoses were assessed by two other attending physicians, who did not know which ones came from humans and which came from AI.
“At each diagnostic touchpoint, o1 either performed nominally better than or on par with the two attending physicians and 4o,” the study said, adding that the differences “were especially pronounced at the first diagnostic touchpoint (initial ER triage), where there is the least information available about the patient and the most urgency to make the correct decision.”
In Harvard Medical School’s press release about the study, the researchers emphasized that they did not “pre-process the data at all” — the AI models were presented with the same information that was available in the electronic medical records at the time of each diagnosis.
With that information, the o1 model managed to offer “the exact or very close diagnosis” in 67% of triage cases, compared to one physician who had the exact or close diagnosis 55% of the time, and to the other who hit the mark 50% of the time.
“We tested the AI model against virtually every benchmark, and it eclipsed both prior models and our physician baselines,” said Arjun Manrai, who heads an AI lab at Harvard Medical School and is one of the study’s lead authors, in the press release.
Techcrunch event
San Francisco, CA
|
October 13-15, 2026
To be clear, the study didn’t claim that AI is ready to make real life-or-death decisions in the emergency room. Instead, it said the findings show an “urgent need for prospective trials to evaluate these technologies in real-world patient care settings.”
The researchers also noted that they only studied how models performed when provided with text-based information, and that “existing studies suggest that current foundation models are more limited in reasoning over nontext inputs.”
Adam Rodman, a Beth Israel doctor who’s also one of the study’s lead authors, warned the Guardian that there’s “no formal framework right now for accountability” around AI diagnoses, and that patients still “want humans to guide them through life or death decisions [and] to guide them through challenging treatment decisions.”
In a post about the study, Kristen Panthagani, an emergency physician, said this is an “an interesting AI study that has led to some very overhyped headlines,” especially since it was comparing AI diagnoses to those from internal medicine physicians, not ER physicians.
“If we’re going to compare AI tools to physicians’ clinical ability, we should start by comparing to physicians who actually practice that specialty,” Panthagani said. “I would not be surprised if a LLM could beat a dermatologist at an neurosurgery board exam, [but] that’s not a particularly helpful thing to know.”
She also argued, “As an ER doctor seeing a patient for a first time, my primary goal is not to guess your ultimate diagnosis. My primary goal is to determine if you have a condition that could kill you.”
This post and headline have been updated to reflect the fact that the diagnoses in the study came from internal medicine attending physicians, and to include commentary from Kristen Panthagani.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.#Harvard #study #offered #accurate #emergency #room #diagnoses #human #doctors #TechCrunchbeth israel,harvard medical school,OpenAI](https://i2.wp.com/techcrunch.com/wp-content/uploads/2024/01/GettyImages-104509077.jpg?resize=1200%2C800&ssl=1 "In Harvard study, AI offered more accurate emergency room diagnoses than two human doctors | TechCrunch
A new study examines how large language models perform in a variety of medical contexts, including real emergency room cases — where at least one model seemed to be more accurate than human doctors.
The study was published this week in Science and comes from a research team led by physicians and computer scientists at Harvard Medical School and Beth Israel Deaconess Medical Center. The researchers said they conducted a variety of experiments to measure how OpenAI’s models compared to human physicians.
In one experiment, researchers focused on 76 patients who came into the Beth Israel emergency room, comparing the diagnoses offered by two internal medicine attending physicians to those generated by OpenAI’s o1 and 4o models. These diagnoses were assessed by two other attending physicians, who did not know which ones came from humans and which came from AI.
“At each diagnostic touchpoint, o1 either performed nominally better than or on par with the two attending physicians and 4o,” the study said, adding that the differences “were especially pronounced at the first diagnostic touchpoint (initial ER triage), where there is the least information available about the patient and the most urgency to make the correct decision.”
In Harvard Medical School’s press release about the study, the researchers emphasized that they did not “pre-process the data at all” — the AI models were presented with the same information that was available in the electronic medical records at the time of each diagnosis.
With that information, the o1 model managed to offer “the exact or very close diagnosis” in 67% of triage cases, compared to one physician who had the exact or close diagnosis 55% of the time, and to the other who hit the mark 50% of the time.
“We tested the AI model against virtually every benchmark, and it eclipsed both prior models and our physician baselines,” said Arjun Manrai, who heads an AI lab at Harvard Medical School and is one of the study’s lead authors, in the press release.
Techcrunch event
San Francisco, CA
|
October 13-15, 2026
To be clear, the study didn’t claim that AI is ready to make real life-or-death decisions in the emergency room. Instead, it said the findings show an “urgent need for prospective trials to evaluate these technologies in real-world patient care settings.”
The researchers also noted that they only studied how models performed when provided with text-based information, and that “existing studies suggest that current foundation models are more limited in reasoning over nontext inputs.”
Adam Rodman, a Beth Israel doctor who’s also one of the study’s lead authors, warned the Guardian that there’s “no formal framework right now for accountability” around AI diagnoses, and that patients still “want humans to guide them through life or death decisions [and] to guide them through challenging treatment decisions.”
In a post about the study, Kristen Panthagani, an emergency physician, said this is an “an interesting AI study that has led to some very overhyped headlines,” especially since it was comparing AI diagnoses to those from internal medicine physicians, not ER physicians.
“If we’re going to compare AI tools to physicians’ clinical ability, we should start by comparing to physicians who actually practice that specialty,” Panthagani said. “I would not be surprised if a LLM could beat a dermatologist at an neurosurgery board exam, [but] that’s not a particularly helpful thing to know.”
She also argued, “As an ER doctor seeing a patient for a first time, my primary goal is not to guess your ultimate diagnosis. My primary goal is to determine if you have a condition that could kill you.”
This post and headline have been updated to reflect the fact that the diagnoses in the study came from internal medicine attending physicians, and to include commentary from Kristen Panthagani.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.#Harvard #study #offered #accurate #emergency #room #diagnoses #human #doctors #TechCrunchbeth israel,harvard medical school,OpenAI")
Post Comment