Researchers at the Massachusetts Institute of Technology (MIT) have developed a new algorithm that aims to make generative artificial intelligence models more suitable for applications in which the answer is not enough for the answer to be “close to correct,” but rather must adhere to strict constraints that cannot be exceeded, such as safety rules, physical laws, or the requirements of the task itself. The method is called “Hard Flow,” and it works on pre-trained models without the need to retrain them.
In many generative AI applications, an approximate answer or approximate image of the request can be accepted, but this margin is not suitable for tasks such as planning a robot’s path inside a crowded factory, where an “almost correct” path may lead to the robot colliding with a worker.
Some current methods rely on imposing strict constraints on each intermediate step during the generation process. However, this approach may limit the model's ability to explore better paths; Because every step is required, from the beginning, to remain within narrow limits, even if some temporary steps can later lead to a better and more efficient final result.
Greater freedom during obstetrics
The basic idea of HardFlow is to allow the model greater freedom during the generation process, imposing strict constraints on the final result, rather than applying them to every intermediate step.
The researchers explain that what matters in the end is the final output of the model; Because the internal stages of the generation process are not used after the result is produced. By relaxing constraints during these stages, the model can explore a greater number of solutions and ultimately arrive at an output that adheres to stringent requirements and, at the same time, achieves better quality.
Utilize control theory
The algorithm recasts the constrained generation process as a path optimization problem, taking advantage of tools used in the field of optimal control. Instead of forcing the model to correct each step directly, HardFlow guides it through incremental adjustments toward the desired result.
The researchers say that using control theory provides a mathematical framework for determining the most efficient way to make these corrections.
But applying this type of optimization around a large neural network is computationally challenging; Because the model may consist of hundreds of interconnected layers. To make the process executable, the team exploited the structure of flow-matching models, broke the problem down into a series of smaller, single-step problems, and then used mathematical transformations and approximations to arrive at an algorithm that could run efficiently during actual use of the model.
Results in robotics and computer vision
The researchers tested the method in a number of scenarios, including moving robots, navigating mazes, and text-guided image editing. In these experiments, Hard Flow achieved full compliance with the specified constraints, with better results than the reference methods used for comparison in terms of solution quality.
In one example, a robotic arm was able to avoid obstacles and reach the target via a faster path, while others resulted in either a collision or choosing paths that took longer. The team also noted that the computation time was similar to or less than most competing methods tested, which supports the possibility of using it in practical applications that require a quick response.
Without retraining the model
One of the features of HardFlow is that it works during the use or deployment phase, and does not require training the model from scratch, or retraining it for new constraints. This means that the method can be added to off-the-shelf generative models, including diffusion and flow-matching models, and then used to impose new requirements related to safety, physics or the nature of the task.
The researchers believe that this approach may expand the use of generative models in areas that require a higher degree of assurance, especially when small violations of constraints are unacceptable.
Combining safety and quality of solution
HardFlow's design is not limited to ensuring compliance with constraints; framing the issue as an improvement process allows for additional goals besides safety. In planning the movement of a robot, for example, the algorithm can search for a path that does not collide with any obstacle, while at the same time minimizing the distance or time required to reach the goal.
The team says that combining these two aspects; That is, respecting constraints and improving the quality of the solution were among the main reasons why the method excelled in tests, compared to some methods that focus only on finding a valid solution.
The study was published in the journal "IEEE Transactions on Pattern Analysis and Machine Intelligence."
In the future, the researchers propose to expand the framework so that the model itself can also be updated, allowing its ability to adhere to constraints and the quality of the outputs to be improved in a way that is more adaptive to the task.
Smart watches offer multiple functions for monitoring health, sports, sleep quality, and more. However, the Huawei Watch GT 7 Pro goes beyond that by offering the ability to assess the risk of coronary artery disease, the risk of high blood sugar, and women’s ovarian health, in addition to other health and sports features in a design that combines the spirit of adventure with touches of modern elegance. Asharq Al-Awsat tested the watch before launching it in the Arab region, and we mention a summary of the experience.
Elegant design
The outer case of the watch features an octagonal design that is consistent with the sharp and streamlined lines typical of previous versions, with precise edges for an elegant and luxurious appearance at all times. The watch offers superior levels of protection thanks to the use of high-quality materials, as the body is made of durable titanium alloy. This significantly enhances its resistance to shocks and ensures a long lifespan.
The front panel is also equipped with a durable screen made of scratch-resistant glass, in addition to using a frame made of “ceramic” with fine nanotechnology in order to provide a solid shield that protects it from daily scratches resulting from friction with hard surfaces, and reduces damage resulting from unexpected shocks during outdoor exercise.
The watch is equipped with an innovative bracelet that provides superior comfort and ease of putting on and taking off thanks to its practical design that keeps the ends securely and with complete elegance. This smart design contributes to reducing friction with clothing during intense motor activities or in daily activities.
The watch is also equipped with practical control buttons designed with great care to facilitate interaction with the screen and quick access to various functions and systems. The design of the buttons includes a “rotating crown” that allows you to zoom in on maps or route details with high accuracy with simple twists, as well as being easy to use for browsing menus and notifications. The side button provides immediate response to running exercises or direct access to shortcuts to favorite applications, with the ability to press the two buttons together to take screenshots to save results and sports information and share them directly.
A proactive health system with advanced medical vision
The watch provides an integrated and comprehensive system for proactive and advanced health management, based on the latest technologies and high-precision sensors to monitor physical condition and monitor vital indicators with extreme accuracy. These functions include, according to the watch designers:
* Electrocardiogram: quickly captures data to help early detect heart disorders, such as atrial fibrillation and premature heartbeats; Which provides valuable insights for proactive medical intervention.
* Coronary artery disease risk assessment: This is done by collecting daily activity data, sleep, and electrocardiogram metrics through the application of research to generate comprehensive risk reports.
* Assessing the risk of high blood sugar: The watch performs this function by relying on high-performance sensors to collect vital signs data around the clock and detect any abnormal metabolic signals.
* Ovarian Health Risk Assessment: This function works by accurately tracking temperature levels and sleep durations throughout the night, then providing personalized recommendations to improve the woman’s overall health.
* New Fitness Index: Analyzes sleep quality, heart rate fluctuations and calories burned to assess recovery status and accurately determine daily stress level.
* Smart Morning Briefing: Provides personalized insights and accurate rest and exercise recommendations based on the previous day's data.
* Smart Health Insights powered by intelligent algorithms: analyzes multiple dimensions of sleep, overall health and stress levels to provide relevant lifestyle guidance.
* Enhanced Health Rings system: Enhances care goals to include new tasks, such as accurately tracking water intake via interactive icons directly from the screen.
* New Community Health System: Provides the ability to monitor the health status of family members remotely and receive immediate alerts when any abnormal readings are detected to share concern and provide health support seamlessly.
Unlimited sporting horizons for breaking records
The watch offers a huge range of advanced sports styles (more than 110 types of sports) specifically designed to meet the passion of exercise enthusiasts of various types, and provide accurate scientific data to enhance performance. These modes include comprehensive winter sports, such as indoor and outdoor skiing, as they support maps of more than three thousand global ski resorts, and provide accurate tracking of tracks, speeds, and levels of slope. Sports capabilities also include the feature of detecting turns and calculating the effort forces generated during sliding, in addition to a cross-country skiing mode designated for long-distance challenges.
The watch also provides cycling enthusiasts with smart climbing plans that clearly display elevation and gradient data, with smart alerts for sharp turns to ensure the highest levels of safety. The watch also supports measuring the heart rate of the lactic threshold for cycling (an advanced professional measurement that helps athletes and cycling enthusiasts train more intelligently and increase their physical efficiency, which means the critical point of the heart rate after which the amount of lactic acid begins to accumulate in the muscles faster than the body can get rid of it), and providing performance analysis supported by artificial intelligence to determine points of improvement.
The activities also include specialized features for the sport of golf, including calculating distances adjusted according to the terrain and slope, allocating the locations of the drill flags, and displaying automatic rotational guidance of the green area with score cards. The watch also includes outdoor running modes with a running energy measurement and a running ability indicator, as well as a built-in mini-exercise application that provides quick sessions to relieve stress and activate the body. The watch allows integration with external fitness applications to support sports, such as tennis, badminton, etc., and tracks movements accurately.
Seamless interactive experience
The user experience through the Huawei Health application on mobile phones is characterized by a flexible and easy design that allows the synchronization of sports and health data instantly and without any complexity. The application allows you to import and manage sports routes, resort maps and training files with one click and go directly to the watch. This makes it easy to review details and graphs of speed and altitude after completing exercises. The application also facilitates the creation of sports teams to monitor locations and health conditions during group exercises outdoors, with the ability to share sports posters and various achievements across social media platforms very smoothly.
Long-term sustainable energy
The watch battery delivers a high-density charge designed with advanced cell technology to provide a long life that extends up to 21 days of light use or 12 days of heavy use. Moreover, the watch supports competitive running style outdoors for long enough periods to cover the toughest races. This ensures stable and sustainable power for users throughout long trips, in addition to supporting fast wireless charging, which recharges the battery with high efficiency in a short time to ensure permanent readiness without interruption.
Technical specifications and chassis capabilities
The technical specifications of the watch feature an accurate and bright display screen that works with high accuracy and ensures superior clarity even under direct sunlight thanks to the high intensity of brightness. The watch offers a built-in storage capacity to store a large number of interfaces, maps, and music files for listening without an Internet connection, and supports Bluetooth wireless technology and highly accurate geolocation, in addition to advanced water and dust resistance and diving support to ensure reliable performance in the harshest environmental conditions.
- Screen diameter: 1.47 inches.
- Screen resolution: 466 x 466 pixels.
- Screen density: 317 pixels per inch.
- Brightness intensity: 3000 lumens.
Image display technology: AMOLED.
- Built-in storage capacity: 64 GB.
- Built-in speakers: Yes to answer calls.
- Operating system: “Harmony OS 6.1” (supports connection to phones running the “Android” and “iOS” operating systems).§
- Support for wireless networks: “Bluetooth 6.0”, Near Field Communication (NFC), and GPS.
-Built-in sensors: acceleration sensor, location, heart rate, barometric pressure, blood oxygen saturation, body temperature measurement, and geographical compass.
- Battery charge: 867 mAh - hour.
-Water and dust resistance: Yes, according to the IP69 standard (can withstand high water pressure for 10 minutes at a depth of 10 meters under the surface of the water).
-Thickness: 11.3 mm.
-Weight: 55.5 grams.
Prices and availability date in Arab markets
The watch is available in the Arab region starting next October 4, in two versions in yellow, dark green, or black, at a price starting at 999 Saudi riyals (about 266 US dollars).
Muse, an AI agent developed by Meta, has helped me manage my dental insurance, book dinner tables, and create my podcast. But it also required giving him my most private data, and putting my trust in him, writes Eli Tan(*).
Secure calls
Last week I answered a call from an unknown number. A voice said: “I have contacted Anthem, a dental insurance company, and I am transferring the call to you now.” The call was coming from an artificial intelligence “agent” created by Meta, the company that owns the Facebook and Instagram platforms. Ten minutes before that moment, the AI agent had called the insurance company, provided it with my membership number, answered a security question (to verify my identity), and waited on the line before transferring the call to me to speak to the relevant employee... just as I had asked him to do.
I was amazed at how easy this agent made handling the insurance process. This was one of many tasks she completed with the help of Muse, which Meta revealed this month.
The first consumer-oriented AI agent
Muse is the first consumer AI agent launched by a major technology company and made available to the public. The technology industry sees these agent tools as the “next frontier” in this field, as they have capabilities that far exceed those of traditional “chatbots”: agents can browse the Internet, log into websites, fill out spreadsheets, and make phone calls on behalf of users, just to name a few.
The test requires submitting personal and banking data
To test Muse, I used it continuously over the course of two weeks, handing over many aspects of my life to it. I had to trust him with my most private data, including linking him to my bank accounts and email. Then I assigned him dozens of daily tasks, like ordering groceries, scheduling my meetings, and forming my fantasy football teams.
After two weeks, I found Muse to be the most useful AI application in my experience. This was clearly demonstrated in a specific situation that occurred after I linked my credit cards to the app and asked it to track my expenses and record them in Google Sheets. I watched the program open tabs with hundreds of rows of data within minutes, then identify two duplicate subscriptions and cancel them for me. Muse then used my email to request a refund for a free trial subscription to a newspaper that had expired, saving me $44.99.
Privacy problem
I'm not entirely sure whether Muse and other AI agents will catch on; These tools require a level of trust with which many may not feel comfortable, especially when dealing with a company like Meta, which has faced privacy issues. Some sites also impose restrictions on these agents within their terms of service. For example, Amazon recently blocked Muse from accessing it.
Melissa Vadasseril, a senior manager at Acoustic, a marketing platform that relies on intelligent agent technologies, says the emergence of Muse reminded her of the 2010 launch of Apple's digital personal assistant, Siri, an innovation that many were late to adopt. “People didn't realize what they could do with it,” she added. As for “Muse,” she says: “I think it is still a bit early for the market, but I am very curious to see how things will develop.”
My experience is for two weeks
My two-week experience with Muse began by downloading the free app and logging in using my Instagram and Facebook accounts.
