Tech

Why Weibo’s tiny VibeThinker-3B has the AI world arguing over benchmarks again

Published

1 month ago

17 June 2026

On Sunday, a team of nine researchers at Sina Weibo — the Chinese social media giant better known for its microblogging platform than for cutting-edge artificial intelligence — quietly posted a 14-page technical report to arXiv that sent shockwaves through the AI research community. Their claim: a language model with just 3 billion parameters can match or exceed the reasoning performance of flagship systems from Google DeepMind, OpenAI, Anthropic, and DeepSeek that are hundreds of times larger.

The model, called VibeThinker-3B, scored 94.3 on AIME 2026 — the American Invitational Mathematics Examination, one of the most demanding standardized math competitions in the world. That figure places it alongside DeepSeek V3.2, a model with 671 billion parameters, and ahead of Gemini 3 Pro, Google’s high-performance flagship reasoning system, which scored 91.7. With a test-time scaling technique the team calls Claim-Level Reliability Assessment, the score climbs to 97.1, edging past virtually every system in the public record.

Within hours of publication, the paper had drawn 62 upvotes on Hugging Face’s daily papers feed, the model repository had accumulated 130 likes, and the GitHub repository had reached 685 stars. But the reaction on social media was not uniformly celebratory. It was, in many cases, deeply skeptical.

“WHAT THE HELL is happening in AI?” wrote the user @orcus108 on X, in a post that accumulated over 161,000 views. “A 3B parameter model just put up coding benchmark scores in the same league as Claude Opus 4.5… I genuinely don’t know if this is a breakthrough or if the benchmarks are broken.”

That tension — between genuine scientific advancement and the growing suspicion that AI benchmarks have become gameable to the point of meaninglessness — sits at the heart of the VibeThinker-3B story. And the answer matters enormously, not just for academic bragging rights, but for the multibillion-dollar question of whether the AI industry’s relentless push toward ever-larger models is the only path to intelligence.

Benchmark scores that defy the scaling laws of modern AI

The results reported in the technical report are, by any conventional standard, extraordinary.

On the mathematics side, VibeThinker-3B achieved 91.4 on AIME 2025, 94.3 on AIME 2026, 89.3 on HMMT 2025 (the Harvard-MIT Mathematics Tournament), 93.8 on BruMO 2025 (the Brown University Math Olympiad), and 76.4 on IMO-AnswerBench, a benchmark comprising 400 problems at the level of the International Mathematical Olympiad. In coding, it posted an 80.2 Pass@1 on LiveCodeBench v6, a benchmark designed to test executable code generation, and achieved a 96.1 percent acceptance rate on unseen LeetCode weekly and biweekly contests from late April through late May 2026. On instruction following, it scored 93.4 on IFEval.

To put the parameter disparity in perspective: DeepSeek V3.2 has 671 billion parameters — roughly 224 times the size of VibeThinker-3B. GLM-5, from Zhipu AI, has 744 billion parameters. Kimi K2.5, from Moonshot AI, exceeds 1 trillion. VibeThinker-3B’s 3 billion parameters could run on a consumer laptop.

The researchers frame this result not as an anomaly but as evidence for a broader theoretical claim. They introduce what they call the “Parametric Compression-Coverage Hypothesis,” which argues that different types of AI capability have fundamentally different relationships to model size. Verifiable reasoning — the kind tested by math competitions and coding challenges, where answers can be definitively checked — is what the paper calls a “parameter-dense” capability: one that can be compressed into a compact core. Open-domain knowledge, by contrast, is “parameter-expansive,” requiring broad coverage across facts, concepts, and edge cases that inherently demands more parameters.

The paper acknowledges this distinction directly. On GPQA-Diamond, a graduate-level science knowledge benchmark, VibeThinker-3B scored just 70.2 — well behind the 91.9 achieved by Gemini 3 Pro and the 87.0 scored by Claude Opus 4.5. The authors write that this gap “is consistent with our claim rather than a contradiction to it: the main finding is not that a 3B model has fully replaced leading general-purpose models, but that a small model can reach first-tier performance on many verifiable reasoning tasks.”

Inside the four-stage training pipeline that powers a tiny reasoning engine

VibeThinker-3B is not built from scratch. It is post-trained on top of Qwen2.5-Coder-3B, a compact foundation model from Alibaba’s Qwen team, through what the Weibo AI researchers call the “Spectrum-to-Signal Principle” — a multi-stage pipeline first introduced in the team’s earlier VibeThinker-1.5B work in November 2025.

The training unfolds in four major phases. The first is a two-stage supervised fine-tuning process that uses curriculum learning: the model first trains on a broad mixture of math, code, STEM reasoning, general dialogue, and instruction-following data, then shifts to a curated subset of harder, longer-horizon reasoning problems. In the second stage, samples with reasoning traces shorter than 5,000 tokens are discarded, and problems that VibeThinker-1.5B can solve more than 75 percent of the time are filtered out, forcing the model to focus on genuinely difficult challenges.

The second phase applies reinforcement learning across multiple domains — mathematics, code, and STEM — using the team’s MaxEnt-Guided Policy Optimization algorithm, or MGPO, which prioritizes training on problems at the model’s current capability boundary rather than problems it already solves easily or finds impossible. Notably, the team found that a strategy that worked well at the 1.5B scale — progressively expanding the context window during RL training — actually hurt performance at 3B. They hypothesize that the stronger starting checkpoint meant that truncating reasoning traces during warm-up was no longer removing noise but disrupting valid reasoning patterns. The solution was to train with a single 64,000-token context window throughout.

Within the math RL phase, the team also introduces what it calls “Long2Short Math RL,” a secondary optimization stage that redistributes rewards to favor shorter correct solutions over longer ones, reducing verbosity without sacrificing accuracy. The technique uses a zero-sum reward redistribution that avoids biasing the overall reward signal while nudging the model toward more efficient reasoning.

The third phase extracts high-quality reasoning trajectories from the RL-trained checkpoints and distills them back into a unified model through supervised fine-tuning. The team uses a “learning-potential score” — essentially the student model’s perplexity on each teacher trajectory — to prioritize traces that are correct but that the student has not yet internalized. The final phase, called Instruct RL, applies reinforcement learning on instruction-following tasks using a combination of rule-based validators for format constraints and rubric-based reward models for open-ended quality assessment.

Francesco Bertolotti, an AI researcher who flagged the paper early on X, described the approach succinctly: “These results were achieved primarily through post-training refinements on Qwen2.5-Coder. The paper doesn’t provide many details, but it appears they distill from RL ckpts and then do a final RL-based instruct RL.” His post drew over 161,000 views.

Real-world testing reveals the gap between benchmark scores and practical AI performance

For every enthusiastic reaction, the paper drew an equally forceful objection. The AI research community in mid-2026 has grown deeply wary of benchmark-driven claims, and VibeThinker-3B arrived in an environment primed for suspicion.

“The benchmarks are literal pattern matching single file coding,” wrote @BigMoonKR on X. “It has no relation to actual coding work. I don’t know how people still don’t get this.”

“Benchmaxxing,” declared @oflu_bedirhan, using a term that has become shorthand in the AI community for models that appear optimized specifically for benchmark performance at the expense of real-world utility.

The most pointed criticism came from users who actually downloaded and tested the model. “Just tried the full precision,” wrote @politilols. “It doesn’t even know what a uv script (so the most popular Python dev tool) is. Haven’t seen that in a single LLM in at least a year now. Benchmaxxed.” When Bertolotti responded that the model seemed more focused on mathematical reasoning than practical coding, the user countered: “They include a livecodebench score. Zero chance that is reflective of the model.”

@Itsdotdev raised a structural criticism: “Look into the benchmarks themselves and it probably won’t be so shocking. Why no DeepSWE? Why none of the standard benchmarks SOTA providers use?” The user @AvenirReym posed a more diagnostic question: “If it holds on a benchmark made after the model’s training cutoff, it’s real. If it only wins on AIME-style sets that have been circulating for years, it’s leakage.”

The paper’s authors appear to have anticipated these objections. The technical report states that training sets “have undergone strict benchmark decontamination,” including n-gram-based filtering to remove “n-gram overlaps with evaluation sets.”

The LeetCode contest evaluation — which covers contests from April 25 to May 31, 2026, dates that postdate any plausible training data cutoff — represents the most robust guard against data contamination concerns. On those contests, VibeThinker-3B passed 123 out of 128 first-attempt submissions, a 96.1 percent rate that exceeded GPT-5.2, Doubao Seed 2.0 Pro, Kimi K2.5, and Claude Opus 4.6 under identical evaluation conditions.

Still, real-world user reports suggest a significant gap between benchmark performance and practical utility — a phenomenon that has become familiar across the industry. “In LM Studio it only responds well to first question, next questions reply to the first question,” reported @luismolinaab.

Why a social media company may have found a crack in the scaling hypothesis

Even the sharpest critics acknowledged that achieving these benchmark numbers at 3 billion parameters — regardless of how transferable they are to production use cases — is a meaningful engineering achievement. “Even if it’s benchmaxxing doing so with 3B parameters is fascinating, goes to show how fast this field is progressing,” wrote @rohityin.

The observation cuts to a question that has consumed the AI industry since the advent of the scaling hypothesis: Is bigger always better? The conventional wisdom, articulated most famously in the Chinchilla scaling laws and reinforced by the commercial dominance of ever-larger foundation models, holds that more parameters and more training data reliably yield better performance. The economic corollary is stark: training and deploying frontier models costs tens or hundreds of millions of dollars, creating enormous barriers to entry.

VibeThinker-3B challenges that consensus — but only partially. The paper is careful to draw a boundary around its claims, distinguishing between tasks with “clear verification signals” and those that require broad factual knowledge. The Parametric Compression-Coverage Hypothesis explicitly argues that small models cannot replace large ones across the board.

“The true significance of VibeThinker-3B does not lie in proving that a 3B model can replace large-scale generalists,” the paper states, “but rather in providing a concrete empirical signal: the development of compact models is no longer merely a passive compromise for deployment efficiency or cost control; it emerges as a promising research trajectory that is fundamentally complementary to the traditional parameter scaling paradigm.”

Perhaps the most surprising element of the work is its provenance. Sina Weibo — publicly traded on Nasdaq and Hong Kong, with a market capitalization that fluctuates in the single-digit billions — is not a company typically associated with frontier AI research. Yet the VibeThinker series is Weibo’s second major open-source AI contribution in seven months.

VibeThinker-1.5B, released in November 2025, demonstrated that a model with just 1.5 billion parameters could outperform the original DeepSeek R1 on several math benchmarks — a result the team achieved for what it claimed was a post-training cost of just $7,800, compared to the $294,000 estimated for DeepSeek R1.

The research team is compact — nine authors, all listed as Sina Weibo Inc. employees. The model is released under the MIT License, one of the most permissive open-source licenses available, and the weights are freely downloadable from both Hugging Face and ModelScope. Within the first day of release, community members had already created GGUF quantizations and derivative models.

Small models, big implications, and the question the AI industry can no longer avoid

The most honest assessment of VibeThinker-3B may be that it is simultaneously less and more than what the benchmarks suggest. Less, because a model that struggles with basic knowledge of popular developer tools is unlikely to replace any production-grade coding assistant anytime soon. More, because the underlying insight — that reasoning ability and factual knowledge are partially decoupled, and that the former can be compressed far more aggressively than previously assumed — has profound implications for how the industry thinks about model design, deployment economics, and the accessibility of advanced AI capabilities.

If the Parametric Compression-Coverage Hypothesis holds, it suggests a future in which small, specialized reasoning engines operate alongside large knowledge-rich models in hybrid architectures — a vision where a 3-billion-parameter model handles the logical heavy lifting while a larger system supplies the factual grounding. Such an architecture could dramatically reduce the cost of deploying AI reasoning capabilities, potentially bringing competition-level mathematical and coding performance to devices with modest hardware.

“The interesting part is that we’re starting to separate knowledge from reasoning,” wrote @RealLambdaFlux on X. “A small model with strong post-training can punch way above its size on tasks with clear feedback.”

@cmitsakis suggested the practical endgame: “I think small models are the future for agents because they can use tools to get the knowledge and they can run fast and cheap.”

Whether that future arrives through VibeThinker-3B specifically, or through the dozens of teams now racing to reproduce and extend these results, the paper has already accomplished something that no benchmark score can fully capture.

It has forced the AI community to confront an uncomfortable possibility: that for years, the industry may have been spending billions of dollars scaling up parameters to improve a kind of intelligence that could have fit, all along, on a laptop. The weights are public. The code is open. And the most important test isn’t on any leaderboard — it’s whether anyone can make a model this small actually useful in the real world.

Source link

Tech

Is Gen Z Actually Bringing Back Manual Transmissions?

Published

5 minutes ago

27 July 2026

NewsAdmin

The manual transmission occupies a strange, almost contradictory position within modern car culture and the auto industry itself. On one hand, car enthusiasts adore the manual transmission for its fun factor and added engagement, so much so that there are entire communities dedicated to “saving the manual.” On the broader industry side, however, things are not so rosy for the manual gearbox, with a decreasing list of manual cars available each year and the vast majority of them coming in niche sports models.

There has, however, been some sense of the manual transmission making a comeback recently, with some attributing this to the increasing popularity of manuals among Gen Z drivers. Is this actually happening though? Are young drivers leading a surge in demand for new manual vehicles that’s causing automakers to rethink their product strategies? That part isn’t so clear.

There may be strong interest in manuals from some Gen Z drivers, but the actual market size of this group, and the influence it has on the industry, might be a bit overstated. The trend highlights the importance of distinguishing between the vocal online commentariat and real-world car buyers. It also underscores the distinction, no matter the age group, between niche car enthusiasts and the mainstream auto market.

Why Gen Z Is Drawn to Manual Transmissions

For decades, a manual transmission with a clutch pedal was default equipment on a huge portion of cars, trucks, and SUVs sold in America. In fact, because of this,some people even called them “standard transmissions.” Many buyers would pay extra for the convenience of an automatic, but as recently as the mid-2000s, it was still fairly common to see manual transmissions on a lot of normal cars and trucks. Gradually, though, the manuals became a niche option that had to be specifically sought out by drivers who wanted to row their own gears.

With the decline of mainstream manual-equipped vehicles, many younger drivers grew up without ever needing to learn how to drive one. Car enthusiasts, though, are different from your average driver, and for them, the decline in manual options has brought an increased reverence for the market’s stick shift models.

It makes sense that Gen Z might be driving recent interest in manual transmissions in ways that older generations didn’t when you consider they’ve grown up in a world full of automatic cars and EVs, the latter of which have single-speed transmissions rather than a multi-speed gearbox. There is, however, a difference between a group showing interest in manual cars and having the means to go out and buy brand-new ones in significant numbers.

Manual Enthusiasm Doesn’t Equal a Market Comeback

The exact degree to which the manual is or isn’t making a comeback is hard to discern, especially when you factor in new car sales. While there is strong demand at times for manuals among enthusiasts, that interest isn’t always enough to sustain mainstream offerings. For example, manual Volkswagen Golf GTI take rates in the U.S. surged to 41% of the hatchback’s orders in 2024. Still, these take rates weren’t enough to make Volkswagen change its mind about bringing back the manual Golf GTI after axing the stick option for the 2025 model year. On the other hand, you have Subaru planning to add three manual cars to its 2027 lineup.

A 2023 article from the Wall Street Journal highlighted growing manual interest among younger drivers, but some of the enthusiasts profiled in the story were driving older ones. The article also brought up the manual share of the new car market rising from 0.9% to 1.7% from 2021 to 2023. However, according to The Washington Post, preliminary government data shows an all-time low of 0.6% of new vehicles manufactured for the U.S. featuring stick shifts. These numbers reinforce the idea of niche manual enthusiasm (especially among younger drivers) versus a broad market transformation.

It’s unlikely we ever return to the days when Camrys, CR-Vs and F-150s came standard with manual gearboxes, but continued interest from young enthusiasts should help keep today’s dwindling number of manual cars from disappearing completely.

Source link

Tech

Apple Planning to Launch a New Upgrade Leasing Program Next Week

Published

14 minutes ago

27 July 2026

NewsAdmin

According to reports, Apple plans to introduce its new Apple Upgrade leasing plan in the United States next week. The program aims to spread the cost of eligible devices into monthly installments. It may also act as an alternative to the existing iPhone Upgrade Plan offered by Apple.

Apple is expected to launch the Apple Upgrade Program in partnership with Klarna. Customers will reportedly need to complete a soft credit check before enrolling. According to reports, the Apple Upgrade Program is expected to work just like a subscription. Customers have the option of paying for their device in full before the end of their lease period.

Customers can also retain the device after paying off the entire amount. If the customer is no longer interested in the device, they can give it back when the lease expires. The customer gets more freedom compared to a one-time purchase. Apple is likely to offer various lease deals depending on the device. iPhones and Apple Watch can be leased for 24 months. Macs and iPads will have a 36-month lease.

Eligible Devices, Exclusions, and Changes for Buyers

The plan is said to cover eligible iPhones, iPads, Macs, and Apple Watches. But not all Apple products are set to be covered by the new lease plan. This is because some products will be excluded from the list, among them the iPhone 16, Apple Watch SE, budget iPad, and MacBook Neo. Apple also plans to exclude business and education purchases.

Unlike the current iPhone Upgrade Program, the new service is not expected to offer AppleCare coverage. Reports also suggest Apple will stop accepting new customers for the existing iPhone Upgrade Program after Apple Upgrade launches. The new leasing option may make premium Apple devices easier to afford. Customers can spread payments over several months instead of paying the full price upfront.

Source link

Tech

Google Maps’ biggest Android Auto upgrade is reaching more users

Published

26 minutes ago

27 July 2026

NewsAdmin

Google is widening the rollout of two long-awaited Google Maps features for Android Auto users: Immersive Navigation and a built-in live speedometer. After months of limited availability, reports suggest both features are now reaching a broader group of users, including people running stable versions of Google Maps instead of beta builds.

The update is part of Google’s ongoing effort to modernize the Android Auto navigation experience. While the company unveiled Immersive Navigation earlier this year as one of the platform’s biggest upgrades in years, the feature has been trickling out slowly, leaving many users without access. That appears to be changing, with more devices now receiving the update through a server-side rollout.

The rollout is still gradual, and availability may vary depending on region and device, but it signals that Google is becoming more confident about deploying the new experience at scale.

Maps becomes more visual while Android Auto gains a useful driving tool

Immersive Navigation makes Google Maps easier to follow while driving by displaying richer road layouts, clearer lane guidance, and more detailed intersections. The updated interface is designed to present navigation information in a way that’s easier to interpret with a glance instead of requiring drivers to study the map.

Alongside the redesigned navigation view, Google is also expanding the rollout of a live speedometer inside Android Auto. Unlike speed limit warnings, the feature displays the vehicle’s current speed directly within Google Maps, reducing the need to switch attention between the infotainment display and the instrument cluster in vehicles where that information isn’t readily visible.

Users receiving the update will also notice another interface change. On Android phones, the Gemini icon has replaced the Google Assistant shortcut in the top-right corner of Google Maps, continuing Google’s transition from Assistant to Gemini across its ecosystem.

A server-side rollout means patience is still required

Although more users are reporting access, Google has not released the features through a standard app update. Instead, the rollout appears to be happening server-side, meaning installing the latest version of Google Maps or Android Auto does not necessarily guarantee immediate access. Reports also suggest the features are appearing on stable builds rather than remaining exclusive to beta users, making them accessible to a much wider audience over time.

For Android Auto users, this update is less about adding flashy new capabilities and more about refining the everyday driving experience. Better lane visualization, more detailed navigation guidance, and an integrated speedometer are incremental improvements individually, but together they make Google Maps feel more complete as an in-car navigation system. As the rollout expands, more drivers should begin seeing those upgrades without needing to join Google’s beta programmes.

Source link

Tech

The Next Solar Eclipse Will Deliver a Rare Sight: Totality at Sunset

Published

40 minutes ago

27 July 2026

NewsAdmin

It’s been two years since the last total solar eclipse, but another one is coming in a few weeks — and with it, a special addition that rarely happens. Some people will get to see the solar eclipse during sunset, giving them an extraordinary view. The only problem is, you may have to jet off on a summer vacation to see it.

The total eclipse will only grace the shores of a few countries. According to NASA, the total eclipse on Aug. 12 will hit Greenland, Iceland, the northern half of Spain and one corner of Portugal. Spain is at the tail end of the eclipse, and as such, it will occur later in the day at around sunset, giving viewers in Spain a spectacular sight.

Other countries in Europe and Africa will also get quite a show. Most of Europe and a sizable portion of northern Africa are still getting a partial solar eclipse, which will happen around sunset. That means tens of millions of people will be in the right place to see a partial or total solar eclipse at sunset.

Russia is the luckiest country this time around. The eclipse starts at sunrise near the uninhabited Taymyr Peninsula, streaks across the world all day, and then Russians on the other side of the country are in the path for the partial solar eclipse at sunset, making Russia the only country that will see the eclipse at both sunrise and sunset.

How often do sunset eclipses happen?

Sunset eclipses don’t occur very often, at least where people can see them. There are two reasons for this.

The first is that solar eclipses usually only last for a few hours, and the timing for that being right at sunset isn’t terribly common.

The other reason is that solar eclipses have long paths, and sometimes those paths cut through the ocean or places where people don’t live. In the last 15 years, only a handful of total eclipses sit in that Venn diagram of occurring late enough in the day to happen at sunset, while also taking place where humans can see it.

The most recent sunset eclipse in North America was in 2014, but it was only a partial eclipse. South America’s most recent one was in July 2019, where people in Chile and Argentina had the best view of a total solar eclipse happening at sunset, while most of the rest of the country had a partial view. Asia’s most recent was in December 2019.

In short, this is a once-in-a-decade event for most of the world.

A map of the Aug. 12 solar eclipse trail — The path of this year’s solar eclipse takes it across the northern portions of Earth before ending in Europe. North Americans won’t see much.
NASA

Will this eclipse be visible in North America?

North America is pretty far from the main action, but a few states are in the path. The math is a little complicated, but if you look at a map of the US and draw a line from Alaska to North Carolina, all the states north of that line will get a little bit of an eclipse (less than 20% coverage), and all the states south of that line are sitting this one out.

If you live in the path and want to check it out, be sure to adhere to proper solar eclipse viewing safety protocols so that you don’t harm your eyes.

Source link

Tech

A significant Apple Watch upgrade is planned, but not for years

Published

53 minutes ago

27 July 2026

NewsAdmin

Apple is reportedly testing its September 2026 Apple Watch lineup, but the new models are expected to be minor updates while more substantial changes are much further off.

Most recent rumors have claimed that the forthcoming Apple Watch Series 12 will not feature any hardware improvements over its predecessor. Now backing up those reports, Bloomberg says that neither the Apple Watch Series 12 nor the expected Apple Watch Ultra 4 will get visible upgrades to their design.

Instead, the updates will be concentrated on giving the new Apple Watches a significant performance boost through a new processor. There may also be health and fitness tracking improvements, though previous reports have claimed that there won’t be new sensors, so this may be a software update.

The new lineup has the codenames N237 and N238 for the Wi-Fi and cellular versions of the Apple Watch Series 12 while the Apple Watch Ultra 4 is codenamed N240. All three models are said to now be in late-stage testing.

It’s claimed that these are the only three models due to be launched in September 2026, and specifically that there will not be a new Apple Watch SE released at this time.

There’s no indication of when an upgraded Apple Watch SE may come, but the report says that work is being done on a significant redesign for the regular Apple Watch. That’s not expected to result in a new device for at least a couple of years, however.

Separately, a recurring rumor has claimed again that the 2026 Apple Watch models will feature a new system for attaching bands. It remains as possible yet unlikely, though, as it has for the last several years that the rumor has circulated.

Also, while this new report predicts no noticeable changes to the Apple Watch Ultra, a rumor in May 2026 suggested that there may well be a significant update. Supply chain sources claimed that the Apple Watch Ultra 4 could double its number of sensors over the previous models.

Source link

Tech

A New Middle Class of Content Creators Is Quietly Quitting the 9-to-5

Published

1 hour ago

27 July 2026

NewsAdmin

“The rise of TikTok, Instagram Reels and Amazon storefronts has created a new kind of white-collar exit strategy,” reports Bloomberg. Workers ditch office jobs not to become celebrities, necessarily, “but to piece together an income online through brand deals, affiliate links and highly personal videos documenting everyday life.”

In many cases, the followers necessary to sustain a living are smaller (and more attainable) than people might assume. A small but loyal audience can now generate enough income to rival a midlevel salary. Welcome to the middle-class creator economy. Last year, 25-year-old Abi Platock balanced a corporate marketing job in New York while posting online in her spare time. She built her audience by posting one or two videos a day, offering career advice, beauty tips and daily vlogs. “I signed my first brand deal in the four-figure range, and for me that was just such a big eye-opening moment,” Platock says of her partnership with deodorant brand Secret. She had 8,000 followers on TikTok at the time. “You can totally make it work without having hundreds of thousands of followers.” Platock, who now has roughly 25,000 followers across platforms, has signed about $25,000 in brand deals so far this year and expects her annual creator income to reach around $50,000 by yearend.

Her experience reflects a broader shift in advertising. Brands are increasingly moving money toward so-called microinfluencers — smaller online personalities who have less than 100,000 followers. “They are hiring a bunch of microcreators at scale instead of hiring a handful of macrocreators for what could potentially be the same cost,” says Ali Grant, co-chief executive officer of the Digital Department, a creator management company. And they perform where it matters most: engagement. An engagement rate of 3% is considered strong, and some microinfluencers exceed 10%, Grant says of the closely watched metric that tracks how often followers interact with content through likes, comments, shares and saves. Microinfluencers average a 3.2% engagement rate, almost triple the 1.1% rate for macroinfluencers (more than 1 million followers), according to growth marketing agency ATTN… A TikTok partnership with a creator who has around 50,000 followers can run a brand more than $3,500 for a single post, Grant says; with 10 times the followers, that fee might just triple, to around $10,000….

The influencer marketing economy ballooned to a projected $33 billion in 2025 up from $1.7 billion in 2015. The segment gained momentum after the COVID-19 pandemic, as dissatisfaction with traditional work pushed many to reconsider conventional career paths, says Brooke Duffy, a professor of communications at Cornell University. “They realized the trade-offs in terms of the investments of time, energy and human capital were not necessarily worth sacrificing so much of one’s personal self for,” she says. Success online can bring greater freedom — and even higher pay than many traditional office jobs, which have a median US salary of $69,000, according to Glassdoor. But the middle-class hustle still requires constant effort to maintain. The career has no promise of lifetime longevity. And unlike traditional workers, creators have no predictable paycheck or job protections, making career stability elusive. Roughly 57% of 3,000 surveyed full-time creators earn below a living wage from content creation, according to a report last year from Influencer Marketing Hub. Income from social media can fluctuate wildly from month to month, driven by shifts in algorithms, sponsorship cycles and platform trends.

“You could have a month where you make zero dollars, or you could have a month where you make $10,000,” Platock says.
The article cites Gallup Poll data released last year that found employee engagement in the U.S. had fallen to its lowest level in a decade [with engagement defined as “the psychological attachment workers have to their work/team/employer]. “Among the hardest-hit groups were Generation Z and workers in finance and technology. Broader workplace challenges, including rapid organizational change, hybrid and remote work transitions, and rising employee expectations are considered drivers of the overall trend.”

“For many workers, influencing can seem like a better deal; flexible schedules and independence wrapped in a veneer of creativity and fun. Almost 60% of Gen Zers say they’d become an influencer if given the opportunity, according to a 2023 survey from Morning Consult.”

Source link

Tech

How to get the best deal on AppleCare One

Published

1 hour ago

27 July 2026

NewsAdmin

Previously exclusively in the US, AppleCare One is now launching in the UK, France, Germany, and Australia, with Apple’s best insurance deal for users with multiple devices — as long as you’re careful in selecting what’s covered.

A year after it launched in the US, AppleCare One is expanding outside of the US. It’s only going to four more countries, and they’re countries you’d expect it to launch in, but that’s a start.

“At Apple, we’re focused on creating and delivering exceptional experiences,” Bob Borchers, Apple’s vice president of Worldwide Product Marketing, said in a statement to AppleInsider. “With AppleCare One, customers in the UK can now enjoy the trusted protection of AppleCare+ in a way that’s simpler and more flexible than ever before — one plan, one price, and the peace of mind that comes with knowing all their eligible products can be covered.”

Full details of the terms, conditions and all pricing have yet to be published, but based on the details provided by Apple UK, the program will cost around the same as it does in the US. It will also offer the same befits, which are:

Up to three products covered
Theft and Loss Coverage (for iPhone, iPad, and Apple Watch)
Unlimited repairs for accidental damage
Free battery replacement
Priority access to Apple support

There are limits in that, for instance, AppleCare One users may only make up to three claims of theft or loss per year. But then there are also extra benefits in that iPad accidental damage from handling (ADH) coverage can include an associated Apple Pencil or Apple-branded iPad keyboard.

Who this does and does not work for

Users who have any single device, such as one iPhone or one iPad, should not take up the new AppleCare One option. They should use AppleCare+, which Apple has also improved.

That AppleCare+ plan used to only feature theft and loss coverage for the iPhone, but it now extends this to the iPad and Apple Watch. AppleCare+ prices vary depending on the model of device, but for example the monthly cost in the US at time of writing is:

iPhone: from $9.99 to $13.99

iPad: from $5.49 to $11.49

Mac: from $3.99 to $17.99

Apple Watch: from $2.99 to $5.99

Apple Vision Pro: $19.99

Each of these comes with an annual version which is roughly equivalent to 10 months at the monthly rate. Note that AppleCare+ only allows annual payment for insuring displays, Apple TV, HomePod, or AirPods.

Those items can, though, be paid for monthly via the new AppleCare One. Again, non-US details will not be fully available until AppleCare One launches on August 4, but the US version does allow adding headphones, for example.

Nonetheless, users who want to insure single devices get no financial benefit from the new AppleCare One. Users who have two devices will definitely benefit if those devices include the Apple Vision Pro.

Covering the Apple Vision Pro by itself with AppleCare+ is exactly the same price as covering it via AppleCare One. So that would be like getting coverage for a second and even third device for free.

Devil in the details

There are ways in which AppleCare One’s coverage of three devices is more than the price of insuring them each with a separate AppleCare+ plan. It depends on if the devices include a Mac, which on its own ranges from $3.99 per month for a Mac mini, to $17.99 per month for a Mac Pro.

Or with the iPhone, the separate monthly cost is $9.99 for an iPhone 17e, rising to $13.99 for an iPhone Air, iPhone 17 Pro, or iPhone 17 Pro Max.

It naturally gets more complicated if you have both an iPhone and a Mac in the equation. For example, if the three devices to be insured consist of an iPhone 17e, Mac mini, and an Apple Watch SE, the total individual cost is $16.97 where AppleCare One is $19.99 and you shouldn’t go near it.

But then if the devices are, say, an iPhone 17 Pro Max, an M5 13-inch iPad Pro, and a Mac Studio, you’d save a startling $12.48 per month by going to AppleCare One. In that case, that’s a hell of a deal.

That’s if you stick to just the basic AppleCare One and its coverage of three devices. It’s possible to add a fourth or any number of more devices, for $5.99 per month each.

Do that by adding, say, an Apple Vision Pro to the example with the iPhone 17 Pro max, 13-inch iPad Pro, Mac Studio and your monthly cost goes up to $25.99. The cost of doing these separately is more than double at $57.46.

One more huge benefit

Not long ago, all of this comparison of coverage costs would be moot because you were limited to which devices could get any AppleCare. It was typically a new device, or a device bought in the last 60 days.

Now with AppleCare One, the coverage is not only cheaper for most people in most circumstances, it is broader. Instead of solely being for new devices, AppleCare One can potentially be used for Apple devices that are up to four years old.

Those devices have to be in good condition, and during online registration users are prompted through questions regarding potential damage. It’s also possible that Apple will require the device to be brought to a store for a visual inspection.

If a user is starting with a new device, then instructions for signing up to AppleCare One will be displayed in Settings. Otherwise it can be done via the Apple website using the user’s Apple Account.

Those users who already have AppleCare plans will be able to switch to AppleCare One. Apple says that their existing plans will be cancelled and a new AppleCare One plan put in place.

Apple’s best deal

As long as you check out the pricing differences between AppleCare+ and AppleCare One, this new program can represent a very significant saving. So it’s unquestionably worth examining the details once Apple has published them for the UK, France, Germany, and Australia, on August 4, 2026.

Note, though, that the US service had some teething problems with eligible devices not always being displayed. If that happens again with the new countries, there are steps you can take to get the correct coverage.

Source link

Tech

Google goes it alone with a new cybercrime crew taxonomy

Published

2 hours ago

27 July 2026

NewsAdmin

Security

So much for Microsoft and CrowdStrike’s plans for consistent names across the industry

Google has created a new taxonomy to describe cybercrime outfits, seemingly abandoning a Microsoft-led effort to create consistent names.

The Big G announced its new schema on Saturday in a post that notes its 2022 acquisition of Mandiant and its subsequent incorporation into a new team called the Google Threat Intelligence Group (CTIG).

Now that two have become one, Google reckons they need consistent naming conventions to describe cybercrime crews.

The result is a two-word schema in which the first word “is a unique and memorable term chosen to represent the specific actor.” If security folk have already applied a particular moniker Google will use it, otherwise it will randomly generate a word “to remove bias.”

Google says the second word “categorizes threat clusters by motivation, attribution, or activity type based on which category we consider to be most important for defense and response strategies.”

Tech

Scanwheel: A Pocket-Sized POV TV

Published

2 hours ago

27 July 2026

NewsAdmin

When you hear the word TV, you probably think of a big LED screen, maybe even the old CRT TVs, but in either case it’s something large and fairly complicated. However, thanks to the persistence of vision, it doesn’t have to be. In this handheld-sized project from [Ancient], the Scanwheel is born, a miniature mechanical TV that uses a spinning disk and some LEDs to produce an image.

The electronics of the Scanwheel are pretty straightforward. The smarts come from a Raspberry Pi Pico, an A4988 motor driver, a couple of LEDs, and a small 21-02485 stepper motor. The Raspberry Pi Pico is used to command the motor speed as well as coordinate the LEDs to turn on at the right time. The case is 3D printed; the base includes space for the various support electronics as well as some small light baffles to ensure the LEDs don’t bleed over outside their intended area. The top of the case is a disk that includes 20 small holes spaced evenly around the perimeter at varying heights, allowing light to only leave the disk when one of these holes is in front of the LEDs.

When you put all these pieces together, spin the motor up to roughly 900 RPM, and turn the LEDs on in a precise order, you end up with a really cool result: a miniature TV. And due to the five different LEDs in this build, you actually have a color 20×20 pixel display in the center and, on either side of that, two more 20×20 black-and-white displays capable of showing different images. Thanks [Ancient] for sharing this awesome build that takes advantage of the persistence of vision effect to create a unique display. Be sure to check out the video below as well as the instructions on how to build your own. And if you enjoy this sort of thing, check out some of our other persistence-of-vision projects as well.

Source link

Tech

Yeast Turns Sugar Into Liquid Power in a Paintball Bottle

Published

2 hours ago

27 July 2026

NewsAdmin

Fermentation Filling High Pressure CO2 Bottles
Yeast spends its days chewing through sugar and splitting the leftovers into alcohol and carbon dioxide. Most people chasing homemade ethanol treat the second half of that reaction as pure waste and let the gas drift away. One maker decided the gas was too useful to ignore and set out to trap every molecule, dry it, chill it, and pack it into the same kind of high-pressure bottles that drive paintball markers and soda siphons.

The numbers appear almost too clean, since 4 kilograms of ordinary sugar dissolved in 14 liters of water already gives a solution that is nearly 22% sugar. If the yeast performs its job and converts everything, the process should result in little more than 2 kg of CO2. That’s enough liquid to fill nearly four 20-ounce paintball cylinders. The problem is that the gas comes out of the fermenter wet and diluted, making the first job (gathering it) difficult, as does maintaining the pressure up and preventing air from entering the system.

Sale

Airmoto Tire Inflator Portable Air Compressor for Cars, Bikes and Bicycles

QUICK & ACCURATE: Built with the highest quality materials, Airmoto is a portable tire inflator for car, trucks, SUVs, motorcycles, and bicycles…
AUTO SHUT OFF: Use the “+” and “-” control buttons on the Airmoto to easily select your desired pressure, simply connect the air hose to your…
MAIN HIGHLIGHTS: Easily change between PSI, kPa, BAR, and KG/CM pressure units with a push of a button; Easy to read large LCD; Digital Tire Pressure…

A 5-gallon water jug serves as a fermentation tank. The carbon dioxide is routed out via an airlock tube and into a recycled water-filter canister. The works is stuffed with silica-gel beads, which reduces moisture slightly, but we later discovered that the dew point remains too high, causing ice to form inside the valves. The next step is to transfer the gas to a beach ball. It takes a few days, but the ball eventually fills up with hundreds of gallons of CO2. It serves the purpose of providing some extra room to keep the pressure near the proper level while the yeast is still active.

Fermentation Filling High Pressure CO2 Bottles
The major issue is turning the squishy substance into liquid. At room temperature, the CO2 must be compressed to roughly 64bar before it can condensate. The problem is that standard shop compressors can only reach a fraction of that capacity. The solution is to simply leave it in the air box. To cool a copper coil, a DIY system makes use of propylene as a refrigerant. This lowers the temperature to roughly -33 degrees Celsius and reduces the condensation pressure to about 13 bar absolute, which is well within the capabilities of a severely modified oil-less air compressor with its over-pressure cut-out switch disabled.

Fermentation Filling High Pressure CO2 Bottles
That copper coil is a 2-inch pipe, approximately 2 feet long, with a thinner copper coil within to convey the propylene. The CO2 from the beach ball enters at the top, meets the chilly surface, and condenses into a liquid that gathers at the bottom. A second coil (the same as the first) is housed in a 96% ethanol-lined thermos. A paintball tank sits in that bath, keeping the metal cool. Once some liquid has accumulated in the coil, a valve opens and the liquid flows into the chilled tank.

Fermentation Filling High Pressure CO2 Bottles
An typical oil-free compressor can move the gas, but only at a very sluggish rate; at 17 bar, it moves like a snail. Switching to a refrigerated compressor provides the necessary pressure, 400 psi or greater, and reduces fill time to 10 or 15 minutes. But now we have a new problem: oil separation. Any lubricant that gets into the tank degrades the purity. Then there’s water, which still freezes inside the tank valve while we pump it out, and this can jam the nozzle until the metal heats up again.

Fermentation Filling High Pressure CO2 Bottles
After filling the bottle to capacity, the scales read 1312 grams with the valve still connected. When the contents were drained, 974 grams remained, indicating that 338 grams of liquid carbon dioxide had been trapped inside. The container wasn’t even full to the brim, but that liquid was unmistakable, and when that valve was opened quickly, the temperature of the tank dropped to the point where it iced over. If you discharge it completely and quickly, you could bring it down to the temperature of dry ice.

Fermentation Filling High Pressure CO2 Bottles
That small charge of ours already has some substantial practical power behind it. By connecting it to a short-stroke pneumatic actuator, he was able to elevate the back end of a full-size pickup approximately 200 millimeters off the ground. The same gas, linked to a vane motor, was able to power a small generator for a few minutes, but you can probably predict where this is going: the intense chill that comes in as the liquid boils away causes the pressure to drop and the motor to turn off.
[Source]