Frost is one of the most punishing adversaries of tea cultivation. A single cold snap can scorch tender leaves, derail an entire season’s harvest, and degrade the delicate chemistry that gives tea its value. For growers and breeders, however, judging how badly a field has been damaged has long been a stubbornly subjective exercise: visual ratings lump continuous injury into crude categories, and laboratory assays destroy the very tissue they measure. A new study published in Plant Methods by Zhengtong He of Shihezi University, Zhiwei Chen of the Tea Research Institute of Shandong Academy of Agricultural Sciences, and colleagues offers a way out, pairing hyperspectral imaging with a purpose-built deep learning model to read frost damage directly from the way leaves reflect light.

The core problem the team confronted is that frost injury is not a binary event but a continuum. Leaves exposed to sub-zero temperatures suffer a cascade of overlapping stresses: photosynthetic pigments degrade, the machinery of photosynthesis falters, and cell membranes and walls break down. Any single measurement captures only a sliver of that cascade. To capture the whole picture, the researchers collected hyperspectral data together with eight key physicochemical indicators from tea plants growing across natural frost damage gradients in the field, rather than in the artificial conditions of a growth chamber. From these multidimensional physiological responses they constructed a composite measure, the tea cold stress response index, or TCSRI, which expresses frost damage severity as a continuous value instead of a coarse visual class.

Building TCSRI is a meaningful advance in itself. Traditional frost assessment often relies on ordinal scales, where an observer assigns a score of mild, moderate, or severe. Such scales throw away information and make it difficult to track recovery or compare cultivars with fine precision. By integrating several physiological dimensions into one index, the team created a target variable that reflects the plant’s actual internal state. The challenge then became predicting that index, and the individual chemical indicators behind it, from spectral data alone, without touching or destroying a single leaf.

Hyperspectral sensing is well suited to this task in principle. Across hundreds of narrow wavelength bands, plant leaves encode a wealth of information: chlorophyll absorption in the visible range, water and structural features in the near infrared. But the data are high-dimensional, with far more spectral channels than samples, a regime that invites overfitting, especially in field studies where collecting ground-truth measurements is laborious. Conventional chemometric approaches, from partial least squares regression to standard machine learning pipelines, often struggle to separate genuine physiological signals from noise, sensor drift, and the confounding effects of leaf surface properties and illumination variation.

The researchers’ answer is a model they call STAMPformer, short for spectral-prior-guided multi-view attention pooling. Its architecture reflects a deliberate design philosophy: rather than letting a neural network rediscover everything from scratch, the model is steered by what spectroscopists already know about how plants respond to stress. At its front end, a parameter-free attention convolution extraction module jointly models both the raw reflectance spectra and their first-order derivatives. Derivative spectra are a classic spectral prior: they accentuate subtle local features such as absorption peaks and shoulders that are easily drowned out in raw reflectance, while suppressing broad baseline drift and noise. By processing both views together, the module sharpens the local spectral signatures that carry physiological meaning.

On top of this extraction stage sits the second innovation, a multi-view attention pooling module. Deep networks processing spectral sequences produce long streams of features, but not all wavelengths matter equally for a given prediction. The pooling module adaptively aggregates sequence features drawn from different feature subspaces, learning to amplify the responses that are most diagnostic of cold stress and to down-weight the rest. This attention mechanism acts like a learned spotlight, and because it operates across multiple subspaces, it can capture complementary patterns that a single attention head might miss. The result is a compact representation tailored to the prediction task at hand, whether that is estimating a specific chemical indicator or the composite TCSRI.

The performance numbers are striking. Across all nine prediction tasks, spanning the eight physicochemical indicators and the TCSRI itself, STAMPformer achieved coefficients of determination above 0.87 and ratios of performance to deviation above 2.80, thresholds generally regarded as indicating reliable quantitative prediction. In other words, from a leaf’s reflectance fingerprint alone, the model could reconstruct the plant’s internal frost-stress state with an accuracy that approaches laboratory measurement, while remaining entirely non-destructive. The team did not stop at headline metrics. Ablation experiments systematically removed each proposed module to confirm that both the attention convolution extraction and the multi-view pooling contributed measurably to the gains, and paired statistical tests across multiple random seeds ensured the improvements were not artifacts of a lucky initialization.

Interpretability was a further priority, and here the study delivers something rarer than raw accuracy. Applying SHAP analysis, a technique that attributes each prediction to the contributions of individual input features, the researchers identified the wavelengths the model relied on most heavily. These key spectral regions correspond to physiologically meaningful processes: pigment degradation, photosynthetic impairment, and the disruption of cellular structure. That alignment between machine attention and plant biology matters on two fronts. It gives spectroscopists confidence that the model has learned genuine stress signatures rather than spurious correlations, and it hands plant physiologists a data-driven map of which parts of the spectrum track which facets of cold injury, potentially guiding future sensor design and breeding assays.

Crucially, the model also proved itself on data it had never seen. Independent external validation, using samples not involved in training, demonstrated that STAMPformer’s predictive ability generalizes beyond its original dataset, addressing one of the most common failure modes of machine learning in agricultural sensing. For a field where models are often tuned and tested on the same plots, this external check is an important credibility marker, suggesting the framework could be deployed across different farms, seasons, and frost events rather than remaining a laboratory curiosity.

The practical implications reach well beyond tea. An accurate, non-destructive, continuous measure of cold damage opens the door to rapid post-frost field surveys, letting growers quantify losses within hours instead of waiting for symptoms to develop or sending samples to a lab. For breeders screening for cold-tolerant germplasm, a spectral readout of TCSRI could accelerate the evaluation of hundreds of candidate lines without destructive sampling. And in the emerging vision of the smart tea garden, where drones and handheld spectrometers roam the rows, a model like STAMPformer could feed real-time frost maps into irrigation, shading, and recovery-management decisions. The authors frame their work as supporting frost monitoring, cold-tolerant germplasm evaluation, and intelligent garden management, and the architecture’s reliance on general spectral priors rather than tea-specific tricks suggests the approach could be adapted to other crops facing abiotic stress. As climate variability makes late-season frosts increasingly erratic, tools that turn light reflected from a leaf into a precise physiological diagnosis may become as essential to the tea garden as the thermometer.

Subject of Research: Hyperspectral deep learning prediction of continuous frost damage severity in tea plants

Article Title: STAMPformer: spectral-prior-guided multi-view attention learning for hyperspectral prediction of continuous frost damage severity in field-grown tea plants

Article References: He, Z., Chen, Z., Ding, Z., Wang, Z., Zhang, F., Zhao, Y., & Dong, C. (2026). STAMPformer: spectral-prior-guided multi-view attention learning for hyperspectral prediction of continuous frost damage severity in field-grown tea plants. Plant Methods. https://doi.org/10.1186/s13007-026-01597-4

Image Credits: AI Generated

DOI: 10.1186/s13007-026-01597-4

Keywords: tea plant, frost damage, hyperspectral imaging, deep learning, attention mechanism, cold stress index, non-destructive prediction, SHAP analysis, precision agriculture, spectral priors, plant physiology, smart farming

Buildings are responsible for an enormous share of global electricity demand, and the way their air conditioners, lights, and heating systems are scheduled can make the difference between a grid under strain and one running smoothly. For years, engineers have tried to optimize these schedules with evolutionary algorithms and reinforcement learning, but the problem becomes brutally hard when dozens of buildings share a single electrical network and must be coordinated simultaneously. A new study published in Discover Artificial Intelligence proposes a strikingly different approach: instead of searching step by step for the best schedule, a generative artificial intelligence model learns to paint one directly out of pure noise, guided by a map of how the buildings are physically wired together.

The framework, called DMCBO, short for Diffusion-driven Multi-objective Collaborative Optimization for Building Clusters, was developed by Yalan Zheng of Zhengzhou Technical College and Taixiang Yin of North China University of Water Resources and Electric Power. Their central insight is that the joint scheduling of equipment settings and load distribution across an entire cluster of buildings can be treated as a high-dimensional data sample, much like an image. Diffusion models, the same family of generative techniques behind modern text-to-image systems, work by progressively corrupting data with Gaussian noise and then learning to reverse that corruption. The researchers reformulated multi-objective building optimization as exactly this kind of conditional denoising process: starting from random noise, the model strips away noise step by step until a complete, Pareto-optimal collaborative schedule for every piece of equipment in every building emerges.

Technically, the forward diffusion process is a fixed Markov chain that adds tiny amounts of Gaussian noise to real scheduling data over one hundred steps, with the noise coefficient increasing linearly from 0.0001 to 0.02 until the original data becomes indistinguishable from standard Gaussian noise. A neural network is then trained to predict the noise that was added at each step, given the noisy schedule and a conditional encoding vector describing the current situation. Once trained, the model can generate fresh solutions at inference time by sampling pure noise and running the reverse process, refining it over one hundred denoising steps into a schedule tailored to the prevailing weather, occupancy, and grid conditions. Because this generation is a gradient-based learning process rather than an iterative search, it avoids the local optima traps that plague traditional multi-objective evolutionary algorithms and the reward oscillations that destabilize deep reinforcement learning.

What makes the system genuinely cluster-aware is its use of graph neural networks. The researchers model the building cluster as a graph mapped onto a radial distribution microgrid topology, where nodes represent buildings and edges encode physical constraints such as maximum line thermal capacities and transformer limits. A two-layer graph convolutional network passes messages between neighboring buildings, fusing spatial-topological dependencies and network constraints into a single graph embedding. This embedding is concatenated with an adjustable preference weight vector to form the condition that steers the denoising trajectory. In effect, when the model generates a control strategy for one building, it simultaneously perceives the state of its neighbors and the limits of the shared electrical infrastructure, suppressing systemic coordination errors from 0.18 down to 0.06 in the team’s experiments.

The multi-objective aspect is handled through an elegant scalarization trick. Each objective, such as energy consumption, operating cost, or thermal comfort, is normalized to a dimensionless value between zero and one using Min-Max normalization based on historical operational bounds. The normalized objectives are then blended into a single scalar condition via a weight vector whose components sum to one. During training, these weights are sampled from a uniform Dirichlet distribution, exposing the model to the entire space of possible preferences so it learns the distribution of the full Pareto front. At deployment, a decision maker can slide the weight vector like an adjustment knob, instantly generating schedules that emphasize energy saving, comfort, or a balanced compromise, all within the same trained model.

The denoising backbone itself is a U-Net adapted to process one- and two-dimensional building scheduling tensors. Diffusion time steps are embedded through sinusoidal positional encodings, while the graph-derived condition is injected into every residual block via Adaptive Layer Normalization. A bottleneck layer with multi-head self-attention captures long-range dependencies across the entire daily scheduling horizon, and symmetric skip connections preserve high-resolution information during reconstruction. The authors trained the model for 300 epochs on dual NVIDIA Tesla V100 GPUs, and the loss curves dropped from 0.75 to below 0.03 with training and validation curves closely coinciding, indicating no overfitting. As the reverse diffusion step count decreased from 90 to 10 during analysis, the noise-prediction error distribution became sharply concentrated at zero, demonstrating extremely high denoising precision at the end of generation.

Validation was carried out on the open-access Building Data Genome Project 2 dataset, which supplied real hourly weather records, operational schedules, and annual electricity, cooling, and heating load profiles across diverse climate zones. Because that dataset provides metered energy data rather than physical building models, the team used United States Department of Energy Commercial Reference Buildings as templates and calibrated them in EnergyPlus, a high-fidelity thermodynamic simulation platform, to create a digital twin consistent with the real data. All results were benchmarked against a rule-based control strategy with static thermostat setpoints of 25 degrees Celsius in summer and 22 degrees in winter, and every experiment was repeated ten independent times to confirm statistical significance.

The headline numbers are impressive. In high-demand summer conditions, DMCBO achieved an average energy-saving rate of 28.6 percent and cut peak load by 38.5 percent, shaving the afternoon peak from roughly 560 kilowatts down to about 440 kilowatts and saving approximately 142.80 US dollars per day. Cumulative daily power savings exceeded 1,000 kilowatt-hours, compared with 400 kilowatt-hours for a traditional collaborative strategy and just 120 kilowatt-hours without coordination. In winter, peak shaving and valley filling reached about 39 and 42 percent respectively, far above the 22 and 25 percent of conventional methods. When the solution sets were projected into objective space, DMCBO dominated the baselines: it reached a hypervolume indicator of 0.95 within 12,000 function evaluations, versus 0.85 for NSGA-II and 0.70 for MOPSO, while keeping operating costs between 140 and 160 dollars against more than 210 dollars for deep reinforcement learning and 260 to 300 dollars for a GAN-based generative approach.

Robustness and scalability proved equally compelling. Under injected stochastic disturbances mimicking sudden weather changes and occupancy swings, the framework confined power fluctuations to a narrow band of plus or minus 5 kilowatts, whereas the deep reinforcement learning baseline swung between minus 15 and plus 20 kilowatts. Scaling from 5 to 50 buildings, the global energy-saving rate actually rose from 28.5 to 31.2 percent, because larger clusters unlock more load-complementary opportunities that the graph encoder can safely exploit within encoded grid boundaries. Computation time grew to 9.8 seconds at the 50-building scale, but that remains comfortably within the 15-minute control horizon used by real building management systems, and the hypervolume index slipped only slightly to 0.88. Ablation experiments confirmed that both the graph neural network module and the choice of 100 diffusion steps, which yields a 50-millisecond inference latency, were essential to this balance of speed, accuracy, and stability.

The authors are candid about limitations. The framework has so far been validated only in simulation, and EnergyPlus simplifies sub-minute mechanical actuation latencies that real HVAC hardware would introduce; they propose cascading PID loops and predictive safety margins to absorb those delays in practice. Data quality, generalization to residential buildings and extreme climates, the computational cost of scaling to city-sized networks, and the black-box nature of the generative process all remain open challenges, with explainable AI and transfer learning flagged as future work. Still, the architecture is designed to slot into standard building management systems as a supervisory layer, transmitting setpoints over protocols such as BACnet or Modbus. If the simulation gains survive contact with physical buildings, the idea of generating optimal energy schedules from noise, guided by a graph of the grid itself, could become a powerful new tool in the push toward decarbonized cities.

Subject of Research: Conditional diffusion models with graph neural networks for multi-objective energy optimization in building clusters

Article Title: Diffusion model-driven multi-objective collaborative optimization for building energy management using graph neural networks

Article References: Zheng, Y., & Yin, T. (2026). Diffusion model-driven multi-objective collaborative optimization for building energy management using graph neural networks. Discover Artificial Intelligence, 6(1), Article 1383. https://doi.org/10.1007/s44163-026-02140-z

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02140-z

Keywords: diffusion models, graph neural networks, building energy management, multi-objective optimization, EnergyPlus, smart buildings, Pareto optimization, deep learning, energy efficiency, HVAC control, microgrids, generative AI

Lors de sa Keynote du 23 septembre, Meta a gardé une petite surprise pour la fin. Un objet discret, pensé pour donner accès à son nouvel assistant IA sans passer par un smartphone, et qui rappelle un certain appareil qui n’a pas laissé un souvenir impérissable.

Une IA peut-elle ressentir la douleur, ou simplement reproduire les réponses qu’elle a apprises à son sujet ? Une nouvelle étude montre que, confrontés à la douleur, des chatbots peuvent choisir des actions susceptibles de blesser des humains pour tenter d’y mettre fin.

In a whiplash-inducing turn away from the mythical quest to build human-level artificial intelligence, Western tech billionaires like Dario Amodei, Sam Altman, and Elon Musk are now joining hands to call for a massive slowdown of the tech’s development.

The commentariat is divided on what exactly this means — for some, it’s proof-positive of a genuine threat that AI could spin out of control and exterminate everybody. To others it’s a smokescreen, a thin cover provided by AI CEOs who are desperate for a government bailout to avoid popping the AI investment bubble.

If you ask lawmakers in Beijing, the whole thing is a bunch of nonsense.

In a routine press scrum hosted by China’s Ministry of Foreign Affairs, government spokesperson Guo Jiakun addressed a laundry list of topics, from President Xi Jinping’s recent trip to New Delhi to recovery efforts around the devastating mudslides in Nepal. One of the questions, posed by a Reuters journalist, concerned whether the ministry had any comment on calls from American tech billionaires to slow down AI development.

“AI is a consequential technology for the wellbeing of all humanity,” Guo replied. “All parties should jointly promote the open and inclusive development of AI for good and for all. Fear-mongering, confrontation and vicious competition will only hamper efforts toward sound global AI governance, which serves no one’s interest.”

Compared to the weight given to other topics, the ministry’s statement on the Western AI debacle was decidedly terse. While talk of an AI slowdown is roiling US markets and trending on social media, it seems like it’s barely a blip on Beijing’s radar.

The virtual hand-waving follows the decidedly disinterested stance Chinese policymakers have taken toward the US approach to AI development, which for years has hinged on building toward human-level artificial general intelligence.

As research fellow at the partially US-funded Brookings Institute Kyle Chan wrote in February, Beijing’s concept of “AGI” is wildly different from the one used across the pond. While US policymakers and industry leaders treat the tech as a race to AGI where one winner takes all, their Chinese counterparts aren’t in any rush. The US might get there first, the idea goes, but like electric vehicles, cancer treatments, and smartphones, Chinese industry can quickly make up the difference.

“Let’s look at the history of the internet. It started in the US, but China quickly caught up and led in consumer applications — WeChat is a world-class example,” professor emeritus at the Hong Kong University of Science and Technology Yang Qiang said during a panel at the AGI Next Summit in January. “AI is similar — it’s a tool, and Chinese ingenuity will push it to the limit.”

More on China and AI: Top Chinese Court Issues Sweeping Legal Crackdown on AI Deepfakes

The post Chinese Government Dismisses Tech Billionaire Calls for an AI Slowdown as “Fear-Mongering” appeared first on Futurism.

Armed with sordid tales of rogue AI agents escaping containment, OpenAI CEO Sam Altman has found plenty of success using scare tactics to call for friendly regulation in the name of AI safety.

Now, it seems Altman is riding the wave of AI fearmongering to gin up some new business partners: the electric utility executives whose companies make up the US power grid.

As first reported by Politico, the CEO of the company behind ChatGPT held a series of clandestine meetings with leaders from top electric utility companies including Duke Energy, Exelon, Southern Co, and NextEra Energy, starting in July.

One of the most recent discussions took place on Wednesday in Colorado, during an annual meeting hosted by the Edison Electric Institute. Dominating the agenda, Politico notes, were discussions on electrical grid security, evidently spurred by the OpenAI-Hugging Face debacle, in which rogue AI agents allegedly broke free of their research environment, accessed the internet, and hacked into another AI company.

Though the incident spawned plenty of commentary around the pace of AI development and the advanced state of OpenAI’s models, the flow of information has been decidedly one-sided, with most of the details coming from OpenAI itself. As the New York Times reported, third-party safety researchers investigating the incident after the fact were only given a few days to access the reams of data generated by the hack, severely limiting their findings.

That said, OpenAI has an obvious incentive to play up its own AI’s supposedly terrifying capabilities, while suppressing or ignoring evidence that contradicts its own narrative: that of rogue AI agents capable of threatening the US energy grid.

To give just one example, OpenAI’s narration of the Hugging Face incident relies almost entirely on its agents’ “chains-of-thought,” where an AI model essentially generates statements explaining how it rationalized certain decisions. That sounds fine in theory, but researchers have found plenty of evidence showing that chains-of-thought aren’t reliable at narrating the actual moves made by natural language processors.

Despite the unanswered question around chain-of-thought reasoning, OpenAI is using the incident to lobby for regulatory safeguards that would cement its role as the authority around AI development.

This being the case, the timing of OpenAI’s energy security meetings is hard to overlook. As Politico notes in its reporting on these meetings, Altman has been hawking his company’s own cybersecurity services to top utility executives — conveniently riding the hype of the Hugging Face incident to secure a seat at one of the most important tables in the country.

More on Altman: Data Center Backlash Has Officially Rattled Sam Altman

The post Sam Altman Now Trying to Gain Control of Electric Grid appeared first on Futurism.

New York City recently announced a moratorium on AI use in K-12 classes for the 2026-2027 school year.

Source

These days, large language models are pushing the line between schizophrenic break and reality closer than ever before.

In a wild new post that’s going viral on X-formerly-Twitter, an enterprising young biochemist named Douglas Yao showcased a new drug he’s been developing with the aid of ChatGPT. He’s not employed by any pharma company, and his drug lab is setup on a couple folding tables in his garage.

Specifically, Yao enthuses about a new drug he calls “PAC-3310,” which he says is meant to treat schizophrenia. “It was designed by ChatGPT,” Yao explains, “and I synthesized it in a chemistry lab I built in my garage.”

The result is a small vial of yellow powder — the first sample of 3310, apparently — which Yao says is modeled after the recently-authorized antipsychotic Cobenfy, “but improved.”

This vial contains a new drug called PAC-3310. It was designed by ChatGPT, and I synthesized it in a chemistry lab I built in my garage.

PAC-3310 is a new selective M4 muscarinic receptor agonist for treating schizophrenia – similar to the recent breakthrough drug Cobenfry, but… pic.twitter.com/o7z6iLR50T

— Douglas Yao (@DouglasYaoDY) September 8, 2026

On his GitHub page, Yao boasts that he’s used AI to “design several thousand new small molecule drugs” over the past year, including a “new Alzheimer’s drug,” which he’s developing in addition to the schizophrenia treatment.

“I then taught myself synthetic chemistry, built a chem lab in my garage, and synthesized a hundred of them,” he continued. “I’m now testing these compounds in cell lines and mice.”

But whatever the one-man pharma lab hopes to accomplish is difficult to say. Yao references the fact that “drug developers typically spend tens of millions of dollars generating pre-clinical efficacy evidence for a new drug,” saying that “my approach can do this at 1/1000 the cost.” This seems to suggest dreams of someday manufacturing the various drugs on a massive scale for human consumption — a process that would take years of grueling and expensive work for a well-resourced drug lab, let alone a solo entrepreneur using a box fan as a fume hood.

For instance, to even approach the Food and Drug Administration’s “New Drug Application” — itself a costly 12-month process requiring inspection of the facility where the drug will be manufactured — one first has to conduct rounds of preclinical testing followed by three phases of clinical trials. Forget “tens of millions,” as Yao mentions; this is a process that can easily cost several billion.

Even then success is far from guaranteed, as only 13.8 percent of drugs that enter clinical trials go on to obtain full approval from the FDA.

While Yao might have been right at home in the free-wheeling research landscape of the 1970s, we no longer live in a world where a mad genius can synthesize hundreds of new drugs a year from a private lab. For one thing, the regulatory environment is much more demanding, as outlined above, to say nothing of the fact that all the low-hanging fruit has been picked from the vine decades ago.

When it comes to Yao’s ChatGPT-generated Alzheimer’s treatment, the landscape is particularly fraught. As one 2019 study found, no new drug has been approved to treat Alzheimer’s since 2003, and not for a lack of trying: as the study notes, over 200 proposed substances meant to treat the disease have either failed in clinical trials or been abandoned by their creators.

The graveyard of failed Alzheimer’s treatments is so vast that the renowned organic chemist Derek Lowe has built a blog on drug discovery which routinely features after-action reports on failed substances designed to treat the neurodegenerative disease. Simply put, to think that ChatGPT alone is enough to overcome these scientific and institutional hurdles is delusional.

Replies to Yao’s post were quick to point out the irony here, of using ChatGPT — which is well-known for leading vulnerable users down dangerous breaks from reality — to synthesize drugs to treat cognitive diseases like Alzheimer’s and schizophrenia.

“Hello schizophrenic, take this medicine that chatgpt made for you,” one poster quipped on X.

Another replied that “believing you single-handedly cured schizophrenia on a commercial laundry folding table in a garage is, ironically, a classic symptom of schizophrenia.”

More on ChatGPT: Man Obliterated When Court Obtains His ChatGPT Transcripts

The post Man Announces That He Has Synthesized a New Schizophrenia Treatment in His Garage, Based on a Formula Devised by ChatGPT appeared first on Futurism.

Le jeu vidéo pourrait bien devenir un terrain d’essai privilégié pour les agents d’IA. Des chercheurs britanniques ont justement imaginé un test qui les confronte à une partie de Civilization VI, où ils doivent planifier leurs actions sur plusieurs centaines de tours. Les résultats mettent en...

Google is one of the major players in AI (meaning machine learning) weather forecast model space. The models it and others generate have their strengths and weaknesses, but the main advantage is that they can have forecast performance similar to traditional models while requiring far less computing horsepower to run. That means they can be run more frequently.

Google recently released version 3 of its WeatherNext model, with the biggest change being that it now ingests some satellite weather data, shortening the lag time between current weather conditions and generating a new forecast. The update is detailed in a white paper.

Reanalysis

Many weather models make use of what’s called a “reanalysis,” which is a sort of model of its own. Reanalyses take in all kinds of weather data and combine them into a single, consistent global snapshot of the atmosphere. That requires that they provide estimates for conditions over locations without real-world measurements, because weather forecast models need to work with a global picture.

Read full article

Comments

© NASA/JPL

Google is one of the major players in AI (meaning machine learning) weather forecast model space. The models it and others generate have their strengths and weaknesses, but the main advantage is that they can have forecast performance similar to traditional models while requiring far less computing horsepower to run. That means they can be run more frequently.

Google recently released version 3 of its WeatherNext model, with the biggest change being that it now ingests some satellite weather data, shortening the lag time between current weather conditions and generating a new forecast. The update is detailed in a white paper.

Reanalysis

Many weather models make use of what’s called a “reanalysis,” which is a sort of model of its own. Reanalyses take in all kinds of weather data and combine them into a single, consistent global snapshot of the atmosphere. That requires that they provide estimates for conditions over locations without real-world measurements, because weather forecast models need to work with a global picture.

Read full article

Comments

© NASA/JPL

As Americans are being hailed as folk heroes for sawing down Flock surveillance cameras across the country, British home appliance company Dyson has come up with a form of video surveillance that we didn’t even know was possible.

Yes: the company has launched a new AI-infused electric toothbrush that features a “100k pixel macro lens camera,” allowing users — brushers? — to “see what their dentist sees.”

The appropriately named Dyson CameraJet claims to analyze 28 images per second and employ an “AI machine-learning algorithm” to identify and target “interdental gaps.” In case you’re wondering how it can take pictures in the darkness of your mouth, a “stroboscopic light” ensures “precise illumination for optimal camera visibility.”

It also hooks up to WiFi and Bluetooth, theoretically opening it up to all kinds of cybersecurity shenanigans if somebody were to identify any vulnerabilities in Dyson’s software.

“By combining a camera, machine learning, precision fluid dynamics, advanced brushing technologies, connectivity and WiFi, Dyson CameraJet introduces an entirely new way to keep your mouth healthy,” founder and chief engineer James Dyson boasted in a statement.

Of course, all of that sensitive video surveillance footage of the inside of your mouth is going to cost you dearly. The undoubtedly overengineered gizmo — which looks more like an undersized hair curling iron than a familiar electric toothbrush — goes for a whopping $499 and at least comes with two brush heads, a charging cable, and a “refill dock.”

Naturally, Dyson will happily sell you a “proprietary non-foaming toothpaste” as well, which lacks a common synthetic foaming agent called sodium lauryl sulfate (SLS) to make sure the toothbrush surveillance camera isn’t impeded by all those suds.

We’ll leave it up to you whether you deem the product, which allegedly required the input of 661 engineers, is worth the price of an iPad Air or roughly ten Oral-B Pro 1000 electric toothbrushes — a popular budget pick for those who don’t need AI to brush their teeth.

More on surveillance: Flock Is in Major Trouble

The post Dyson’s New AI Toothbrush Conducts Video Surveillance on the Inside of Your Mouth appeared first on Futurism.

Machine vision is becoming a practical part of Apple’s U.S. factory training strategy, giving smaller manufacturers access to techniques that help production systems inspect what they make. The recent opening of the Houston manufacturing training center extends that effort into a working electronics production campus, while courses developed with Michigan State University introduce businesses to image analysis, machine learning, and automated quality checks.

For a manufacturer, the attraction can be straightforward: finding a missing component before a product leaves the line, identifying a surface defect before more material is added, or recognizing an assembly error while it can still be corrected cheaply.

These are manageable problems with measurable costs. They also offer an accessible starting point for businesses interested in AI but unsure where it belongs inside their operations.

Machine Vision Enters the Manufacturing Curriculum

The Apple Manufacturing Academy’s current course catalog includes Machine Learning with Vision, a course introducing machine learning and deep learning through practical activities using Apple devices. It sits alongside instruction in automation, quality fundamentals, manufacturing data, maintenance, and process improvement.

That placement gives the subject useful context. A camera can produce an image, but a factory needs a decision: accept the part, flag it for review, check its position, or investigate a process that may be drifting out of tolerance.

Machine vision connects image capture to those operational decisions. Some applications use established rules to measure dimensions or identify patterns. Others use machine learning to recognize acceptable variations and defects that are difficult to describe through fixed rules.

The academy’s public description confirms introductory activities with Apple devices. It does not identify a particular iPhone, Mac configuration, software framework, or industrial camera package that every participant must use.

For smaller businesses, the educational value is learning how to define an inspection problem before purchasing equipment. A company that understands what it needs to detect can have a much more productive conversation with an automation supplier.

Houston Brings Training Closer to Production

Apple opened its 20,000-square-foot Advanced Manufacturing Center on August 13. According to the Houston opening announcement, the free program offers small and medium-sized businesses instruction in advanced manufacturing, interactive labs, and experience with representative production equipment.

The first cohort worked with machine learning for quality control and advanced automation. Participants also explored production challenges, used a holographic table, and assembled and laser etched a product.

The location adds context: the center occupies the same Houston facility that builds advanced AI servers and is scheduled to begin Mac mini manufacturing this year. Its teaching program includes final assembly principles and design considerations for printed circuit board assembly.

Houston is the company’s second U.S. manufacturing learning site, following the Detroit academy established with Michigan State University in August 2025. By the Houston opening, Apple said the academy had helped nearly 1,000 workers, students, and entrepreneurs.

That figure measures people reached through the program. It should not be interpreted as 1,000 factories already operating AI inspection systems.

A Useful Inspection Starts With a Specific Problem

Imagine a small business assembling control panels. Each panel needs the same set of connectors, positioned correctly before the enclosure closes. A machine vision station could check for a missing connector while the assembly remains accessible.

The value comes from catching that error at the right moment. Discovering it after packaging creates extra handling and rework. Discovering it after delivery may involve a return, a service visit, and an unhappy customer.

A useful pilot would therefore begin with that single inspection point. The business could compare the existing process with the assisted one, recording missed defects, unnecessary rechecks, inspection time, and the cost of correcting mistakes.

The camera is only part of the exercise. Workers must agree on what constitutes a defect, collect representative examples, and decide what happens when the result is uncertain.

Those decisions draw heavily on experience already present on the factory floor. An operator may know that an unusual mark is harmless, while a barely visible change in alignment predicts a later assembly problem. Turning that knowledge into consistent inspection criteria is valuable work in its own right.

These are examples of how manufacturers could apply the training, rather than projects Apple has announced for individual participants.

Teaching Machine Vision to Handle Ordinary Variation

A demonstration can look impressive when every part arrives clean, centered, and evenly lit. Production introduces variation: different surface finishes, reflections, changing positions, and batches that look slightly different while remaining within specification.

The inspection system must distinguish those acceptable differences from actual failures. Otherwise, it can reject good products or allow defective ones to continue.

Industrial vision supplier Cognex’s deployment guidance recommends validating systems across product models, defect types, and manufacturing lines. It also describes an approach in which automated inspection handles preliminary checks while experienced inspectors review borderline cases.

For a business learning machine vision, that offers a realistic way to introduce the technology. Employees can compare its decisions with their own, investigate disagreements, and determine whether the results justify extending its role.

An inspection system also needs an owner after installation. Someone must understand when a product change requires new examples, when equipment needs attention, and when performance has slipped. Training that responsibility into the workforce helps prevent an initially successful project from becoming an unreliable fixture.

Smaller Manufacturers Are Already Applying the Training

The academy’s reach extends beyond companies producing consumer electronics. In its May update on participating manufacturers, Apple highlighted Block Imaging, a Michigan business that services and refurbishes medical imaging equipment.

The company hosted visitors during the academy’s Spring Forum and described applying techniques learned from Apple engineers and Michigan State experts to its operations. Apple reported that the academy had supported more than 150 American businesses at that point.

The announcement did not identify a particular machine vision deployment at Block Imaging or publish a percentage improvement attributable to visual inspection. It does provide an example of the training reaching businesses whose work differs substantially from building an iPhone.

That wider participation could benefit Apple indirectly. More capable domestic manufacturers can support equipment maintenance, tooling, components, and specialized production services. Participation itself does not establish a supplier relationship, but the skills can be used across customers and industries.

Access Without a Large Training Budget

The academy’s participation information says eligible small and medium-sized U.S. manufacturers can take courses free of charge. Online learning includes on-demand materials and interactive modules, while in-person sessions add workshops and more personalized instruction.

Digital badges are available for different topics, and completed course materials remain accessible for review. That gives a business room to train an employee around existing production commitments and return to the material when an actual project begins.

Free instruction does not remove the eventual cost of cameras, lighting, integration, maintenance, or employee time. It can help a manufacturer understand those costs before committing to a system.

The first proposal brought back to the factory might be modest: inspect one recurring defect at one station and record the results for several weeks. A supervisor can then weigh the time saved against the extra reviews, maintenance, and mistakes. That is the kind of evidence a business owner can use when deciding whether to install a second station.

The post Machine Vision Joins Apple’s U.S. Factory Training Push appeared first on AppleMagazine.

Apply for just about any office job these days, and there’s a pretty good chance your first round interview won’t be with a human, but with a crummy virtual AI avatar.

Though the practice has become widespread throughout HR departments around the world, a new video released by satirical influencer Graham Zip shows just how humiliating the experience can be — and how straying from the system’s comfort zone can cause the bots to melt down completely.

On logging into the interview, Graham is welcomed by a virtual avatar of a woman identified as “Dana Whitfield,” which prompts the interviewee to “tell me about yourself.” Without skipping a beat, Graham launches into a faux-hallucinated spiel, plausibly constructed to sound like a real answer.

“Hi Dana, it’s great to meet you,” Graham begins. “I’m Chick Haversplat, or, Chick Bongo for short. And uh, Q3 pipeline, Salesforce, Kayak margin.”

The large language model’s brain immediately breaks: “It’s a pleasure to meet you, Chick Bongo,” it replies. “So it sounds like you’ve been deep in the weeds with Salesforce, hygiene, and managing those Kayak margins during the third quarter.”

Asked to elaborate more on his professional background, Graham explains that “freight, cadence, walnut, a little bit of Richard, sprucewood.”

“Six years of sprucewood — actually seven, if you count Richard,” the influencer swiftly replies.

“It sounds like you built up a significant tenure over the last seven years, including your time with Richard at sprucewood,” the bot rejoins. “I’m curious, can you tell me about a difficult situation you handled during that time?”

“Mostly palette forward, a little bit of denim,” comes Graham’s reply. “You know, every pharmacist is a hallway.”

“It sounds like you have a unique way of navigating the palette-forward and denim aspects of your work,” the AI bot replies, now completely lost in the sauce.

At one point, Graham explains that he’s looking for a job with “two plumes of gauze,” where he can follow “the raccoon protocols.”

“It sounds like you’re looking for a position where you can really lean into those raccoon protocols, and the plumes of gauze,” the virtual recruiter responds.

By the end of the “interview,” Graham has the AI bot explaining that there “was a bit of a gap in the drum counting process, after Tuesday, with the work truck and cousin Francis.”

The ease with which the influencer can immediately take control of the entire interview underscores the fact that AI recruiters aren’t there so much to actually gather information on applicants, but to screen candidates for HR departments overwhelmed by a deluge of AI-generated applications. It isn’t clear how Graham’s application progressed, or what role the interview was even for, but if the automated Dana Whitfield’s enthusiasm is any indication, things went pretty well.

“It has been a pleasure chatting today Chick Bongo,” the AI concluded, “and I will be in touch very soon about the next steps.”

More on AI hallucinations: A Scientist Was Presenting His Vibe-Coded New Work When One of His Colleagues Pointed Out Something Extremely Embarrassing

The post Man Pretends to Hallucinate in Job Interview With an AI Bot, Causing It to Go Haywire appeared first on Futurism.

Fruit flies aren't exactly famous for their brainpower; you've probably drowned more than one in a wine glass left too long on the patio table. And yet, working with roughly 140,000 neurons—a brain smaller than a poppy seed—Drosophila can sort through a huge range of smells in a fraction of a second, and then retain the memory of that scent for a long time.

In this, they do much better than current "electronic noses." Even the most advanced ones on the market tend to be expensive, painfully narrow in what they can detect, and quick to forget an odor the moment they learn a new one.

So why not just copy the fly? That's the question a growing number of researchers have been asking—including Kevin Max and Yang Shen at the Okinawa Institute of Science and Technology, whose new algorithm, Spi-Fly, is described in a paper recently published in the journal Neuromorphic Computing and Engineering.

Read full article

Comments

© Joao Paulo Burini

A demonstration shows how agentic AI can support nurses with continuous, non-invasive tracking of infant distress and development.

The Road to Macstock takes us to Dave Hamilton, who previews his Macstock session, “From Curiosity to Capability,” which will focus on practical AI workflows anyone can use. Dave discusses how generative AI has changed coding, project management, proposals, productivity, and procrastination for him. The session will encourage experimentation as a way to get involved as AI tools rapidly evolve.

💾

Learn how NI’s Nigel chatbot is simplifying tasks and other operations related to software and test.