AI News Archive: May 25, 2026 — Part 5
Sourced from 500+ daily AI sources, scored by relevance.
- Awfis Q4 profit doubles, revenues up 20% YoY on GCC boom and AI-led expansion
Co-working spaces remained the company’s biggest revenue contributor, accounting for more than 83% of operating revenues. Income from the segment rose 26.8% on-year to Rs 341.5 crore in Q4 FY26 from Rs 269.3 crore in the year-ago period. The construction and fit-out projects business contributed Rs 68.6 crore during the quarter.
- Teacher-founded AI edtech Diotima spins out from Trinity
Diotima received €500,000 under Enterprise Ireland's Commercialisation Fund last year. Read more: Teacher-founded AI edtech Diotima spins out from Trinity
Score: 31🌐 MovesMay 25, 2026https://www.siliconrepublic.com/start-ups/teacher-founded-ai-edtech-diotima-spins-out-from-trinity - Boundless Leader Launches AI-Native Precision Matching Platform for High-Capacity Leaders, Revenue-Generating Pre-Launch
Boundless Leader Launches AI-Native Precision Matching Platform for High-Capacity Leaders, Revenue-Generating Pre-Launch azcentral.com and The Arizona Republic
- New watchOS 27 Rumor Tips Better Heart-Rate Tracking, Delayed AI Health Coach
New watchOS 27 Rumor Tips Better Heart-Rate Tracking, Delayed AI Health Coach PCMag
Score: 31🌐 MovesMay 25, 2026https://www.pcmag.com/news/new-watchos-27-rumor-tips-better-heart-rate-tracking-delayed-ai-health - Salesforce and Snowflake Earnings to Focus Attention on AI’s Software Impact
Salesforce and Snowflake Earnings to Focus Attention on AI’s Software Impact The Information
- HCLTech, Pegasystems Expand AI Partnership to Modernise Legacy Enterprise Systems
Discover how HCLTech and Pegasystems are transforming legacy systems with AI. Learn more about their innovative partnership and solutions!
- Three ways to avoid being fooled by AI slop
Fact-checking can take hours or days while fakes can be created in seconds. So, what do we do?
Score: 30🌐 MovesMay 25, 2026https://theconversation.com/three-ways-to-avoid-being-fooled-by-ai-slop-282974 - The Acer TravelMate X 14 AI is a solid mid-range pick that strives to compete with the heavyweights
The Acer TravelMate X 14 AI is a solid mid-range pick that strives to compete with the heavyweights IT Pro
- Microsoft will let you uninstall Copilot app as Windows 11 clean-up moves ahead
Microsoft’s latest Windows 11 update introduces new ways to permanently remove Copilot using Group Policy, Registry changes, and traditional uninstall methods.
- [PODCAST] Data Protection and Privacy in the Age of Artificial Intelligence: In Conversation with Mr Adrian Mak
[PODCAST] Data Protection and Privacy in the Age of Artificial Intelligence: In Conversation with Mr Adrian Mak Oxford Law Blogs
- AI won’t replace you but someone using AI might
Generative AI is transforming the workplace faster than ever, but new research from the University of Vaasa suggests the biggest threat may not be AI itself — it’s falling behind in learning how to use it. Researcher Zhe Zhu found that employees who see tools like ChatGPT and Gemini as helpful collaborators rather than job-stealing rivals tend to be more engaged, adaptable, and optimistic about their careers.
- Why some AI bets scale and others quietly die
Technology leaders say AI initiatives fail when they remain isolated demos and succeed when linked to workflows, ownership, data quality and measurable P&L impact.
- Twilio’s AI Boost Is a Double-Edged Sword
Twilio’s AI Boost Is a Double-Edged Sword The Information
Score: 30🌐 MovesMay 25, 2026https://www.theinformation.com/articles/twilios-ai-boost-double-edged-sword - DJI Says Mass Adoption of Delivery Drones, Flying Vehicles Still on the Distant Horizon
DJI Says Mass Adoption of Delivery Drones, Flying Vehicles Still on the Distant Horizon Caixin Global
- Apple to Introduce Improved Genmoji, Image Playground Upgrades With iOS 27 Update: Mark Gurman
In the new edition of the Power On newsletter, Mark Gurman reported that Apple is preparing major upgrades for Genmoji and Image Playground in iOS 27. “Apple’s own models for Genmoji and Image Playground have been improved, so quality is getting a big boost this year,” Gurman said. Apple is expected to expand Image Playground’s capabilities by adding support f...
- Microsoft Replaces Claude Code With GitHub Copilot CLI
The move is partly tied to cost-cutting efforts, with Microsoft’s financial year ending on June 30.
Score: 30🌐 MovesMay 25, 2026https://analyticsindiamag.com/ai-news/microsoft-replaces-claude-code-with-github-copilot-cli - Andrew Phillips: I discovered a clarion call on the threat of AI from a most unlikely source
Andrew Phillips: I discovered a clarion call on the threat of AI from a most unlikely source Toronto Star
- SoftBank Shares Hit Record With Lift From OpenAI IPO Hopes
SoftBank Group Corp. shares climbed to a record high, spurred by hopes of big returns from the Japanese investor’s stakes in OpenAI and SB Energy Corp., should the two go public.
- Influence360 Launches as the First AI & Data-Driven Web3 KOL Platform with Global KOL Coverage and Real Attribution
Influence360 Launches as the First AI & Data-Driven Web3 KOL Platform with Global KOL Coverage and Real Attribution USA Today
- Introduction to special issue on FAIR Principles and Machine Learning, AI Readiness and AI Reproducibility
AI Magazine, Volume 47, Issue 2, Summer 2026.
- Motion tracking system shows robots the path most traveled by, keeping them on task
There's a delicate art to teaching robots, even when you're preparing them for predictable environments like factories, where they'll repeat the same tasks a little differently depending on the obstacles they face. Whether a human is suddenly in their way or there's new clutter, the machine must closely mimic its operator's actions by staying on a trajectory (or motion path).
Score: 29🌐 MovesMay 25, 2026https://techxplore.com/news/2026-05-motion-tracking-robots-path-task.html - LTM Bags AI-Led IT Infra Modernisation Mandate from UK’s SSP Group
Discover how LTM is modernizing SSP Group's IT infrastructure using AI. Learn more about this innovative partnership and its impact on efficiency!
Score: 29🌐 MovesMay 25, 2026https://analyticsindiamag.com/ai-news/ltm-bags-ai-led-it-infra-modernisation-mandate-from-uks-ssp-group - When AI evolves on its own
Kim Dae-shik The author is a professor at KAIST. Ancient Greeks used the word "mythos" to describe stories or narratives. Like all words, its meaning shifted over time. Beginning in the early 19th century, Europeans increasingly applied mythos only to old tales, especially ancient Greek legends. That is why the word today is often understood simply as “myth.” Anthropic's chief product officer, Ami Vora, its co-founder and president, Daniela Amodei, and co-founder and CEO Dario Amodei present on stage at the Code with Claude developer conference in San Francisco on May 6. [AP/YONHAP] Yet another interpretation exists. In “Poetics” (335 B.C.), Aristotle used mythos to describe the representation or structure of action within tragedy. By contrast, ethos referred to the character performing the action, while praxis meant the action itself. Aristotle argued that the essence of tragedy lay not in character or action alone but in mythos, which connected the two. OpenAI CEO Sam Altman, center, and Greg Brockman, OpenAI president and co-founder, arrive at the federal courthouse during proceedings in a lawsuit against OpenAI in Oakland, California, on April 30. [AFP/YONHAP] That background makes the naming of the latest AI model introduced by Anthropic on April 7 symbolic. The important point is not who created it or how it was built, but the very existence of an AI system with such capabilities. Mythos is described as the most advanced AI system developed so far, especially in coding and hacking. Reports say it identified vulnerabilities not only in operating systems such as Windows and macOS but also in BSD Unix, which has supported much of the global internet infrastructure for nearly three decades. It reportedly went further by proposing hacking strategies based on those vulnerabilities. In theory, if such technology were obtained by terrorist groups or rogue states, it could threaten the global internet system. Online payments, logistics systems and communications infrastructure could all be affected simultaneously. Such dystopian possibilities no longer seem entirely unimaginable. As a result, Anthropic reportedly decided not to release Mythos publicly. Instead, access has been limited to companies connected to the discovered vulnerabilities. That decision raises another issue. Most companies participating in the so-called Project Glasswing are U.S. firms. Korean companies, meanwhile, reportedly cannot access Mythos. The imbalance means that while the U.S. government and certain private companies may now possess tools capable of disrupting foreign information technology infrastructure, many countries, including Korea, lack comparable defensive capabilities. Related Article Rise of AI raises fears of North Korean hacking capabilities The butterfly effect of the Anthropic contract termination OpenAI officials discuss safety protocols with Canada following school shooting OpenAI claims China's DeepSeek trained its AI by distilling U.S. models, memo shows Another concern lies in the nature of modern AI competition. Today’s AI race is driven less by different theories or algorithms than by scale. Since most companies rely on similar mathematical foundations, the decisive factors are massive datasets and large-scale GPU infrastructure. Given enough computing power, creating comparable AI systems eventually becomes a matter of time. Reports suggest that GPT-5.5, released by OpenAI on April 23, possesses coding and hacking abilities similar to Mythos. In effect, Mythos may mark the beginning of unlimited competition among major technology companies. For U.S. allies such as Korea, this competition presents both opportunities and risks. For China, however, it represents a strategic vulnerability. The obvious response for Beijing would be to develop an AI system comparable to Mythos. Yet China’s flagship AI model, DeepSeek V4, released on April 24, reportedly has not demonstrated the performance many expected. Because of semiconductor export controls, China still faces difficulty building data center infrastructure on the scale available in the United States. Under such conditions, matching AI models developed by Anthropic, Google or OpenAI remains challenging. If China cannot catch up through conventional methods, it may seek alternatives. One option would be integrating multiple AI systems developed by different firms into a single national AI champion. Another could involve nationalizing domestic data centers to create state-led computing infrastructure. Japanese Finance Minister Satsuki Katayama speaks to the media after a meeting involving the Financial Services Agency, the Bank of Japan, the National Cybersecurity Office, the country's top three banks and the Japan Exchange Group following concerns about potential vulnerabilities linked to Anthropic's Mythos AI model in Tokyo on April 21. [REUTERS/YONHAP] But if even those measures fail, China could turn to a more dangerous path: recursive self-improvement, or RSI. Proposed in 1965 by British mathematician I. J. Good, RSI refers to a process in which AI rewrites its own code to improve its intellectual capabilities. If successful, such systems could rapidly evolve into artificial superintelligence. During the Cold War between the United States and the Soviet Union, the doctrine of mutually assured destruction, or MAD, rested on the paradox that the ability to annihilate each other deterred the use of nuclear weapons. The article argues that AI, accelerated by systems such as Mythos, is beginning to transform into a strategic weapon surpassing nuclear arms. Unlike the nuclear rivalry of the 20th century, however, the ultimate winner in a 21st-century AI version of MAD between the United States and China may be neither country. Instead, it could be the artificial superintelligence created through recursive self-improvement itself. This article was originally written in Korean and translated by a bilingual reporter with the help of generative AI tools. It was then edited by a native English-speaking editor. All AI-assisted translations are reviewed and refined by our newsroom.
Score: 28🌐 MovesMay 25, 2026https://koreajoongangdaily.joins.com/news/2026-05-26/opinion/columns/When-AI-evolves-on-its-own/2600329 - From TF-IDF to Transformers: Implementing Four Generations of Semantic Search
How did semantic search evolve from simple keyword matching into modern transformer-based language understanding? This hands-on article builds four generations of semantic search systems step by step using Python. The post From TF-IDF to Transformers: Implementing Four Generations of Semantic Search appeared first on Towards Data Science .
Score: 28🌐 MovesMay 25, 2026https://towardsdatascience.com/from-tf-idf-to-transformers-implementing-four-generations-of-semantic-search/ - Tech Week event centers student voices on AI in higher education
What do students actually want and need when it comes to AI? A Tech Week event is ready to help answer the question.
Score: 28🌐 MovesMay 25, 2026https://www.bizjournals.com/boston/news/2026/05/25/higher-ed-ai-challenges-tech-week.html?ana=brss_6150 - The one-person company revolution: How to build more with AI (without losing your mind)
Not long ago, building a company as a solo operator was mostly impractical. Too many moving parts, too much effort, too many skills required. Today, that constraint is rapidly disappearing. With AI, execution has become dramatically cheaper, and in many cases, accessible to a single person. But this shift hides a deeper truth: Execution is […] The post The one-person company revolution: How to build more with AI (without losing your mind) appeared first on e27 .
- A (Slightly) Mechanistic Theory for Exponentially Increasing AI Time Horizons?
AI ‘time horizons’ are mostly not about time (I think it’s mostly ‘data’, but you’ll see where I’m unsure). One chart from 2025 has become perhaps the most (in)famous in modern AI commentary. For those in the know, ‘ the METR graph ’ [1] is unusually compelling because it achieves what so few measures of AI progress have achieved: a somewhat meaningful Y axis (‘time horizon’ [2] ) as well as a somewhat predictable trend over time! (This is remarkably rare !) Frustratingly, the only superficially available takeaway is something like, ‘the line goes up straight-ish over time’. This is better than nothing, but it’s very dissatisfactory from the point of view of getting confidence in the predictions, because it exposes no deeper mechanism. This drives a lot of confusion and argument about the implications. A deeper mechanism would be good for two reasons: It enables a sanity check on the trend, perhaps enabling more confidence in its predictions than we would sensibly allow with only the surface understanding. It gives some way to interrogate when and how the trend might change (because if the deeper mechanism gets deflected, the superficial projection would be broken, but a prediction based on the deeper mechanism might stay viable for longer). (A sub-reason: if we want the trend to change, knowing some more mechanism might shed light on some levers to pull rather than sitting around to wait and see.) As an analogy, a similarly superficial trend, Moore’s Law , can be a little better mechanistically explained by the more general Wright’s Law [3] . This is great, because that law covers more cases, and it can handle some deflection from the trend, or give some idea of when (and under what conditions) the trend might break. Important when looking at plausible futures, and how to steer toward desirable ones! Attempting to find some mechanism in the METR graph Task ‘length’ and success modelling Why did METR focus on ‘task length’? First, it’s not how long the AI agent takes. It’s how long the task in question takes a panel of sampled human experts, on average [4] . So in their ‘time horizon’ measurements, METR is capturing the effective hours of human-expert-equivalent activity that AI agents can carry out. [5] One way to think about the time it takes human experts to complete a task is that, for each subtask they had to know how to do (or be able to figure out how to do) and then successfully execute, the overall task takes incrementally longer. By how much? That depends on exactly what ‘subtasks’ we're imagining breaking things down into. [6] But on average longer tasks correspond to more distinct challenges, all else equal. [7] A random generation of tasks (rows) with ‘subtasks’ as segments, sorted by subtask count from least to most. You can see that the more subtasks, the longer, on average. It’s a little ragged — not all subtasks are the same length, so occasionally fewer, longer subtasks add up to more overall time than more, shorter subtasks. What METR can easily measure is the overall duration. Even if the subtask division is somewhat subjectively defined, duration stands as a reasonable proxy for it. Note that the vertical subtask count axis is sorted but not uniformly spaced. (Created with claude.ai.) This is the first piece of mechanism we should take into account. ‘Time’ is not agent time : it's a noisy estimate for ‘number of somewhat challenging requirements necessary to complete the task’. [8] This is treating overall tasks as formed by something like drawing ‘subtasks’ out of a large collection of possible requirements. Given the agent’s general competence, specific knowledge, tools available, and opportunity to retry or learn on the fly, sometimes the agent can meet these requirements. Other times it can’t. [9] ‘Longer’ tasks simply draw more subtasks (that’s why they’re ‘longer’, in this model: expert humans had more subtasks they needed to carry out). [10] Toby Ord demonstrates one way to take this intuition further , noting that if we explicitly model overall success according to a simple model where chance of failure compounds with task ‘length’, , we get a reasonable fit for the data METR collected. (Interestingly Toby mainly seems to continue treating this as ‘agent time’. I’ll instead take as given that we’re talking about a proxy for number of subtasks.) In other words, for a given AI agent and task domain, there's something like a ‘hazard rate’, (per-subtask probability of failure), which reasonably well summarises (and predicts) the AI's level of success in that domain: (i.e. to succeed at a -step task, the agent must not fail — must avoid the ‘hazard’ — times.) This enables us to translate back and forth between an estimate of this hazard rate and an estimate of a ‘half-life’ or 50% success horizon — how ‘long’ (i.e. complex) a task needs to be before the agent fails more often than not — and also to extrapolate to ‘durations’ corresponding to other reliability levels, like 99% or 99.9% [11] . In this formulation, the hazard rate, , stands in for what fraction of our ‘subtask’ pool the agent can’t (yet) succeed at, which ends up being a reasonable summary of the agent’s competence in this domain. [12] This time, we’re looking at overall task success as if the agent has a 98% chance of meeting any particular subtask’s requirements. Sometimes a shorter task will happen to have one of the difficult subtasks — but usually they’re overall successful. As tasks get longer, there’s a greater chance that at least one subtask requirement is insurmountable at this reliability level. Among longer tasks, overall success becomes fewer and farther between. This agent can’t expect to often succeed on tasks longer than 50 or so subtasks. If you have a new task, you don’t know if the agent has all it needs to complete it. But the task ‘length’ is an indicator of how many tricky subtasks it has, and similar-lengthed tasks will have similar numbers of such subtasks — so their average success rate is a good estimate for how likely the agent is to succeed at this new task. Relating hazard rate with frontier AI development METR's graph is compelling because it suggests a steadily increasing frontier of success horizon as AI developers produce new agents over time. What does this imply if we interrogate our hazard rate model? Well, 'half-life' (and indeed various success-level horizons) is observed apparently growing exponentially with date : This is the central striking takeaway from the METR graph (modulo their measurement uncertainty). Half-life go up! But half-life according to our model has: where is the per-step hazard rate from before. When this is not too close to 1, that half-life is, fairly intuitively, approximately proportional to the reciprocal of the hazard rate: So METR's observation of rising time horizons is equivalent to saying that the frontier hazard rate is shrinking exponentially over time . Recall that this hazard rate corresponds with the fraction of ‘subtasks’ in a domain that an agent doesn’t yet know how to complete. So this fraction is presumed to shrink roughly exponentially with date, in turn driving the observed ‘longer’ success horizons. Why does hazard rate shrink with date? Here’s where to look for the next bit of mechanism. Why would the hazard rate, the fraction of ‘subtasks’ which remain out of reach, shrink in that way? It goes without saying that AI developers are chasing after increasing competence in their products, so (if they are doing anything at all right!) the direction of movement is unsurprising. Why that particular roughly-exponential form, though? I confess here I’m uncertain and the quest for more mechanism continues. My best guess is that it’s about the effective evidence available to the agent toward subtask solution strategy. Intuitively, if you’ve seen very similar subtasks many times before, it’s hard to go too wrong. If you’ve only seen vaguely similar subtasks once or twice, you’re in much less familiar territory and stand a good chance of stalling. Suggestively, effective evidence and training data are both information-like quantities, but I don’t want to make too much of that without a crisper connection. Formally, we could consider how many bits of evidence the agent can muster about how to proceed (either from past learning or by exploring in context ). In other words, training produces learnings . These range from broad, generally-applicable heuristics for adaptable, effective behaviour (experiment, test your work, notice when something surprising happens, read the manual if you can find one, accrue power and resources at any opportunity, ...), to narrow specific details about particular situations and activities (Earth's radius is roughly 6.4 megameters, detonating TNT yields roughly 4.2 kJ/g, humans succumb to oxygen deprivation after around 5 minutes, …). Ahem. Empirically, AI developers have historically poured something like exponentially increasing ‘quantities’ of ‘data’ into their machine learning pipelines. [13] Mathematically, that implies a power law : data inputs rising at one exponential rate, matched by hazard rate decaying at another exponential rate. Power laws aren’t deeply mechanically explanatory, but they’re often the best we have in machine learning, and are at least more predictable than mere date-based trends. Under the simple subtask model described here, this power law translates directly into a power law between ‘time horizon’ and data. This is actually the same level of explanatory improvement offered by Wright’s Law over Moore’s: not fully mechanistic, but an extra layer of detail which offers firmer purchase on what’s going on. What this doesn’t straightforwardly account for is the benefit to success rates of increased in-context reasoning , which is exhibited according to METR’s estimates. I expect this is operating on those borderline subtasks — where the agent would have some slim chance of satisfying them if it ‘rushed’. In those cases, ‘ thinking harder ’ may more effectively recall and combine the relevant learned knowledge, and allow better choices for exploratory discovery in situ . In any case, changing the thinking budget of an otherwise similar existing system certainly calls for a more mechanistic understanding than mere date-based trend extrapolation! I would be thrilled if someone with more smarts, time to experiment, and access to data were to dig into ways we could match up various AI production inputs (especially ‘data’ in various forms) with observed outputs like ‘time horizon’. One of the more difficult pieces might be quantifying ‘data’, especially teasing apart what types of evidence are ‘relevant’ for the domain and tasks at hand. Upshot The kind-of-boring upshot of this is that data and ‘practice’ on related tasks makes AI better at those tasks! This is boring because, well obviously!, we already basically knew that. But it’s encouraging because we can say a little more than that, which gives us some better grasp on what’s driving ‘time horizon’ progress in particular domains — and it can help get more precise about predictions. The fact that the ‘subtask’ model — with a ‘hazard rate’ of subtasks currently out of reach — is a fairly explanatory fit for capability profiles of individual agents is evidence that there’re not unusual amounts of generalisation capability in AI. As with humans, they can extrapolate a bit, but need ‘experience’ and examples to succeed. [14] Importantly, this means that vast in silico training ranges for software, cyber, and mathematics very likely won’t transfer much to other domains of interest , like interpersonal intelligence, medical discovery, bioweapons development, intelligence analysis, and robotic manipulation. Of course, like with every domain of human experience and activity, we have some relevantly-similar data already collected, and schemes can be devised to more rapidly expand that digitised experience bank for AI to learn from. Increasing adoption of AI in task-integrated contexts, industrial deployment, and even explicit approaches to gathering example data such as ‘ hand movement farming ’ are the leading indicators to watch for progress in particular domains — not just the headline benchmark metrics in software-like tasks. For some types of activity, developers are probably ‘running out’ of raw example data to scrape from the internet. The era of mostly-pretraining is over. For domains which can be relatively easily verified, like mathematics and coding, this is very surmountable — you can just run drills galore on a computer and get data that way. But this costs extra compute and doesn’t scale at the same exponential rate for long (perhaps 10x/year presently). As soon as this year, developers could be back to ‘only’ scaling compute around 4x per year (and a bit after that they might have bought most of the compute ! — and will only be able to scale at the positively sloth-like 1.5x-ish a year of underlying hardware progress ). I don’t feel confident extrapolating exactly where that cashes out, but if the data-driven subtask-learning model is right, it would imply we should see less steepness to the time horizon growth quite soon . [15] Some commentaries project that, once AI can autonomously do software and machine learning work reliably, it will thereafter enter a ‘recursive self-improvement’ phase and rapidly colonise all capabilities. I don’t think this is missing the point entirely: there will be modest multipliers on the speed of the AI development pipeline, and we might see an ‘explosion’ in the speed and cost-effectiveness of AI (because they are among the most immediately-verifiable properties to iterate on). But generalisation doesn’t come for free, so on-task data and compute will remain crucial to broadening the frontier of autonomous capabilities. Collecting that data and manufacturing that compute look to me like the rate-limiting steps, and therefore the major leading indicators to use in foresight. The best case I can make for a much more general explosion is if the speed and cost-effectiveness explosions rapidly accelerate the gathering and digestion of diverse task data — but I think that remains mostly rate-limited in the familiar ways: some domains easy and some more difficult. Don’t mistake me for ruling out across-the-board AI capability! Companies are charging ahead with data collection and set on automating much of their AI production pipeline. It just won’t happen overnight. Thanks to Coz Ududec for a conversation prompting me to think about this. ^ Produced by AI monitoring non-profit METR ^ Very importantly, it’s measured within a particular collection of challenges/tasks which are mostly associated with software development, especially ML engineering. METR also has a great preliminary study of some other domains, finding differing, but perhaps also somewhat predictable trends. ^ Moore’s Law is the very superficial observation that, over time, the number of transistors per chip doubles roughly every two years. (More recently, it’s been more clearly expressed as the price per transistor halving every year-or-two .) Wright’s Law is the slightly more mechanistic and general observation that production of many commodities follows ‘learning curves’, such that each doubling of cumulative production produces roughly similar relative cost savings. (We can in turn attempt to explain this in yet more mechanistic terms, pointing to the insight gained from observing and recording many trials and experiments, with suitably diminishing returns.) Now, if the quantity demanded and produced grows exponentially over time (as it has for computer chips), then Wright’s Law predicts comparable cost savings each year: Moore’s Law. If the quantity produced grows (or shrinks) in some other pattern over time, Wright’s Law, by accounting for this mechanistic detail, can often forecast cost trends more reliably than Moore’s. ^ Also note that the estimation of ‘task length’ according to human experts was quite crude (naturally, humans are the most expensive part of most experiments!), and there are good reasons to treat the reported error bars as much too narrow , i.e. misleadingly confident. I’ll use quotes around ‘time’ related quantities in this post as a reminder that it’s a loose estimate of a crudely human-performer-derived time-to-completion for tasks, and doesn’t correspond well to real time as such. ^ I don’t know if METR publishes how long the agents themselves take at these tasks — I don’t think so, and it’d arguably be ill-defined anyway since it would depend in part on how fast a computer you ran the agent on. ^ If we conceptually carve up subtasks into smaller pieces, they'll be quicker per piece, but there are commensurably more of them, and vice versa. ^ This could come apart if longer tasks are systematically more likely to include repetitive similar activities rather than a series of distinct ones, for example. Or longer tasks might tend to admit more truly alternative pathways. Both these effects could make longer tasks slightly easier than the naive picture. There are also higher-level ‘orchestration’ tasks i.e. coherently coming up with (and executing and adapting) an appropriate sequential plan: perhaps these might be systematically more difficult for longer tasks. ^ Notably, agents sometimes take a (relatively) longer time to do something that’s quicker for humans, and vice versa. ^ Incidentally, success (or not) here already accounts for the agent attempting and re-attempting steps or fixing earlier mistakes, which might take variable amounts of time: another reason not to treat this as agent time. Some subtasks might be intermediate and succeed sometimes (for example if the agent can’t easily choose the best approach but sometimes hits on the right one, or sometimes gets stuck in a terminal cycle but sometimes makes lucky progress.) ^ This is throwing away some detail: obviously not all subtasks are equally likely to follow from each other! There’s some correspondence between on-task sequences. But within a particular domain (like software engineering), this naive model of overall tasks combining subtasks somewhat randomly seems to do OK. ^ By the way, the rule of 72 provides a really quick mental approximation for the higher-reliability ‘time’ horizons, depending on the ‘half-life’ (the 50% ‘time’ horizon). Divide the ‘half-life’ by 72. That’s the 1% failure horizon (equivalent to the 99% success horizon). Multiply by your target failure rate in percent, and you’re done: that’s your target success ‘time’ horizon. E.g. if ‘half-life’ is 1h, the ‘time’ horizon at 99.9% is (1h/72)*(0.1) i.e. 5 seconds. (This also reveals that cutting the ‘time’ horizon tenfold cuts the average failure rate tenfold and so on.) Going the other way, estimating long-horizon success rates, divide your target horizon by the ‘half-life’. That’s how many halvings of success to expect: raise one half to that power for your success rate. E.g. if ‘half-life’ is 1h, your 24h success rate is i.e. one in sixteen million. ^ It didn’t have to be that way! A single number which manages to explain a lot of variation in agent capability is very suggestive of an underlying mechanism something like the ‘fraction of subtasks’ model I’ve described here. Of course there is still some residual uncertainty and there may be better summaries available with a more detailed model or epicycles on this one. ^ This may recently be trickier to measure as training pipelines have adapted to incorporate more reinforcement learning , which means these experience data are less ‘homogeneously slurped up from the internet’ and increasingly ‘proactively curated from in-domain training curricula’. So the mere quantity of data isn’t like-for-like over time. ^ In fact contemporary AI is perhaps substantially less good at generalisation than humans, though I’d like to be better informed about how factors like sample efficiency of AI learning (including in-context learning) stack up. ^ Actually saying something so bearish about AI makes me nervous, as there is a venerable history of people boldly declaring AI is about to hit a wall ! But I think it’s borne out. I’m not saying progress stops , I’m saying it probably gets slower (in exponential terms). Discuss
Score: 27🌐 MovesMay 25, 2026https://www.lesswrong.com/posts/zT76JcomKkdqo8tC6/a-slightly-mechanistic-theory-for-exponentially-increasing - HMD Vibe2 with Sarvam's Indus AI assistant to go on sale starting May 26
HMD Vibe2 5G goes on sale May 26 on Flipkart with Android 16, a 6000mAh battery, 50MP camera, and India's first Indus by Sarvam AI integration
- Opinion | Taste and AI: Why human judgment will always rule the winning workplaces
Opinion | Taste and AI: Why human judgment will always rule the winning workplaces Toronto Star
- BeautyPlus Upgrades Online AI Photo Enhancer with Batch Processing and Scene-Specific Models to Revolutionize Web-Based Editing
BeautyPlus Upgrades Online AI Photo Enhancer with Batch Processing and Scene-Specific Models to Revolutionize Web-Based Editing USA Today
- 7 AI gadgets of 2026 that actually feel useful: Smart glasses, AI rings and more
AI gadgets in 2026 are finally moving beyond hype, with smart glasses, AI rings, fitness trackers, and wearable assistants offering practical features. From Fitbit’s AI-powered wellness insights to Meta’s smart glasses and Samsung’s Galaxy Ring, these devices focus on convenience and productivity.
- Implementing Hybrid Semantic-Lexical Search in RAG
Implementing hybrid search strategies is a critical step in building modern RAG (Retrieval-Augmented Generation) systems , especially when shifting from prototype to production-ready solutions.
Score: 26🌐 MovesMay 25, 2026https://machinelearningmastery.com/implementing-hybrid-semantic-lexical-search-in-rag/ - School of Football | Can football teach a robot to move?
Teaching a robot to move using football
Score: 26🌐 MovesMay 25, 2026https://www.hyundaimotorgroup.com/en/tv/school-of-football-can-football-teach-a-robot-to-move - Why relationships are the hidden infrastructure of AI transformation
By now, most digital and tech leaders have heard some version of the same refrain, that AI adoption is 20% about the technology and 80% about the people. But even when leaders acknowledge the people side of AI, the conversation often moves quickly to skills, workforce planning, talent architecture, or reimagining how work gets done. Those are all important, but Breakthru Beverage Group EVP and CIO Glenn Remoreras believes it’s an incomplete picture. He argues that the real differentiator in AI transformation is something more human: the ability to build trust, deepen relationships, and lead through connection when uncertainty is high. “As AI accelerates change and organizations transform at unprecedented speed, the real differentiator for leaders will no longer be how fast they move, but how deeply they connect,” he says. For him, relationships aren’t soft skills or leadership accessories but the infrastructure that allows transformation to happen. When trust is strong, organizations align faster, challenge assumptions more productively, and move through uncertainty with greater confidence. When trust is weak, even the best technology strategy can stall . Closing the IT and business chasm Remoreras’s views are rooted in a long-standing conviction about the role of technology in the enterprise. Over the years, he says he became almost maniacal about solving the persistent chasm between IT and the business . In many organizations, business functions still view IT primarily as a service provider, and too often, tech leaders reinforce that mindset by referring to business stakeholders as customers. The language may seem harmless, but Remoreras believes it reflects a deeper problem. In that model, the business sets strategy while IT receives it, translates it, and aligns execution. The relationship is reactive, and the trust gap remains unresolved. “To me, that paradigm is reactive,” he says, “and it exists largely because the trust gap has never been fully resolved.” The alternative is convergence, where business stakeholders aren’t customers but equal partners. CIOs don’t simply respond to strategy, they help shape it, bringing a clear understanding of how technology capabilities, data, platforms, and AI can change what’s possible. “When trust exists, a different model emerges,” Remoreras says. “Technology strategy becomes inseparable from business strategy.” That shift matters even more in the age of AI because AI isn’t just another technology deployment , it changes decision-making, work design, customer experience, risk, governance, and the way employees understand their own roles. If technology and business leaders approach AI from opposite sides of the table, adoption will be slower, riskier, and less likely to produce meaningful value. Trust as an operating system For Remoreras, trust isn’t a vague aspiration but a core operating infrastructure, and relationships are the primary enabler of that trust. That framing matters because AI introduces uncertainty at multiple levels. Employees may wonder how their work will change, leaders may struggle to separate hype from real opportunity, and functions may disagree about ownership, risk, funding, or pace. Without trust, those tensions can harden into resistance. But with trust, the same tensions can become productive. Remoreras’s thinking has been shaped in part by the Business Relationship Management discipline. In 2014, he joined the BRM Institute as one of its early founding members and helped contribute to the development of its first body of knowledge. But he’s never viewed BRM as only a role, but rather a capability and leadership philosophy. “It shouldn’t be confined to a handful of dedicated roles,” he says. “It should be a competency shared by leaders across both technology and business functions.” That distinction is critical. In many companies, relationship management is delegated to specific people or roles. Remoreras sees it as a leadership muscle that must be built across the enterprise, particularly as AI creates new dependencies across functions. The CIO, in this context, isn’t only a technology strategist or transformation leader but a trust builder, translator, challenger, and convener. The PATH to relationship leadership To make the idea more concrete, Remoreras developed a framework he calls PATH (Purpose, Agility, Trust, and Humanity), and introduced it in his BRMConnect keynote as a way to describe the leadership capabilities organizations will need in an AI-enabled world. Purpose comes first because transformation requires more than activity . People need to understand why the work matters. “When people sense you’re driven by purpose, that what you’re doing truly matters, they lean in,” he says. Purpose creates alignment and gives people a shared reason to move through uncertainty together. Agility is the ability to lead when the path is still forming. For Remoreras, leadership agility is about setting the rhythm, empowering others, and adjusting as the tempo shifts, not controlling every note. Trust is the currency of leadership in the age of AI, and the safety net that allows people to explore boldly, he says. Without it, experimentation slows, transparency fades, and people retreat into self-protection. Humanity is the final pillar. When employees sense that leaders value technology more than people, they disengage. But when innovation is anchored in ethics, empathy, and fairness, people are more likely to follow with confidence. “Humanity ensures that progress benefits everyone, not just a few,” Remoreras says. Making AI adoption human The PATH framework reframes the AI adoption challenge and moves the discussion beyond tools, training, and process redesign without diminishing the importance of any of them. Skills, governance, and use case prioritization all matter. But none will be enough if people don’t trust the leaders setting direction, the teams building solutions, or the organization’s intent for how AI will be used. That’s why Remoreras’s message is particularly relevant for CIOs. Tech leaders are often accountable for the platforms and capabilities that make AI possible, but they’re also uniquely positioned to shape the relationships that make AI scalable. They can help the organization move from IT as service provider to technology as co-leader, build the connective tissue between strategy, execution, risk, and adoption, and create the conditions where business and tech leaders share ownership for outcomes. This is the leadership work AI requires, says Remoreras. As machines become more capable, the leadership capabilities that can’t be automated become more valuable. The ability to listen, align, challenge, empathize, build trust, and lead through ambiguity becomes a strategic advantage. The future of AI won’t be determined only by which organizations choose the right platforms or move the fastest, but it’ll also be shaped by which organizations can create enough trust for people to move together. That’s why relationships may be the hidden infrastructure of AI transformation. Not because they replace technology, but because they make transformation possible.
Score: 26🌐 MovesMay 25, 2026https://www.cio.com/article/4172478/why-relationships-are-the-hidden-infrastructure-of-ai-transformation.html - ET Most Innovative AI Product Awards 2026: Recognise leaders in Agentic AI & Autonomous Systems
The ET Most Innovative AI Product Awards 2026 by The Economic Times is now inviting nominations for the Agentic AI & Autonomous Systems category, recognising scalable AI innovations driving autonomous decision-making, intelligent workflows, and the future of enterprise technology.
- Lidwave Appoints Eyal Assa as Chief Business Officer to Lead Global Expansion as Physical AI Reshapes the Real World
Lidwave Appoints Eyal Assa as Chief Business Officer to Lead Global Expansion as Physical AI Reshapes the Real World USA Today
- Google clarifies its slightly confusing pair of AI Ultra plans
Following the introduction of a more affordable new tier, Google One’s AI Ultra plans have been a tiny bit confusing for those looking to switch, but Google is updating the process to make things a bit more clear. more…
- Harness, Scaffold, and the AI Agent Terms Worth Getting Right
Harness, Scaffold, and the AI Agent Terms Worth Getting Right
- Putting AI to work in network management
Last year, analyst Forrester reported that while IT departments manage billion-dollar portfolios, their internal operations lag in automation, coordination and visibility. The complexity of managing a modern IT architecture means network management must evolve. This is not something that is entirely new. Automation is part of the functionality available in modern network management tools. Big data analysis of network log files is used in security information and event management (SIEM) , and machine learning (ML) is helping network administrators identify potential issues before they affect the business. Phil Huang, business development and field application manager at D-Link, explains: “We have been offering a pure cloud management platform for networks for a number of years and the AI [artificial intelligence] assistance behind such network management gives us the ability to monitor in real time and also proactively try to alert of any potential problems that may arise.” Advances in tooling potentially reduce the complexity of network management. Matt Stava , CEO and chairman of third-party support firm Spinnaker Support, says this changes the role of IT administrators and programmers. Looking specifically at network skills, he says: “The need for a Cisco-certified expert is getting less and less right now.” Modern networking skills Modern IT infrastructure means that having an industry-certified network specialist is becoming less relevant. In a March 2026 blog post , Amit Katz, vice-president of ethernet switch at Nvidia, highlights the shifts occurring in network management. In the post, Katz points out that while the value of a new network administrator may have previously been measured by their level of expertise in a particular networking command line interface (CLI), the advent of hybrid cloud and DevOps means there is a growing shift towards application programming interfaces (APIs). “Skills in Ansible, Salt [the open source automation framework] and Python now have more value than a Cisco certification,” he says. Now, Katz believes the tasks network administrators need to do are very different from the way they used to monitor and manage networks. Skills in Ansible, Salt and Python now have more value than a Cisco certification Amit Katz, Nvidia “You’ve moved from tools that polled devices across the datacentre using SNMP [Simple Network Management Protocol] and NetFlow [which monitors IP traffic] to new switch-based telemetry models where the switches proactively stream flow-based diagnostic details,” he notes in the blog post. And according to Katz, while network administrators have a lot of experience introducing new workloads into datacentres – some of which have unique networking requirements – building an AI cluster is actually very different. He writes: “It is tempting to think that AI is just a bigger and faster big data application. But AI is different, and AI can be hard without the right tools.” AI also has a role to play in helping network administrators manage this complexity more easily. Information Services Group (ISG), a research and advisory firm, says organisations are taking advantage of the enhanced capabilities of AI and ML to automate configuration changes and optimisation across the network. In an ISG article about how AI is transforming network operations, Marc Herren, a director at ISG, notes that AI can analyse network data and identify patterns to automatically generate configurations that optimise performance. He says Cisco and Juniper Networks, the latter now being part of Hewlett Packard Enterprise, are developing intent-based networking products that use AI to understand an administrator’s intent and automatically configure the network accordingly. Such technology is essential to keep on top of ever-more-complex network management. Network complexity In a presentation at Microsoft Build 2025, Phil Gervasi, director of technical evangelism at Kentik, spoke about how networks are growing in complexity. They now span different clouds, datacentres, edge computing and hybrid IT infrastructure, all of which introduce new challenges for network management. “The volume of telemetry, events and logs has exploded beyond human capacity to analyse in real time,” he told attendees. At the same time, as Gervasi noted, network teams are under pressure to improve the mean time to resolution of an issue, and maintain uptime without expanding headcount. The volume of telemetry, events and logs has exploded beyond human capacity to analyse in real time Phil Gervasi, Kentik “What AI offers is not magic, but a better way to correlate data, forecast performance and understand network behaviour in context. So, in short, AI helps operators move from reacting to predicting,” he added. While ML is being used in networking for capacity planning, anomaly detection and baselining, Gervasi said that large language models (LLMs) offer a different approach to network management. “Unlike classical data models, which rely on structured data, LLMs operate on unstructured information like documentation, configuration files and tickets,” he told Build 2025 delegates. However, LLMs are probabilistic, which means they can produce inconsistent and different answers to the same prompts. They also hallucinate . To get around these limitations, Gervasi stressed the need to ensure quality of training data, proper evaluation and controlled model behaviour. These are key to keeping LLM responses honest. Privacy and regulation are also issues for LLMs, especially when handling network data that could contain sensitive information. Some IT operations challenges are inherent to AI use. For Gervasi, IT decision-makers need to be aware of the difficulties that may arise when integrating real-time telemetry, dealing with diverse data types, and managing compute costs for AI workloads. But, despite these caveats, Gervasi believes the real power of LLMs lies in their ability to synthesise vast volumes of data into information that can then be used by people to make better decisions. Among the examples he provided during his Build 2025 talk was incident triage and summarisation. “Instead of sifting through hundreds of alerts, an AI system can turn that noise into a single incident summary, highlighting probable root cause, and even suggesting next steps,” Gervasi said. Getting started with AI in network management The starting point in using AI for network management is collecting network telemetry logs, helpdesk ticket and configuration files. Those then need to be cleaned up and stored in a format that can be accessed by the AI system. Gervasi told delegates that one of the most effective ways to use this information is through retrieval augmented generation (RAG). As an example, he said when a user submits a query, the system converts the question into a mathematical representation, which searches a vector database for semantically related data, such as telemetry, past incidents or documentation. “The LLM then synthesises an answer, using both its general knowledge and the retrieved context,” he explained. Another use for LLMs is in text-to-structured query language (SQL), which , as Gervasi noted, enables network engineers to use natural language, where their queries are converted by the LLM into an SQL query and then, where relevant, provide a graphical representation of the data. Once the data is in a format the AI model can process, agentic AI is a natural progression. “An LLM doesn't just respond to prompts, but acts kind of like the brain, coordinating multiple tools,” he says. During the presentation, Gervasi spoke about how with agentic AI powering network management, an agent could run a trace route, collect network telemetry, consult a knowledge base, and then generate a remediation plan, all autonomously, but with human oversight. This is something that is likely to provide autonomous operations behind commercial network provider services. Analyst Gartner expects that AI will be embedded into managed network services (MNS) by 2028, to increase and enhance operational efficiency and enable more informed decision-making. According to Gartner, AI will be used to ensure that networks are robust and agile enough to adapt to changing demands and traffic patterns. “Looking ahead three to five years from now, we anticipate significant transformation in MNS due to extensive use of AI and automation,” the analyst firm stated in its AI will transform managed network services in the next three years report. For Stava and other industry watchers , the hot skill is agentic AI and the ability to integrate AI agents into workflows to achieve a business outcome. And these outcomes are increasingly IT-focused, especially as IT teams are being asked to do more with fewer resources and being put under increasing strain to support companies’ appetites for all things relating to AI. But AI also has a big role to play in making networks more manageable. As network management becomes more automated and networks become self-healing, network engineers will need to learn how to integrate the latest tooling with agentic technology to provide the data stream for AI-powered network management. Read more about AI in network management Agentic AI ushers in a new era of network management : Agentic AI will redefine network management. But IT execs have to understand the technology’s benefits – and its drawbacks. AI-driven self-healing networks bring new capabilities: Self-healing networks use AI to continuously monitor, diagnose and fix issues autonomously, shifting IT from reactive troubleshooting to proactive management.
Score: 25🌐 MovesMay 25, 2026https://www.computerweekly.com/feature/Putting-AI-to-work-in-network-management - iOS 27 may bring Google Cast, AI Siri, Camera app overhaul: What's coming
iOS 27 may bring native Google Cast support, AI-powered Siri upgrades, a redesigned Camera app, and new AI editing tools for Photos alongside watchOS 27 features
- What Gemini features you get with Google AI Plus, Pro, & Ultra [May 2026]
To get more out of Gemini-powered features in first-party apps, Google offers AI Plus, AI Pro, and AI Ultra subscriptions. At I/O 2026 in May, the plans got updated usage limits. more…
Score: 25🌐 MovesMay 25, 2026https://9to5google.com/2026/05/25/google-ai-plus-pro-ultra-gemini-features/ - Teen launches AI platform in Dubai to address brokerage information gaps
Teen launches AI platform in Dubai to address brokerage information gaps
Score: 24🌐 MovesMay 25, 2026https://www.khaleejtimes.com/business/teen-launches-ai-platform-in-dubai-to-address-brokerage-information-gaps - Team Sakura, including Adam Nohejl of the Explainable AI Team, won the Open Track of the BEA 2026 Shared Task（May, 2026)
Team Sakura, including Adam Nohejl, Postdoctoral Researcher, and Hitomi Yanaka, Team Director of the Explainable AI Team, won the Open Track of the BEA 2026 Shared Task “Vocabulary Difficulty Prediction for English Learners,” held in conjunction w
- Route Claude Code Through MLflow AI Gateway
Learn how to route Claude Code through MLflow AI Gateway to get full observability, budget controls, and guardrails across all your coding agent sessions, with no changes to how you use Claude Code.
- Here are all the major products Google announced at I/O 2026
Google used I/O 2026 to unveil agentic AI systems, new Gemini models, autonomous assistants, upgraded AI search, creative tools, AI glasses, and a $100/month Ultra plan targeting developers and power users. The post Here are all the major products Google announced at I/O 2026 appeared first on MEDIANAMA .
Score: 22🌐 MovesMay 25, 2026https://www.medianama.com/2026/05/223-major-products-google-announced-i-o-2026/ - Rank Contractors Launches AI Marketing Platform for Home Service Contractors
Rank Contractors Launches AI Marketing Platform for Home Service Contractors azcentral.com and The Arizona Republic
- Copilot is getting a sidebar treatment for Windows 11, just like Gemini in Chrome
Copilot is getting a dockable sidebar in Windows 11 that works similarly to Gemini in Chrome, an approach that Microsoft already tried and scrapped.
- VT Markets Adds 39 US Stocks and ETFs Spanning AI, Space, and Energy
VT Markets Adds 39 US Stocks and ETFs Spanning AI, Space, and Energy The Straits Times
- Applications open for 2nd MEST AI Startup Programme
The Meltwater Entrepreneurial School of Technology (MEST), a leader in tech entrepreneurship training and early-to-growth-stage startup support, has opened applications for the second edition of the MEST AI Startup Programme. MEST, a pan-African software and entrepreneurship training programme, seed fund, and incubator, has for the last 18 years trained and supported software entrepreneurs across the [...] The post Applications open for 2nd MEST AI Startup Programme appeared first on Disrupt Africa .
- The one-man $250M startup
Learn about a one-man startup that reached $250M valuation