Notion’s High-Contrast Mode Makes the Interface Clearer, Not Every Page
The new display setting improves text, icons and borders, but workspace authors still own the accessibility of their content.
Notion has added a high-contrast display mode for people who find its muted interface difficult to read. The setting increases the visual separation of text, icons and borders. It is available now in the desktop app and on the web.
That sounds like a small appearance option, and in operation it is. Open Settings, choose Preferences, then Appearance and select high contrast. It works alongside the existing light and dark themes rather than replacing them.
The useful part is precisely that simplicity. Notion pages can contain dense databases, pale dividers, secondary labels and controls that recede into the background. Increasing contrast can make those elements easier to locate without asking every workspace author to redesign a page.
There are two practical limits to know before treating it as a complete accessibility solution. The option is not available on mobile, and independent release tracking indicates that the preference is stored per device. Turning it on at work does not automatically carry the choice to another computer.
The setting also belongs to the viewer, not the workspace administrator or page owner. That is valuable because contrast needs vary, but it means a team cannot assume that enabling it once has improved a shared document for everyone. Each reader must choose the presentation on each supported device.
High contrast modifies Notion’s application interface. It makes the product’s own words, icons, outlines and navigation more distinct. That can help people with reduced contrast sensitivity, but it can also be useful on a dim display, an aging monitor or a laptop used in bright light.
The feature answers a long-running criticism of Notion’s visual style. Users have complained for years that gray text, faint borders and desaturated colors can be difficult to distinguish, especially in dark mode. A 2024 accessibility assessment by design students at Pratt Institute also identified limited contrast control as one of several barriers in the product.
Notion now acknowledges the tradeoff directly. In announcing the mode, the company said its muted design can make content hard to read. Giving the reader a preference is better than assuming one level of subtlety suits every set of eyes.
It is still important not to call the feature proof of compliance with the Web Content Accessibility Guidelines, or WCAG. Notion has not published measured contrast ratios for the new theme. The W3C standard generally calls for a ratio of at least 4.5 to 1 between normal text and its background, with 3 to 1 permitted for large text. Visual information needed to identify controls and meaningful graphics should generally reach 3 to 1 against adjacent colors.
Those are testable thresholds, not synonyms for a setting named “high contrast.” A product can offer a stronger theme while particular elements still miss a target.
The mode cannot repair everything inside a workspace. Authors choose colored text, callout backgrounds, status labels, images, charts and embedded material. If a project board communicates urgency only through red, yellow and green tags, stronger interface contrast does not give those colors a second meaning. A person who cannot reliably distinguish them may still lose the message.
The same applies to screenshots containing tiny text and to diagrams whose lines differ only by color. High contrast around the image does not alter the image itself. Page creators should add clear labels, useful alternative text and written summaries where visual material carries important information.
Keyboard navigation, screen-reader behavior and focus order are separate concerns as well. The Pratt assessment found interaction problems that changing colors alone would not address. This release should be judged as one targeted improvement, not a replacement for broader accessibility work.
For teams, the sensible approach is to make people aware of the new option while continuing to design shared pages for ordinary viewing conditions. Do not require colleagues to discover a personal setting before they can read a status, find a control or interpret a dashboard. WebAIM notes that some people with low vision use operating-system or browser contrast overrides, but content authors still need to provide sufficient contrast in the default experience.
As a product decision, Notion’s new mode is refreshingly narrow. It does not promise to redesign work with AI or automate another routine. It gives readers more control over a visual system that the company admits can be too quiet. The next meaningful steps would be mobile support, preference syncing and published testing that shows which elements meet recognized contrast targets.
Until then, the feature is worth enabling for anyone who squints at Notion’s gray details. It makes the interface clearer. It does not remove the responsibility to make the information inside it understandable.
LinkedIn’s “AI Slop” Button Is a Feed Signal, Not an AI Detector
The new report option targets low-quality and inauthentic posts while leaving the definition deliberately open.
LinkedIn is adding a new reason for reporting a post or comment: “Seems like AI slop.” The wording is blunt, but the mechanism is more modest than the label suggests. It does not establish that a machine wrote something. It gives LinkedIn another signal about content that members find repetitive, inauthentic or unhelpful.
The company announced the change on July 30. LinkedIn chief product officer Hari Srinivasan said member reports would help tune its systems because “slop” is difficult to define and the definition keeps changing. LinkedIn is also testing private notices in creator analytics when readers perceive a post as inauthentic or too heavily shaped by AI.
Alongside those reports, LinkedIn says it has improved classifiers that identify low-quality content. Material that receives these signals may appear less often in recommendations, especially to people who do not follow the author. Reporting a post also hides it from the person who filed the report.
That makes the new option part of feed ranking and moderation. It is not a public warning label, a plagiarism finding or a dependable AI detector.
LinkedIn is not banning AI-assisted writing. Srinivasan explicitly separated AI use from slop, noting that people can use tools to refine their thoughts without surrendering their own point of view.
The more revealing product change is elsewhere in the composer. LinkedIn is removing its “enhance your post” feature, which could rewrite a draft, and replacing it with proofreading designed to preserve the author’s voice. The platform is retreating from a tool that helped make professional posts sound more alike while asking members to identify the sameness that now irritates them.
This is a quality problem before it is an authorship problem. A human can produce generic engagement bait. An AI tool can help clarify a useful argument. The report option deliberately asks how a post feels to readers, not how it was produced.
That choice has practical advantages. Automated AI-text detectors remain unreliable, particularly after a person edits the output. A behavioral signal can instead capture repetition, manufactured anecdotes and formulaic comments that technically evade detection.
The same flexibility is also the feature’s weakness. LinkedIn has not disclosed how many reports affect distribution, how its classifiers weigh them, whether a human reviews disputed cases or what appeal process creators will have. The company has also not specified the full rollout schedule.
Crowdsourced labels can be gamed. Coordinated users could target an unpopular author or mistake polished, translated or non-native English for automated prose. Writers who favor conventional business language may look formulaic even when every sentence is their own. A report should therefore be treated as noisy feedback, not evidence.
The platform says its defenses block hundreds of thousands of automated comment attempts each day and billions of other automation attempts in recent months. Those company figures describe a genuine scale problem, but they combine obvious bot activity with the harder question of judging ordinary posts. Stopping a comment bot is more straightforward than deciding whether a thoughtful post has been polished too much.
For professionals and company pages, the safest response is not to hunt for supposed AI tells. It is to make posts harder to confuse with templates. Include first-hand observations, specific evidence and a conclusion that follows from the writer’s experience. AI can still help with structure or proofreading, but batch production and interchangeable commentary now carry a clearer distribution risk.
LinkedIn’s move matters because it turns an argument about AI authorship into an ongoing product decision about attention. Members will help define what the feed should suppress, while classifiers translate those judgments into reach. The button may reduce obvious filler. Whether it improves the feed without punishing legitimate voices will depend on safeguards LinkedIn has not yet explained.
How eROSITA Turns Repeated Sky Scans Into Two Million X-Ray Sources
The telescope’s second public catalogue shows how photons become positions, probabilities and usable astronomical objects.
The eROSITA consortium has released its second public catalogue of the X-ray sky. Built from three complete surveys, Data Release 2 contains nearly two million sources in its main catalogue, roughly twice the number in the first release.
Most are point-like detections associated with stars or actively feeding supermassive black holes. About 64,000 appear extended, a category that includes galaxy clusters, nearby galaxies and supernova remnants. A separate catalogue adds nearly 15,000 sources detected at higher X-ray energies.
Those numbers are the end of a long conversion process. eROSITA does not take a conventional photograph and label the objects it sees. It records individual X-ray events, combines repeated passes, models the instrument and the background, then assigns probabilities to possible sources.
eROSITA is an array of seven aligned X-ray telescopes aboard the Spectrum-Roentgen-Gamma spacecraft. Each telescope module uses mirrors designed to guide incoming X-rays onto a CCD detector. Together they provide a field of view about one degree wide.
During survey operations, the spacecraft rotated once every four hours around an axis pointed near the Sun. That axis shifted by about one degree per day as Earth moved around the Sun. The geometry drew great circles across the sky and produced full coverage roughly every six months.
A typical sky position passed through the field of view several times during each survey, often in exposures lasting up to 40 seconds. Data Release 2 stacks the first three surveys, covering 556 days of operations. Repeated passes add more photons from persistent sources, which improves the chance of finding objects too faint to clear the detection threshold in a single survey.
The public release covers the western Galactic hemisphere. That is not an instrumental limit. Survey data rights were divided between the German and Russian eROSITA consortia, and this release comes from the German side.
Raw detections first need cleaning and calibration. The pipeline reconstructs each event, estimates its energy and determines where the telescope was pointing. It rejects damaged frames, invalid detector patterns and periods affected by background flares.
The sky is then divided into 4,700 overlapping tiles. Overlap reduces the chance that a source near the edge of one tile is measured poorly or missed. For each tile, software builds images, exposure maps and models of the background expected without an astronomical source.
The detection stage compares the recorded pattern with the telescope’s point spread function, which describes how a point source is blurred by the optics. It also tests extended models for objects that cover a larger patch of sky. The resulting fit estimates position, brightness and whether the source is point-like or extended.
Every candidate receives a detection likelihood. This is related to the probability that random background fluctuations could have produced the signal. The main public catalogue requires a likelihood of at least six. Raising that threshold creates a cleaner sample but discards more real, faint objects. Lowering it improves completeness while admitting more false detections.
That trade-off is why the catalogue includes warning flags. Bright diffuse regions, crowded areas and fragments of large extended objects can confuse the detection algorithm. Very bright optical stars can also generate false X-ray-like events in the CCDs, a problem called optical loading.
An X-ray position alone usually does not reveal what produced it. The team therefore compared point sources with optical and infrared catalogues from the DESI Legacy Imaging Surveys, Gaia and CatWISE.
The matching is probabilistic. The software considers the separation between positions, their uncertainties, the local density of possible matches and whether an optical or infrared object has properties typical of an X-ray emitter. It also repeats the calculation after shifting X-ray positions to estimate how often convincing matches occur by chance.
Within the DESI imaging footprint, the team identified counterparts for about 1.4 million sources and estimates that roughly 88 percent are outside the Milky Way. Many are active galactic nuclei, the bright regions powered by matter falling toward supermassive black holes.
This cross-matching turns a detection list into something researchers can filter by likely object type, distance and reliability. It also exposes uncertainty rather than erasing it. A faint source can be genuine but variable, a background fluctuation, or associated with the wrong visible object.
Data Release 2 is catalogue-focused. Unlike the first release, it does not publish the full event files, spectra and sky maps for every detection. It provides validated source tables, counterpart catalogues, documentation and an updated service for calculating upper flux limits where no source was detected.
That last tool matters because absence is also a measurement. Researchers comparing another survey with eROSITA need to know whether an object was truly quiet or simply too faint for the local exposure and background.
eROSITA has been in safe mode since February 2022, so the release does not represent new observing. Its advance comes from combining existing scans with improved calibration and processing. A third German data release is planned for the second half of 2028.
The practical lesson is that a sky catalogue is not a finished picture of the Universe. It is a structured set of measurements, models and confidence levels. The value of nearly two million entries depends on preserving that chain from photon to probability.