AI Detector Eraser

AI Detector Eraser — independent reviews, comparisons, pricing and step-by-step guides on Aizhi.

• 

  Verbal overshadowing

  Verbal overshadowing is a phenomenon where giving a verbal description of sensory input impairs formation of memories of that input. This was first reported by Schooler and Engstler-Schooler (1990) where it was shown that the effects can be observed across multiple domains of cognition which are known to rely on non-verbal knowledge and perceptual expertise. One example of this is memory, which has been known to be influenced by language. Seminal work by Carmichael and collaborators (1932) demonstrated that when verbal labels are connected to non-verbal forms during an individual's encoding process, it could potentially bias the way those forms are reproduced. Because of this, memory performance relying on reportable aspects of memory that encode visual forms should be vulnerable to the effects of verbalization. == Initial findings == Schooler and Engstler-Schooler (1990) were the first to report findings of verbal overshadowing. In their study, participants watched a video of a simulated robbery and were instructed to either verbally describe the robber or engage in a control task. Those who engaged in giving a verbal description were less likely to correctly identify the robber from a test lineup, compared to those who engaged in the control task. A larger effect was detected when the verbal description was provided 20, rather than 5, minutes after the video, and immediately before the test lineup. A meta-analysis by Meissner and Brigham (2001) supported the effects of verbal overshadowing, showing a small but reliably negative effect. == General effects of verbal overshadowing == The effects of verbal overshadowing have been generalized across multiple domains of cognition that are known to rely on non-verbal knowledge and perceptual expertise, such as memory. Memory has been known to be influenced by language. Seminal work by Carmichael and collaborators (1932) demonstrated that labels attached to, or associated with, non-verbal forms during memory encoding can affect the way the forms were subsequently reproduced. Because of this, memory performance that relies on reportable aspects of memory that encode visual forms should be vulnerable to the effects of verbalization. Pelizzon, Brandimonte, and Luccio (2002) found that visual memory representations appear to incorporate visual, spatial, and temporal characteristics. It is explained as follows: With the temporal code (where the only information available is the sequence of the stimuli), performance levels remain high, unless participants are required to retrieve the stimuli in a different order from that used at encoding (visual cue). In this case, performance is significantly impaired, even in the presence of a visual cue. The study showed that order information acts as a link between the two separate representations of figure and background, hence preventing verbal overshadowing at encoding (temporal component) or attenuating its influence at retrieval (spatial component).(p. 960) Hatano, Ueno, Kitagami, and Kawaguchi found that verbal overshadowing is likely to occur when participants verbally described targets in detail. Detailed verbal descriptions resulted in more frequently inaccurate descriptions that in turn created inaccurate representations in the memories of participants. Inaccuracies are also likely to occur when face recognition comes immediately after verbalization. Other forms of non-verbal knowledge affected by verbal overshadowing include the following: [Verbal overshadowing] has also been observed when participants attempt to generate descriptions of other 'difficult-to-describe' stimuli such as colors (Schooler and Engstler-Schooler, 1990) or abstract figures (Brandimonte et al., 1997), or other non-visual tasks such as wine tasting (Melcher and Schooler, 1996), decision making (Wilson and Schooler, 1991), and insight problem-solving. (p. 871) (Schooler et al., 1993) Verbalization of stimuli leads to the disruption of non-reportable processes that are necessary for achieving insight solutions, which are distinct from language processes. Schooler, Ohlsson, and Brooks (1993) found that face recognition requires information that cannot be adequately verbalized, giving rise to difficulty in describing factors in recognition judgments. Subjects were less effective in solving insight problems when compelled to put their thoughts in words, which suggests that language may interfere with thought. The verbal overshadowing effect was not seen when participants engaged in articulatory suppression. Performance was reduced in both the verbal and non-verbal description conditions. This is evidence that verbal encoding plays a role in face recognition. By testing with distracting faces presented between study and test, Lloyd-Jones and Brown (2008) suggested a dual-process approach to recognition memory took place, that verbalization influenced familiarity-based processes at first, but its effects were later seen on recollection, when discrimination between items became more difficult. == Verbal overshadowing in facial recognition == The verbal overshadowing effect can be found for facial recognition because faces are predominately processed in a holistic or configurable manner. (Tanaka & Farah, 1993; Tanaka & Sengco, 1997) Verbalizing one's memory for a face is done using a featural or analytic strategy, leading to a drift from the configurable information about the face and to impaired recognition performance. However, Fallshore & Schooler (1995) found that the verbal overshadowing effect was not found when participants described faces of races different from their own. A study by Brown and Lloyd-Jones (2003) found that there was no verbal overshadowing effect found in car descriptions; it was only seen in facial descriptions. The authors noted that descriptions were no different on any measure including accuracy. It is suggested that less expertise in verbalizing faces rather than cars invokes a stronger shift in verbal and featural processing. This supports the concept of a transfer inappropriate retrieval framework and addresses some limitations of the effect. Wickham and Swift (2006) suggested that the verbal overshadowing effect is not seen in describing all faces, and one aspect that determines this is distinctiveness. Results showed that typical faces produce verbal overshadowing, while distinctive faces did not. In studies of eyewitness reports, variation in response criteria given by participants influenced the quality of the descriptions generated and accuracy on identification task, known as the retrieval-based effect. Face recognition was also impaired when subjects described a familiar face, such as a parent, or when describing a previously seen but novel face. Dodson, Johnson, and Schooler (1997) found that recognition was also impaired when participants were provided with a description of a previously seen face, and they were able to ignore provided versus self-generated descriptions more easily. This finding of verbal overshadowing suggested that eyewitness recognition is not only affected by their own descriptions, but of descriptions heard from others, such other eyewitness testimonies. == Voice recognition == The verbal overshadowing effect has also been found to affect voice identification. Research shows that describing a non-verbal stimuli leads to a decrease in recognition accuracy. In an unpublished study by Schooler, Fiore, Melcher, and Ambadar (1996), participants listened to a tape-recorded voice, after which they were asked either to verbally describe it or to not do so, and then asked to distinguish the voice from 3 similar distractor voices. The results showed that verbal overshadowing impaired accuracy of recognition based on gut feeling, suggesting an overall verbal overshadowing for voice recognition. Due to the forensic relevance of voices heard over the telephone and harassing phone calls that are often a problem for police, Perfect, Hunt, and Harris (2002) examined the influence of three factors on accuracy and confidence in voice recognition from a line-up. They expected to find an effect, because voice represents a class of stimuli that is difficult to describe verbally. This meets Schooler et al.'s (1997) modality mismatch criterion, meaning that describing the speakers age, gender, or accent is difficult, making voice recognition susceptible to the verbal overshadowing phenomenon. It was found that the method of memory encoding had no impact on performance, and that hearing a telephone voice reduced confidence but did not affect accuracy. They also found that providing a verbal description impaired accuracy but had no effect on confidence. The data showed an effect of verbal overshadowing in voice recognition and provided yet another disassociation between confidence and performance. Although there was a difference in confidence level, witnesses were able to identify voices over the telephone as accurately as voices heard direc

  Read more →
• 

  Simultaneous localization and mapping

  Simultaneous localization and mapping (SLAM) is a process where a computer constructs or updates a map of an unknown environment while simultaneously keeping track of an entity's location within it. While this initially appears to be a chicken or the egg problem, there are several algorithms known to solve it in, at least approximately, tractable time for certain environments. Popular approximate solution methods include the particle filter, extended Kalman filter, covariance intersection, and GraphSLAM. SLAM algorithms are based on concepts in computational geometry and computer vision, and are used in robot navigation, robotic mapping and odometry for virtual reality or augmented reality. SLAM algorithms are tailored to the available resources and are not aimed at perfection but at operational compliance. Published approaches are employed in self-driving cars, unmanned aerial vehicles, autonomous underwater vehicles, planetary rovers, newer domestic robots and even inside the human body. == Mathematical description of the problem == Given a series of controls u t {\displaystyle u_{t}} and sensor observations o t {\displaystyle o_{t}} over discrete time steps t {\displaystyle t} , the SLAM problem is to compute an estimate of the agent's state x t {\displaystyle x_{t}} and a map of the environment m t {\displaystyle m_{t}} . All quantities are usually probabilistic, so the objective is to compute P ( m t + 1 , x t + 1 | o 1 : t + 1 , u 1 : t ) {\displaystyle P(m_{t+1},x_{t+1}|o_{1:t+1},u_{1:t})} Applying Bayes' rule gives a framework for sequentially updating the location posteriors, given a map and a transition function P ( x t | x t − 1 ) {\displaystyle P(x_{t}|x_{t-1})} , P ( x t | o 1 : t , u 1 : t , m t ) = ∑ m t − 1 P ( o t | x t , m t , u 1 : t ) ∑ x t − 1 P ( x t | x t − 1 ) P ( x t − 1 | m t , o 1 : t − 1 , u 1 : t ) / Z {\displaystyle P(x_{t}|o_{1:t},u_{1:t},m_{t})=\sum _{m_{t-1}}P(o_{t}|x_{t},m_{t},u_{1:t})\sum _{x_{t-1}}P(x_{t}|x_{t-1})P(x_{t-1}|m_{t},o_{1:t-1},u_{1:t})/Z} where Z {\displaystyle Z} is the normalization constant, which ensures all the probabilities sum up to 1. Similarly the map can be updated sequentially by P ( m t | x t , o 1 : t , u 1 : t ) = ∑ x t ∑ m t P ( m t | x t , m t − 1 , o t , u 1 : t ) P ( m t − 1 , x t | o 1 : t − 1 , m t − 1 , u 1 : t ) {\displaystyle P(m_{t}|x_{t},o_{1:t},u_{1:t})=\sum _{x_{t}}\sum _{m_{t}}P(m_{t}|x_{t},m_{t-1},o_{t},u_{1:t})P(m_{t-1},x_{t}|o_{1:t-1},m_{t-1},u_{1:t})} Like many inference problems, the solutions to inferring the two variables together can be found, to a local optimum solution, by alternating updates of the two beliefs in a form of an expectation–maximization algorithm. == Algorithms == Statistical techniques used to approximate the above equations include Kalman filters and particle filters (the algorithm behind Monte Carlo Localization). They provide an estimation of the posterior probability distribution for the pose of the robot and for the parameters of the map. Methods which conservatively approximate the above model using covariance intersection are able to avoid reliance on statistical independence assumptions to reduce algorithmic complexity for large-scale applications. Other approximation methods achieve improved computational efficiency by using simple bounded-region representations of uncertainty. Set-membership techniques are mainly based on interval constraint propagation. They provide a set which encloses the pose of the robot and a set approximation of the map. Bundle adjustment, and more generally maximum a posteriori estimation (MAP), is another popular technique for SLAM using image data, which jointly estimates poses and landmark positions, increasing map fidelity, and is used in commercialized SLAM systems such as Google's ARCore which replaces their prior augmented reality computing platform named Tango, formerly Project Tango. MAP estimators compute the most likely explanation of the robot poses and the map given the sensor data, rather than trying to estimate the entire posterior probability. New SLAM algorithms remain an active research area, and are often driven by differing requirements and assumptions about the types of maps, sensors and models as detailed below. Many SLAM systems can be viewed as combinations of choices from each of these aspects. === Mapping === Topological maps are a method of environment representation which capture the connectivity (i.e., topology) of the environment rather than creating a geometrically accurate map. Topological SLAM approaches have been used to enforce global consistency in metric SLAM algorithms. In contrast, grid maps use arrays (typically square or hexagonal) of discretized cells to represent a topological world, and make inferences about which cells are occupied. Typically the cells are assumed to be statistically independent to simplify computation. Under such assumption, P ( m t | x t , m t − 1 , o t ) {\displaystyle P(m_{t}|x_{t},m_{t-1},o_{t})} are set to 1 if the new map's cells are consistent with the observation o t {\displaystyle o_{t}} at location x t {\displaystyle x_{t}} and 0 if inconsistent. Modern self driving cars mostly simplify the mapping problem to almost nothing, by making extensive use of highly detailed map data collected in advance. This can include map annotations to the level of marking locations of individual white line segments and curbs on the road. Location-tagged visual data such as Google's StreetView may also be used as part of maps. Essentially such systems simplify the SLAM problem to a simpler localization only task, perhaps allowing for moving objects such as cars and people only to be updated in the map at runtime. === Sensing === SLAM will always use several different types of sensors, and the powers and limits of various sensor types have been a major driver of new algorithms. Statistical independence is the mandatory requirement to cope with metric bias and with noise in measurements. Different types of sensors give rise to different SLAM algorithms which assumptions are most appropriate to the sensors. At one extreme, laser scans or visual features provide details of many points within an area, sometimes rendering SLAM inference unnecessary because shapes in these point clouds can be easily and unambiguously aligned at each step via image registration. At the opposite extreme, tactile sensors are extremely sparse as they contain only information about points very close to the agent, so they require strong prior models to compensate in purely tactile SLAM. Most practical SLAM tasks fall somewhere between these visual and tactile extremes. Sensor models divide broadly into landmark-based and raw-data approaches. Landmarks are uniquely identifiable objects in the world which location can be estimated by a sensor, such as Wi-Fi access points or radio beacons. Raw-data approaches make no assumption that landmarks can be identified, and instead model P ( o t | x t ) {\displaystyle P(o_{t}|x_{t})} directly as a function of the location. Optical sensors may be one-dimensional (single beam) or 2D- (sweeping) laser rangefinders, 3D high definition light detection and ranging (lidar), 3D flash lidar, 2D or 3D sonar sensors, and one or more 2D cameras. Since the invention of local features, such as SIFT, there has been intense research into visual SLAM (VSLAM) using primarily visual (camera) sensors, because of the increasing ubiquity of cameras such as those in mobile devices. Follow up research includes. Both visual and lidar sensors are informative enough to allow for landmark extraction in many cases. Other recent forms of SLAM include tactile SLAM (sensing by local touch only), radar SLAM, acoustic SLAM, and Wi-Fi-SLAM (sensing by strengths of nearby Wi-Fi access points). Recent approaches apply quasi-optical wireless ranging for multi-lateration (real-time locating system (RTLS)) or multi-angulation in conjunction with SLAM as a tribute to erratic wireless measures. A kind of SLAM for human pedestrians uses a shoe mounted inertial measurement unit as the main sensor and relies on the fact that pedestrians are able to avoid walls to automatically build floor plans of buildings by an indoor positioning system. For some outdoor applications, the need for SLAM has been almost entirely removed due to high precision differential GPS sensors. From a SLAM perspective, these may be viewed as location sensors which likelihoods are so sharp that they completely dominate the inference. However, GPS sensors may occasionally decline or go down entirely, e.g. during times of military conflict, which are of particular interest to some robotics applications. === Kinematics modeling === The P ( x t | x t − 1 ) {\displaystyle P(x_{t}|x_{t-1})} term represents the kinematics of the model, which usually include information about action commands given to a robot. As a part of the model, the kinematics of the robot is included, to improve estimates of sensing under con

  Read more →
• 

  Trustworthy AI

  Trustworthy AI refers to artificial intelligence systems that are designed to have transparent reasoning, are explainable (XAI), accountable, robust, fair and honest, respectful of data privacy, and steerable or alignable with human goals. == Terminology == Recent work in AI ethics distinguishes trustworthiness and trustability as two different conditions relevant to trustworthy AI. Trustworthiness is concerned with whether an AI system or the institutions deploying it merit trust by being reliable, fair, and accountable. Trustability, on the other hand, is the prior question of whether a given entity is even the kind of thing to which interpersonal trust can coherently apply as opposed to mere instrumental reliance. Some philosophers argue that current AI systems are best understood as tools that are not genuine targets of interpersonal trust. They argue that trust should be directed toward the human and institutional arrangements that govern the systems' design, deployment, and oversight. This stance supports interpreting "trustworthy AI" as trustworthy governance and use of AI rather than trust in the artifacts themselves. Transparency in AI involves making the processes and decisions of such systems understandable to users and stakeholders. Accountability ensures that there are protocols for addressing adverse outcomes or biases that may arise, with designated responsibilities for oversight and remediation. Robustness and security aim to ensure that AI systems perform reliably under various conditions and are safeguarded against malicious attacks. Harmlessness can be achieved by refusal training: training the models to avoid problematic requests, and by adding filters to detect and prevent discussion on biased, unethical, or dangerous outputs. There is research on how to train AI so that it aligns with human goals. == Techniques and ITU standardization == Trustworthy AI creation is a goal of AI governance and policymaking. To achieve transparency and data privacy, several privacy-enhancing technologies (PETs) can be used. These include: Homomorphic encryption for computing with encrypted data without ever decrypting it. Federated learning and secure multi-party computation (MPC) for distributing the model training without sharing information between the learning centers and computing servers. Differential privacy for exposing statistical data while guaranteeing that no private information is exposed. Zero-knowledge proof - providing proven validity for statements without disclosing any extra information. A work programme for achieving Trustworthy AI was set up by the International Telecommunication Union, an agency of the United Nations, initiated under its AI for Good programme. Its origin lies with the ITU-WHO Focus Group on Artificial Intelligence for Health, where a strong need for both privacy and analytics created demand for a standard in these technologies. In 2020, AI for Good moved online, and the TrustworthyAI seminar series was established to initiate discussions on these topics. This eventually led to standardization activities. === Multi-party computation === Secure multi-party computation (MPC) is being standardized under "Question 5" (the incubator) of ITU-T Study Group 17. === Homomorphic encryption === Homomorphic encryption allows for computing on encrypted data, where the outcomes or result is still encrypted and unknown to those performing the computation, but can be deciphered by the original encryptor. It is often developed with the goal of enabling use in jurisdictions different from the data creation (under, for instance, GDPR). ITU has been collaborating since the early stage of the HomomorphicEncryption.org standardization meetings, which has developed a standard on homomorphic encryption. The fifth homomorphic encryption meeting was hosted at ITU HQ in Geneva. === Federated learning === Zero-sum masks as used by federated learning for privacy preservation are used extensively in the multimedia standards of ITU-T Study Group 16 (VCEG) such as JPEG, MP3, H.264, and H.265 (commonly known as MPEG). === Zero-knowledge proof === Previous pre-standardization work on the topic of zero-knowledge proof has been conducted in the ITU-T Focus Group on Digital Ledger Technologies. === Differential privacy === The application of differential privacy in the preservation of privacy was examined at several of the "Day 0" machine learning workshops at AI for Good Global Summits. == Mozilla "Rebel Alliance" == In January 2026, the Mozilla Foundation and its subsidiaries announced a strategic shift to deploy their entire $1.4 billion reserve into building what foundation president Mark Surman termed a "rebel alliance" for trustworthy AI. Framed by Surman as a mission-driven alternative to the market dominance of OpenAI and Anthropic, the initiative seeks to establish an open-source AI stack by 2028. The alliance includes several startups funded via Mozilla Ventures, specifically focusing on decentralized governance and transparency: Trail: A firm developing AI compliance frameworks for regulated industries. Transformer Lab: A developer of open-source tools for AI model management. Oumi: A platform for training and deploying open-source models. The "rebel alliance" terminology is a historical reference to Mozilla's efforts in 1998 to challenge Microsoft's browser monopoly. While the $1.4 billion in funding is significant, it has been contrasted with the tens of billions in capital raised by proprietary competitors like OpenAI.

  Read more →
• 

  RealSense

  RealSense is an American technology company that develops depth cameras and computer-vision systems used in robotics, access control, industrial automation and healthcare. The company’s stereoscopic 3D cameras and software are marketed as a perception platform for “physical AI”, particularly for humanoid robots and autonomous mobile robots (AMRs). RealSense was incubated for more than a decade inside Intel’s perceptual computing and depth-sensing group before being spun out as an independent company in July 2025 with a US$50 million Series A round backed by a semiconductor-focused private equity firm and strategic investors including Intel Capital and the MediaTek Innovation Fund. Following the spin-out, RealSense announced a strategic collaboration with Nvidia to integrate its AI depth cameras with the Nvidia Jetson Thor robotics platform, the Isaac Sim simulation environment and the Holoscan Sensor Bridge for low-latency sensor fusion. In November 2025, Swiss access-solutions provider dormakaba acquired a minority stake in RealSense and formed a partnership to develop AI-powered biometric access-control and security systems for data centres, airports and other critical infrastructure. == History == === Origins in Intel Perceptual Computing === Intel began developing depth-sensing and perceptual-computing technologies in the early 2010s under the Perceptual Computing brand, with research spanning gesture control, facial recognition and eye-tracking systems. The work led to a series of 3D cameras and developer challenge programmes intended to stimulate software ecosystems for natural-user interfaces. In 2014 Intel rebranded the effort as Intel RealSense, positioning the technology as a family of depth cameras and vision processors for PCs, mobile devices and embedded systems. Early devices such as the F200 and R200 were integrated into laptops and tablets from OEMs including Asus, HP, Dell, Lenovo and Acer, and were also sold as standalone webcams by partners such as Razer and Creative. === Refocus on robotics and near-closure === By the late 2010s Intel had steered RealSense away from mainstream PC peripherals toward robotics, industrial and embedded applications, adding stereo and lidar-based depth cameras to the portfolio. In August 2021, trade publication CRN reported that Intel planned to wind down the RealSense business as part of a broader restructuring, raising questions about the future of the product line. Despite that announcement, Intel continued to invest in new custom silicon for depth cameras, and RealSense remained widely used in mobile robots and automation projects. === Spin-out as RealSense Inc. (2025) === On 11 July 2025, Intel completed the spin-out of its RealSense 3D-camera business into a new privately held company, RealSense Inc., and the new entity announced a US$50 million Series A funding round. The round was led by a semiconductor-focused private equity investor with participation from Intel Capital, MediaTek Innovation Fund and other strategics. Independent coverage described RealSense as serving more than 3,000 active customers and supplying depth cameras to a large share of global AMR and humanoid robot platforms. The company stated that it would continue to support the existing Intel RealSense product roadmap while accelerating development of AI-enabled cameras and perception software. === Strategic partnerships and investments === In October 2025 RealSense and Nvidia announced a strategic collaboration centered on integrating RealSense AI depth cameras with Nvidia’s Jetson Thor robotics compute modules, the Isaac Sim simulation environment and the Holoscan Sensor Bridge for multi-sensor streaming. The collaboration is positioned as enabling “physical AI” workloads such as whole-body humanoid control, real-time mapping and safety-critical human–robot interaction. On 19 November 2025, dormakaba announced that it had acquired a minority stake in RealSense and entered into a partnership to co-develop intelligent access-control solutions, including biometric gates for airports and enterprise facilities. The partnership aims to combine RealSense’s depth and facial-authentication technology with dormakaba’s installed base of sensors, doors and turnstiles. == Products == === Depth-camera families === RealSense’s products are sold as modular components (depth modules, vision processors and complete cameras) and as integrated systems with on-device AI. The company continues to offer and support the Intel RealSense D400 family of active-stereo depth cameras (including the D415, D435 and D455), which are widely used in robotics and automation. These devices combine a RealSense Vision Processor from the D4 family with dual infrared imagers and, on some models, an RGB camera. Earlier generations of Intel RealSense cameras, including the F200, R200, SR300 and the L515 lidar camera, remain in use in niche and legacy applications but are no longer the focus of the independent company’s roadmap. === D555 PoE depth camera === The first new hardware platform announced after the spin-out was the RealSense Depth Camera D555, a ruggedised stereo-depth device aimed at industrial and robotics deployments. The D555 uses the longer-range D450 optical module with a global shutter and integrates RealSense’s Vision SoC V5, a new generation of vision processor optimised for neural-network inference and depth computation. Key features highlighted in technical coverage include: Power over Ethernet (PoE), allowing power and data to be delivered over a single cable and supporting both RJ45 and ruggedised M12 connections; an IP-rated enclosure designed for harsh indoor and outdoor environments; a built-in inertial measurement unit (IMU) to support simultaneous localisation and mapping (SLAM) and motion tracking; native support for ROS 2 and integration with the open-source RealSense SDK. According to independent reporting, the D555 is used in AI-enabled embedded-vision applications in mobile robots and fixed industrial systems, and was among the first RealSense products to be tightly integrated with Nvidia’s Jetson Thor and Holoscan platforms for low-latency sensor fusion. === Software and SDK === RealSense cameras are supported by a cross-platform, open-source software stack historically branded as Intel RealSense SDK 2.0. The SDK provides device drivers, depth and point-cloud processing, tracking and calibration tools, and bindings for languages such as C++, Python and C#. The independent company has continued to maintain and extend the SDK for new hardware, including D555 and other Vision SoC V5-based devices, and publishes reference integrations for ROS 2 and industrial-automation frameworks. === Biometrics and access-control products === In addition to general-purpose depth cameras, RealSense offers facial-authentication hardware and software, commonly referred to as RealSense ID, for biometric access control and identity verification. These products combine an active depth sensor with a dedicated neural-network pipeline running on embedded processors, aimed at applications such as secure doors, turnstiles and kiosks. Use-case material published by partners describes deployments of RealSense-based biometric readers in school lunch programmes, agricultural biosecurity checkpoints and enterprise facilities. The dormakaba partnership announced in 2025 extends this portfolio to integrated biometric gates and sensor-equipped doors in airports and data centres. == Applications == === Robotics and automation === RealSense depth cameras are used in autonomous mobile robots, humanoid robots, drones and industrial automation systems for tasks such as obstacle avoidance, navigation and manipulation. Reuters reported in 2025 that RealSense cameras were embedded in around 60 percent of the world’s AMRs and humanoid robots, citing customers including Unitree Robotics and ANYbotics. Developers and integrators use RealSense systems with platforms such as Nvidia Jetson, ROS and proprietary motion-planning stacks. === Biometrics and security === RealSense technology is also applied in biometric access control and surveillance, where depth and infrared imaging are used to improve anti-spoofing performance for facial recognition. The dormakaba investment and collaboration is aimed at integrating these capabilities into boarding gates, staff entrances and secure facilities, with RealSense providing perception hardware and algorithms and dormakaba providing access-control infrastructure and global distribution. == Reception == Early coverage of Intel RealSense for consumer PCs noted that the technology’s impact would depend on the availability of compelling software and use cases for depth-sensing cameras. Later reporting on the spin-out has characterised the new company as part of a broader wave of investment in robotics and physical AI, with some analysts suggesting that RealSense’s installed base and patent portfolio give it an advantage as dep

  Read more →
• 

  Stripe, Inc.

  Stripe, Inc. is an Irish and American multinational financial services and software as a service (SaaS) company dual-headquartered in South San Francisco, California, United States, and Dublin, Ireland. The company primarily offers payment-processing software and application programming interfaces for e-commerce websites and mobile applications. Stripe is the largest privately owned financial technology company with a valuation of about $159 billion and over $1.9 trillion in payment volume processed in 2025, processing transactions for 5 million businesses in that year. == History == Irish entrepreneur brothers John and Patrick Collison founded Stripe in Palo Alto, California, in 2010, and serve as the company's president and CEO, respectively. In 2011 the company received a $2 million investment, including contributions from Elon Musk, PayPal founder Peter Thiel, Irish entrepreneur Liam Casey, and venture capital firms Sequoia Capital, Andreessen Horowitz, and SV Angel. In March 2013, Stripe made its first acquisition, Kickoff, a chat and task-management application. In 2012 the company moved from Palo Alto to San Francisco. In October 2019, the company announced that it would be moving from the South of Market area to Oyster Point in the neighbouring city of South San Francisco in 2021. In February 2021, Mark Carney, former governor of the Bank of Canada and of the Bank of England, was appointed to the company's board. Carney stepped down from his role with the company in 2025 in order to run for the leadership of the Liberal Party. Stripe acquired accountancy platform Recko in October 2021 whose solution was to be added to Stripe's existing suite of financial tools. In January 2022, Stripe entered a five-year partnership with Ford Motor Company. Through the deal, Stripe would handle transactions for consumer vehicle orders and reservations. That same month, Stripe partnered with Spotify to help the company monetize subscriptions. In April 2022, Twitter announced that it would partner with Stripe, Inc. (digital payments processor) for piloting cryptocurrency pay-outs for limited users in the platform. In April 2022, Stripe announced its strategic partnership with UK-based financial technology company ION. The Wall Street Journal reported in July 2022 that the company's internal share price had fallen, causing its implied valuation to drop from $95 billion to $74 billion. In November 2022, the company announced it intended to initiate layoffs, terminating some 14% of its workforce. Throughout 2022 and 2023, the company announced a number of large enterprise customers, including Airbnb, Amazon, Microsoft, Uber, BMW, Maersk, Zara, Lotus, Alaska Airlines, Le Monde, and Toyota. The company also announced in March 2023 that OpenAI is working with Stripe to commercialize its generative AI technology. In January 2025, Stripe sent layoff notices to nearly 300 workers, primarily affecting roles in Product, Operations and Engineering. The company experienced controversy when the company sent a cartoon picture of a duck to the laid-off employees. Stripe's Chief People Officer Rob McIntosh later apologized for the mistake. After re-enabling cryptocurrency pay-ins in April 2024, starting with USDC, Stripe completed the acquisition of Bridge in February 2025. The acquisition of the two-year-old stablecoin platform company is valued at $1.1 billion. In June 2025, the company acquired Privy, which powers crypto wallets. In September 2025, Stripe announced it was powering Instant Checkout in ChatGPT and released Agentic Commerce Protocol for agentic commerce, which was co-developed with OpenAI. In October 2025, the company opened its second headquarters in Dublin, Ireland. In February 2026, Stripe was valued at $159 billion in a tender offer posted for employees and shareholders. The tender offer was about a 70% increase from Stripe's previous valuation published in February 2025, where it was valued at $91.5 billion. Stripe also announced that its total volume increased to $1.9 trillion USD in 2025, a 34% increase from 2024. == Technology company == === Payment processing === Stripe provides application programming interfaces that web developers can use to integrate payment processing into their websites and mobile applications. The company introduced Stripe Connect in 2012, a multiparty payments solution that lets software developers embed payments natively into their products. In April 2018, Stripe released antifraud tools, branded "Radar", that block fraudulent transactions. The same year, it expanded its services to include a billing product for online businesses, allowing businesses to manage subscription recurring revenue and invoicing. Stripe's point-of-sale service called Terminal was made available to US users on 11 June 2019. Terminal had previously been invitation-only. Terminal is currently available in Australia, Canada, France, Germany, Ireland, the Netherlands, New Zealand, Singapore, and the United Kingdom. The service offers physical credit-card readers designed to work with Stripe. On 5 September 2019, Stripe launched a merchant cash-advance scheme called Stripe Capital. The scheme allows Stripe merchants to request an advance on future payments they expect to process through their Stripe merchant account. In June 2021, the company launched Stripe Tax, a service to allow businesses to automatically calculate and collect sales tax, VAT, and GST, initially rolling out to 30 countries and all US states. As of 2025, it has been made available in 102 countries. In May that year, Stripe introduced Payment Links, a no-code product allowing businesses to create a link to a checkout page and begin accepting payments on social platforms or direct channels. In January 2022, Stripe agreed to acquire Terminal manufacturing partner BBPOS, allowing the company to bring the hardware development of Terminal readers in-house. In February, it was announced as Apple's first partner on in-person Tap to Pay, which enables businesses to accept contactless payments using an iPhone and a partner-enabled iOS app. In May, Stripe announced Data Pipeline, a tool for Stripe users who store data with Amazon Redshift or Snowflake Data Cloud. Data Pipeline syncs Stripe data and reports with Amazon Redshift or Snowflake Data Cloud, where they can be queried in combination with other business information. That month, the company also introduced Stripe Financial Connections, enabling businesses to establish direct connections with their customers’ bank accounts to verify accounts for payments and pay-outs, check balances to reduce payment failures, and cut fraud by confirming bank account ownership. In September 2023, Stripe announced that its optimized checkout suite allowed businesses to offer their customers more than 100 payment methods. In May 2025, Stripe announced a new AI foundational model for payments, and introduced stablecoin powered accounts. === Corporate finance === In July 2018, Stripe introduced Stripe Issuing, a product that allows online businesses and platforms to create their own physical and digital credit and debit cards. === Atlas === On 14 February 2016, the company launched the Atlas platform to help start-ups register as US corporations, targeting foreign entrepreneurs. The platform was originally invitation-only. In March 2016, Cuba was added to the list of countries covered under the program. Originally, companies registered using Atlas were set up as Delaware-based C corporations. As of 30 April 2018, the option to be registered as limited liability companies was added. Companies set up using Atlas automatically had a business bank account and Stripe merchant account set up. === Link === In May 2021, Stripe launched Link, a service for saving and auto-filling payment details when paying via Stripe. The service supported payments in over 185 countries and Stripe reported plans to make it available to platform businesses through its API. In September 2025, Patrick Collison announced that Link had surpassed 200 million users. === Other === In 2018, Stripe started a publishing company named Stripe Press to promote ideas that support businesses. In 2019, Stripe began offering loans and credit cards to businesses in the United States. The company stated that loans are approved automatically using machine-learning models, with no human intervention. The following year, the company introduced Stripe Treasury, which provides its platform users APIs to embed financial services, allowing their customers to send, receive, and store funds. In October 2020, Stripe announced Stripe Climate, a service for businesses to fund atmospheric carbon research and capture. In 2022, Stripe started a new subsidiary called Frontier that would direct spending on carbon removal. It announced $925 million in funding from major Silicon Valley companies to fund start up companies performing carbon capture to kick-start the industry. Stripe Identity, launched in Ju

  Read more →
• 

  Tag (metadata)

  In information systems, a tag is a keyword or term assigned to a piece of information (such as an Internet bookmark, multimedia, database record, or computer file). This kind of metadata helps describe an item and allows it to be found again by browsing or searching. Tags are generally chosen informally and personally by the item's creator or by its viewer, depending on the system, although they may also be chosen from a controlled vocabulary. Tagging was popularized by websites associated with Web 2.0 and is an important feature of many Web 2.0 services. It is now also part of other database systems, desktop applications, and operating systems. == Overview == People use tags to aid classification, mark ownership, note boundaries, and indicate online identity. Tags may take the form of words, images, or other identifying marks. An analogous example of tags in the physical world is museum object tagging. People were using textual keywords to classify information and objects long before computers. Computer based search algorithms made the use of such keywords a rapid way of exploring records. Tagging gained popularity due to the growth of social bookmarking, image sharing, and social networking websites. These sites allow users to create and manage labels (or "tags") that categorize content using simple keywords. Websites that include tags often display collections of tags as tag clouds, as do some desktop applications. On websites that aggregate the tags of all users, an individual user's tags can be useful both to them and to the larger community of the website's users. Tagging systems have sometimes been classified into two kinds: top-down and bottom-up. Top-down taxonomies are created by an authorized group of designers (sometimes in the form of a controlled vocabulary), whereas bottom-up taxonomies (called folksonomies) are created by all users. This definition of "top down" and "bottom up" should not be confused with the distinction between a single hierarchical tree structure (in which there is one correct way to classify each item) versus multiple non-hierarchical sets (in which there are multiple ways to classify an item); the structure of both top-down and bottom-up taxonomies may be either hierarchical, non-hierarchical, or a combination of both. Some researchers and applications have experimented with combining hierarchical and non-hierarchical tagging to aid in information retrieval. Others are combining top-down and bottom-up tagging, including in some large library catalogs (OPACs) such as WorldCat. When tags or other taxonomies have further properties (or semantics) such as relationships and attributes, they constitute an ontology. In folder system a file cannot exist in two or more folders so tag system has been thought more convenient. But transitioning to tag system requires awareness of difference between properties of two systems. In folder system the information of classification is put outside of the file and we can change folder at once. In tag system the information of classification is put inside the file so changing its tag means changing the file and it needs to be saved again and takes time. Metadata tags as described in this article should not be confused with the use of the word "tag" in some software to refer to an automatically generated cross-reference; examples of the latter are tags tables in Emacs and smart tags in Microsoft Office. == History == The use of keywords as part of an identification and classification system long predates computers. Paper data storage devices, notably edge-notched cards, that permitted classification and sorting by multiple criteria were already in use prior to the twentieth century, and faceted classification has been used by libraries since the 1930s. In the late 1970s and early 1980s, Emacs, the text editor for Unix systems, offered a companion software program called Tags that could automatically build a table of cross-references called a tags table that Emacs could use to jump between a function call and that function's definition. This use of the word "tag" did not refer to metadata tags, but was an early use of the word "tag" in software to refer to a word index. Online databases and early websites deployed keyword tags as a way for publishers to help users find content. In the early days of the World Wide Web, the keywords meta element was used by web designers to tell web search engines what the web page was about, but these keywords were only visible in a web page's source code and were not modifiable by users. In 1997, the collaborative portal "A Description of the Equator and Some ØtherLands" produced by documenta X, Germany, used the folksonomic term Tag for its co-authors and guest authors on its Upload page. In "The Equator" the term Tag for user-input was described as an abstract literal or keyword to aid the user. However, users defined singular Tags, and did not share Tags at that point. In 2003, the social bookmarking website Delicious provided a way for its users to add "tags" to their bookmarks (as a way to help find them later); Delicious also provided browseable aggregated views of the bookmarks of all users featuring a particular tag. Within a couple of years, the photo sharing website Flickr allowed its users to add their own text tags to each of their pictures, constructing flexible and easy metadata that made the pictures highly searchable. The success of Flickr and the influence of Delicious popularized the concept, and other social software websites—such as YouTube, Technorati, and Last.fm—also implemented tagging. In 2005, the Atom web syndication standard provided a "category" element for inserting subject categories into web feeds, and in 2007 Tim Bray proposed a "tag" URN. == Examples == === Within a blog === Many systems (and other web content management systems) allow authors to add free-form tags to a post, along with (or instead of) placing the post into a predetermined category. For example, a post may display that it has been tagged with baseball and tickets. Each of those tags is usually a web link leading to an index page listing all of the posts associated with that tag. The blog may have a sidebar listing all the tags in use on that blog, with each tag leading to an index page. To reclassify a post, an author edits its list of tags. All connections between posts are automatically tracked and updated by the blog software; there is no need to relocate the page within a complex hierarchy of categories. === Within application software === Some desktop applications and web applications feature their own tagging systems, such as email tagging in Gmail and Mozilla Thunderbird, bookmark tagging in Firefox, audio tagging in iTunes or Winamp, and photo tagging in various applications. Some of these applications display collections of tags as tag clouds. === Assigned to computer files === There are various systems for applying tags to the files in a computer's file system. In Apple's Mac System 7, released in 1991, users could assign one of seven editable colored labels (with editable names such as "Essential", "Hot", and "In Progress") to each file and folder. In later iterations of the Mac operating system ever since OS X 10.9 was released in 2013, users could assign multiple arbitrary tags as extended file attributes to any file or folder, and before that time the open-source OpenMeta standard provided similar tagging functionality for Mac OS X. Several semantic file systems that implement tags are available for the Linux kernel, including Tagsistant. Microsoft Windows allows users to set tags only on Microsoft Office documents and some kinds of picture files. Cross-platform file tagging standards include Extensible Metadata Platform (XMP), an ISO standard for embedding metadata into popular image, video and document file formats, such as JPEG and PDF, without breaking their readability by applications that do not support XMP. XMP largely supersedes the earlier IPTC Information Interchange Model. Exif is a standard that specifies the image and audio file formats used by digital cameras, including some metadata tags. TagSpaces is an open-source cross-platform application for tagging files; it inserts tags into the filename. === For an event === An official tag is a keyword adopted by events and conferences for participants to use in their web publications, such as blog entries, photos of the event, and presentation slides. Search engines can then index them to make relevant materials related to the event searchable in a uniform way. In this case, the tag is part of a controlled vocabulary. === In research === A researcher may work with a large collection of items (e.g. press quotes, a bibliography, images) in digital form. If he/she wishes to associate each with a small number of themes (e.g. to chapters of a book, or to sub-themes of the overall subject), then a group of tags for these themes can be attached to each of the items in

  Read more →
• 

  TasteDive

  TasteDive (formerly named TasteKid) is an entertainment recommendation engine for films, TV shows, music, video games, books, people, places, and brands. It also has elements of a social media site; it allows users to connect with "tastebuds", people with like minded interests. == History == TasteDive was founded in 2008 as TasteKid by brothers Andrei Oghina and Felix Oghina. In 2019, it was acquired by Qloo headquartered in NYC. "Qloo has built for developers and enterprises what TasteDive has built for individuals". == Description == When a user types in the title of a film or TV show, the site's algorithm provides a list of similar content. It provides recommendations for TV shows to watch based on films liked by the user, and vice versa. It also provides recommendations for music, video games, and books, and includes film and TV trailers and music videos. An account is free and is not required to receive recommendations, but recommendations are more accurate for those with an account. The more a user explores the site, the more the site learns about the user's preferences and the better the results become. The site also has a social media aspect where one can see activity and gain recommendations from other users, how many others in the community like or dislike any recommendation, and how popular their tastes are within the TasteDive community. The main competitors of TasteDive are Taste App, Trakt.tv and Tastoid.

  Read more →
• 

  Shakey the robot

  Shakey the Robot was the first general-purpose mobile robot able to reason about its own actions. While other robots would have to be instructed on each individual step of completing a larger task, Shakey could analyze commands and break them down into basic chunks by itself. Due to its nature, the project combined research in robotics, computer vision, and natural language processing. Because of this, it was the first project that melded logical reasoning and physical action. Shakey was developed at the Artificial Intelligence Center of Stanford Research Institute (now called SRI International). Some of the most notable results of the project include the A search algorithm, the Hough transform, and the visibility graph method. == History == Shakey was developed from approximately 1966 through 1972 with Charles Rosen, Nils Nilsson and Peter Hart as project managers. Other major contributors included Alfred Brain, Sven Wahlstrom, Bertram Raphael, Richard Duda, Richard Fikes, Thomas Garvey, Helen Chan Wolf and Michael Wilber. The project was funded by the Defense Advanced Research Projects Agency (DARPA) based on a SRI proposal submitted in April 1964 for research in "Intelligent Automata", later "Intelligent Automata to Reconnaissance". It was originally designed to have two retractable arms. Now retired from active duty, Shakey is currently on view in a glass display case at the Computer History Museum in Mountain View, California. The project inspired numerous other robotics projects, most notably the Centibots. == Software == The robot's programming was primarily done in LISP. The Stanford Research Institute Problem Solver (STRIPS) planner it used was conceived as the main planning component for the software it utilized. As the first robot that was a logical, goal-based agent, Shakey experienced a limited world. A version of Shakey's world could contain a number of rooms connected by corridors, with doors and light switches available for the robot to interact with. Shakey had a short list of available actions within its planner. These actions involved traveling from one location to another, turning the light switches on and off, opening and closing the doors, climbing up and down from rigid objects, and pushing movable objects around. The STRIPS automated planner could devise a plan to enact all the available actions, even though Shakey himself did not have the capability to execute all the actions within the plan personally. An example mission for Shakey might be something like, an operator types the command "push the block off the platform" at a computer console. Shakey looks around, identifies a platform with a block on it, and locates a ramp in order to reach the platform. Shakey then pushes the ramp over to the platform, rolls up the ramp onto the platform, and pushes the block off the platform. == Hardware == Physically, the robot was particularly tall, and had an antenna for a radio link, sonar range finders, a television camera, on-board processors, and collision detection sensors ("bump detectors"). The robot's tall stature and tendency to shake resulted in its name: We worked for a month trying to find a good name for it, ranging from Greek names to whatnot, and then one of us said, 'Hey, it shakes like hell and moves around, let’s just call it Shakey.' == Research results == The development of Shakey provided far-reaching impact on the fields of robotics and artificial intelligence, as well as computer science in general. Some of the more notable results include the development of the A search algorithm, which is widely used in pathfinding and graph traversal, the process of plotting an efficiently traversable path between points; the Hough transform, which is a feature extraction technique used in image analysis, computer vision, and digital image processing; and the visibility graph method for finding Euclidean shortest paths among obstacles in the plane. == Media and awards == In 1969 the SRI published "SHAKEY: Experimentation in Robot Learning and Planning", a 24-minute video. The project then received media attention. This included an article in the New York Times on April 10, 1969. In 1970, Life referred to Shakey as the "first electronic person"; and in November 1970 National Geographic Magazine covered Shakey and the future of computers. The Association for the Advancement of Artificial Intelligence's AI Video Competition's awards are named "Shakeys" because of the significant impact of the 1969 video. Shakey was inducted into Carnegie Mellon University's Robot Hall of Fame in 2004 alongside such notables as ASIMO and C-3PO. Shakey has been honored with an IEEE Milestone in Electrical Engineering and Computing. Shakey was showcased in the BBC's Towards Tomorrow: Robot (1967) documentary.

  Read more →
• 

  Neural field

  In machine learning, a neural field (also known as implicit neural representation, neural implicit, or coordinate-based neural network), is a mathematical field that is fully or partially parametrized by a neural network. Initially developed to tackle visual computing tasks, such as rendering or reconstruction (e.g., neural radiance fields), neural fields emerged as a promising strategy to deal with a wider range of problems, including surrogate modelling of partial differential equations, such as in physics-informed neural networks. Differently from traditional machine learning algorithms, such as feed-forward neural networks, convolutional neural networks, or transformers, neural fields do not work with discrete data (e.g. sequences, images, tokens), but map continuous inputs (e.g., spatial coordinates, time) to continuous outputs (i.e., scalars, vectors, etc.). This makes neural fields not only discretization independent, but also easily differentiable. Moreover, dealing with continuous data allows for a significant reduction in space complexity, which translates to a much more lightweight network. == Formulation and training == According to the universal approximation theorem, provided adequate learning, sufficient number of hidden units, and the presence of a deterministic relationship between the input and the output, a neural network can approximate any function to any degree of accuracy. Hence, in mathematical terms, given a field y = Φ ( x ) {\textstyle {\boldsymbol {y}}=\Phi ({\boldsymbol {x}})} , with x ∈ R n {\displaystyle {\boldsymbol {x}}\in \mathbb {R} ^{n}} and y ∈ R m {\displaystyle {\boldsymbol {y}}\in \mathbb {R} ^{m}} , a neural field Ψ θ {\displaystyle \Psi _{\theta }} , with parameters θ {\displaystyle {\boldsymbol {\theta }}} , is such that: Ψ θ ( x ) = y ^ ≈ y {\displaystyle \Psi _{\theta }({\boldsymbol {x}})={\hat {\boldsymbol {y}}}\approx {\boldsymbol {y}}} === Training === For supervised tasks, given N {\displaystyle N} examples in the training dataset (i.e., ( x i , y i ) ∈ D t r a i n , i = 1 , … , N {\displaystyle ({\boldsymbol {x_{i}}},{\boldsymbol {y_{i}}})\in {\mathcal {D_{train}}},i=1,\dots ,N} ), the neural field parameters can be learned by minimizing a loss function L {\displaystyle {\mathcal {L}}} (e.g., mean squared error). The parameters θ ~ {\displaystyle {\tilde {\theta }}} that satisfy the optimization problem are found as: θ ~ = argmin θ 1 N ∑ ( x i , y i ) ∈ D t r a i n L ( Ψ θ ( x i ) , y i ) {\displaystyle {\tilde {\boldsymbol {\theta }}}={\underset {\boldsymbol {\theta }}{\text{argmin}}}\;{\frac {1}{N}}\sum _{({\boldsymbol {x_{i}}},{\boldsymbol {y_{i}}})\in {\mathcal {D_{train}}}}{\mathcal {L}}(\Psi _{\theta }({\boldsymbol {x}}_{i}),{\boldsymbol {y}}_{i})} Notably, it is not necessary to know the analytical expression of Φ {\displaystyle \Phi } , for the previously reported training procedure only requires input-output pairs. Indeed, a neural field is able to offer a continuous and differentiable surrogate of the true field, even from purely experimental data. Moreover, neural fields can be used in unsupervised settings, with training objectives that depend on the specific task. For example, physics-informed neural networks may be trained on just the residual. === Spectral bias === As for any artificial neural network, neural fields may be characterized by a spectral bias (i.e., the tendency to preferably learn the low frequency content of a field), possibly leading to a poor representation of the ground truth. In order to overcome this limitation, several strategies have been developed. For example, SIREN uses sinusoidal activations, while the Fourier-features approach embeds the input through sines and cosines. == Conditional neural fields == In many real-world cases, however, learning a single field is not enough. For example, when reconstructing 3D vehicle shapes from Lidar data, it is desirable to have a machine learning model that can work with arbitrary shapes (e.g., a car, a bicycle, a truck, etc.). The solution is to include additional parameters, the latent variables (or latent code) z ∈ R d {\displaystyle {\boldsymbol {z}}\in \mathbb {R} ^{d}} , to vary the field and adapt it to diverse tasks. === Latent code production === When dealing with conditional neural fields, the first design choice is represented by the way in which the latent code is produced. Specifically, two main strategies can be identified: Encoder: the latent code is the output of a second neural network, acting as an encoder. During training, the loss function is the objective used to learn the parameters of both the neural field and the encoder. Auto-decoding: each training example has its own latent code, jointly trained with the neural field parameters. When the model has to process new examples (i.e., not originally present in the training dataset), a small optimization problem is solved, keeping the network parameters fixed and only learning the new latent variables. Since the latter strategy requires additional optimization steps at inference time, it sacrifices speed, but keeps the overall model smaller. Moreover, despite being simpler to implement, an encoder may harm the generalization capabilities of the model. For example, when dealing with a physical scalar field f : R 2 → R {\displaystyle f:\mathbb {R} ^{2}\rightarrow \mathbb {R} } (e.g., the pressure of a 2D fluid), an auto-decoder-based conditional neural field can map a single point to the corresponding value of the field, following a learned latent code z {\displaystyle {\boldsymbol {z}}} . However, if the latent variables were produced by an encoder, it would require access to the entire set of points and corresponding values (e.g. as a regular grid or a mesh graph), leading to a less robust model. === Global and local conditioning === In a neural field with global conditioning, the latent code does not depend on the input and, hence, it offers a global representation (e.g., the overall shape of a vehicle). However, depending on the task, it may be more useful to divide the domain of x {\displaystyle {\boldsymbol {x}}} in several subdomains, and learn different latent codes for each of them (e.g., splitting a large and complex scene in sub-scenes for a more efficient rendering). This is called local conditioning. === Conditioning strategies === There are several strategies to include the conditioning information in the neural field. In the general mathematical framework, conditioning the neural field with the latent variables is equivalent to mapping them to a subset θ ∗ {\displaystyle {\boldsymbol {\theta }}^{}} of the neural field parameters: θ ∗ = Γ ( z ) {\displaystyle {\boldsymbol {\theta }}^{}=\Gamma ({\boldsymbol {z}})} In practice, notable strategies are: Concatenation: the neural field receives, as input, the concatenation of the original input x {\displaystyle {\boldsymbol {x}}} with the latent codes z {\displaystyle {\boldsymbol {z}}} . For feed-forward neural networks, this is equivalent to setting θ ∗ {\displaystyle {\boldsymbol {\theta }}^{}} as the bias of the first layer and Γ ( z ) {\displaystyle \Gamma ({\boldsymbol {z}})} as an affine transformation. Hypernetworks: a hypernetwork is a neural network that outputs the parameters of another neural network. Specifically, it consists of approximating Γ ( z ) {\displaystyle \Gamma ({\boldsymbol {z}})} with a neural network Γ ^ γ ( z ) {\displaystyle {\hat {\Gamma }}_{\gamma }({\boldsymbol {z}})} , where γ {\displaystyle {\boldsymbol {\gamma }}} are the trainable parameters of the hypernetwork. This approach is the most general, as it allows to learn the optimal mapping from latent codes to neural field parameters. However, hypernetworks are associated to larger computational and memory complexity, due to the large number of trainable parameters. Hence, leaner approaches have been developed. For example, in the Feature-wise Linear Modulation (FiLM), the hypernetwork only produces scale and bias coefficients for the neural field layers. === Meta-learning === Instead of relying on the latent code to adapt the neural field to a specific task, it is also possible to exploit gradient-based meta-learning. In this case, the neural field is seen as the specialization of an underlying meta-neural-field, whose parameters are modified to fit the specific task, through a few steps of gradient descent. An extension of this meta-learning framework is the CAVIA algorithm, that splits the trainable parameters in context-specific and shared groups, improving parallelization and interpretability, while reducing meta-overfitting. This strategy is similar to the auto-decoding conditional neural field, but the training procedure is substantially different. == Applications == Thanks to the possibility of efficiently modelling diverse mathematical fields with neural networks, neural fields have been applied to a wide range of problems: 3D scene reconstruction: neural fields can be used to model t

  Read more →
• 

  Interactive activation and competition networks

  Interactive activation and competition (IAC) networks are artificial neural networks used to model memory and intuitive generalizations. They are made up of nodes or artificial neurons which are arrayed and activated in ways that emulate the behaviors of human memory. The IAC model is used by the parallel distributed processing (PDP) Group and is associated with James L. McClelland and David E. Rumelhart; it is described in detail in their book Explorations in Parallel Distributed Processing: A Handbook of Models, Programs, and Exercises. This model does not contradict any currently known biological data or theories, and its performance is close enough to human performance as to warrant further investigation.

  Read more →
• 

  Semantic parameterization

  Semantic parameterization is a conceptual modeling process for expressing natural language descriptions of a domain in first-order predicate logic. The process yields a formalization of natural language sentences in Description Logic to answer the who, what and where questions in the Inquiry-Cycle Model (ICM) developed by Colin Potts and his colleagues at the Georgia Institute of Technology. The parameterization process complements the Knowledge Acquisition and autOmated Specification (KAOS) method, which formalizes answers to the when, why and how ICM questions in Temporal Logic, to complete the ICM formalization. The artifacts used in the parameterization process include a dictionary that aligns the domain lexicon with unique concepts, distinguishing between synonyms and polysemes, and several natural language patterns that aid in mapping common domain descriptions to formal specifications. == Relationship to other theories == Semantic Parameterization defines a meta-model consisting of eight roles that are domain-independent and reusable. Seven of these roles correspond to Jeffrey Gruber's thematic relations and case roles in Charles Fillmore's case grammar: The Inquiry-Cycle Model (ICM) was introduced to drive elicitation between engineers and stakeholders in requirements engineering. The ICM consists of who, what, where, why, how and when questions. All but the when questions, which require a Temporal Logic to represent such phenomena, have been aligned with the meta-model in semantic parameterization using Description Logic (DL). == Introduction with Example == The semantic parameterization process is based on Description Logic, wherein the TBox is composed of words in a dictionary, including nouns, verbs, and adjectives, and the ABox is partitioned into two sets of assertions: 1) those assertions that come from words in the natural language statement, called the grounding, and 2) those assertions that are inferred by the (human) modeler, called the meta-model. Consider the following unstructured natural language statement (UNLS) (see Breaux et al. for an extended discussion): UNLS1.0 The customer1,1 must not share2,2 the access-code3,3 of the customer1,1 with someone4,4 who is not the provider5,4. The modeler first identifies intensional and extensional polysemes and synonyms, denoted by the subscripts: the first subscript uniquely refers to the intensional index, i.e., the same first index in two or more words refer to the same concept in the TBox; the second subscript uniquely refers to the extensional index, i.e., two same second index in two or more words refer to the same individual in the ABox. This indexing step aligns words in the statement and concepts in the dictionary. Next, the modeler identifies concepts from the dictionary to compose the meta-model. The following table illustrates the complete DL expression that results from applying semantic parameterization.

  Read more →
• 

  Zvi Mowshowitz

  Zvi Mowshowitz is an American writer and member of the rationalist community who primarily discusses new developments in artificial intelligence. He is a former competitive Magic: The Gathering player and was CEO of MetaMed. == Career == Mowshowitz is an alumnus of Columbia University and holds a bachelor's degree in mathematics. He co-founded and was the CEO of MetaMed, a medical research analysis firm. He has worked at Jane Street Capital, and has worked for the gambling industry in Las Vegas. He attempted to launch a blockchain game, Emergents, in 2020. === Magic: The Gathering === Mowshowitz held a developer intern position at Wizards of the Coast R&D in 2005. He created the deck TurboZvi. His first-place finishes at major competitions were the 1999 World Championships as part of the four-person United States national team, the 2001 Pro Tour Tokyo, and two 2003 Grand Prix. He has placed in the top eight of four Pro Tours, and earned over $140,000 playing Magic competitively. In 2007, Mowshowitz was elected into the Magic Hall of Fame. Last updated: 12 May 2013Source: Wizards.com Mowshowitz has written about Magic for several outlets, including the official Magic website. === Later career === Mowshowitz is on the board of directors for the Center for Applied Rationality, and is a member of the rationalist community. He also founded Balsa Research, a nonprofit think tank which advocated for the repeal of the Jones Act, increasing the housing supply, and reform of the National Environmental Policy Act. In 2023, Mowshowitz wrote an article for Vox on the topic of artificial intelligence safety. Mowshowitz has a blog on Substack under the name "Don't Worry about the Vase". He has written on topics such as artificial intelligence, economics, and the COVID-19 pandemic. == Personal life == Mowshowitz is the son of American biochemist Deborah Mowshowitz. His parents have both worked as Columbia University professors.

  Read more →
• 

  GPT-4o

  GPT-4o ("o" for "omni") is a multilingual, multimodal generative pre-trained transformer developed by OpenAI and released in May 2024. It can process and generate text, images and audio. Upon release, GPT-4o was free in ChatGPT, though paid subscribers had higher usage limits. GPT-4o was removed from ChatGPT in August 2025 when GPT-5 was released, but OpenAI reintroduced it for paid subscribers after users complained about the sudden removal. GPT-4o's audio-generation capabilities are used in ChatGPT's Advanced Voice Mode. On July 18, 2024, OpenAI released GPT-4o mini, a smaller version of GPT-4o which replaced GPT-3.5 Turbo on the ChatGPT interface. The image generation model GPT Image 1, which is based on GPT-4o, replaced DALL-E 3 in ChatGPT in March 2025. OpenAI retired GPT-4o from ChatGPT on February 13, 2026. However, as of February 2026 the voice mode is still powered by GPT-4o or GPT-4o mini, depending on the usage and plan. == Background == Multiple versions of GPT-4o were originally secretly launched under different names on Arena (formerly LMArena and Chatbot Arena) as three different models. These three models were called gpt2-chatbot, im-a-good-gpt2-chatbot, and im-also-a-good-gpt2-chatbot. On 7 May 2024, OpenAI CEO Sam Altman tweeted "im-a-good-gpt2-chatbot", which was commonly interpreted as a confirmation that these were new OpenAI models being A/B tested. == Capabilities == When released in May 2024, GPT-4o achieved state-of-the-art results in voice, multilingual, and vision benchmarks, setting new records in audio speech recognition and translation. GPT-4o scored 88.7 on the Massive Multitask Language Understanding (MMLU) benchmark compared to 86.5 for GPT-4. Unlike GPT-3.5 and GPT-4, which rely on other models to process sound, GPT-4o natively supports voice-to-voice. The Advanced Voice Mode was delayed and finally released to ChatGPT Plus and Team subscribers in September 2024. On 1 October 2024, the Realtime API was introduced. When released, the model supported over 50 languages, which OpenAI claims cover over 97% of speakers. GPT-4o has knowledge up to October 2023 and a context length of 128k tokens. === Corporate customization === In August 2024, OpenAI introduced a new feature allowing corporate customers to customize GPT-4o using proprietary company data. This customization, known as fine-tuning, enables businesses to adapt GPT-4o to specific tasks or industries, enhancing its utility in areas like customer service and specialized knowledge domains. Previously, fine-tuning was available only on the less powerful model GPT-4o mini. The fine-tuning process requires customers to upload their data to OpenAI's servers, with the training typically taking one to two hours. OpenAI's focus with this rollout is to reduce the complexity and effort required for businesses to tailor AI solutions to their needs, potentially increasing the adoption and effectiveness of AI in corporate environments. == GPT-4o mini == On July 18, 2024, OpenAI released a smaller and cheaper version, GPT-4o mini. According to OpenAI, its low cost is expected to be particularly useful for companies, startups, and developers that seek to integrate it into their services, which often make a high number of API calls. Its API costs $0.15 per million input tokens and $0.6 per million output tokens, compared to $2.50 and $10, respectively, for GPT-4o. It is also significantly more capable and 60% cheaper than GPT-3.5 Turbo, which it replaced on the ChatGPT interface. The price after fine-tuning doubles: $0.3 per million input tokens and $1.2 per million output tokens. == Controversies == === Scarlett Johansson controversy === As released, GPT-4o offered five voices: Breeze, Cove, Ember, Juniper, and Sky. A similarity between the voice of American actress Scarlett Johansson and Sky was quickly noticed. On May 14, Entertainment Weekly asked themselves whether this likeness was on purpose. On May 18, Johansson's husband, Colin Jost, joked about the similarity in a segment on Saturday Night Live. On May 20, 2024, OpenAI disabled the Sky voice. Scarlett Johansson starred in the 2013 sci-fi movie Her, playing Samantha, an artificially intelligent virtual assistant personified by a female voice. As part of the promotion leading up to the release of GPT-4o, Sam Altman on May 13 tweeted a single word: "her". OpenAI stated that each voice was based on the voice work of a hired actor. According to OpenAI, "Sky's voice is not an imitation of Scarlett Johansson but belongs to a different professional actress using her own natural speaking voice." CTO Mira Murati stated "I don't know about the voice. I actually had to go and listen to Scarlett Johansson's voice." OpenAI further stated the voice talent was recruited before reaching out to Johansson. On May 21, Johansson issued a statement explaining that OpenAI had repeatedly offered to make her a deal to gain permission to use her voice as early as nine months prior to release, a deal she rejected. She said she was "shocked, angered, and in disbelief that Mr. Altman would pursue a voice that sounded so eerily similar to mine that my closest friends and news outlets could not tell the difference." In the statement, Johansson also used the incident to draw attention to the lack of legal safeguards around the use of creative work to power leading AI tools, as her legal counsel demanded OpenAI detail the specifics of how the Sky voice was created. Observers noted similarities to how Johansson had previously sued and settled with The Walt Disney Company for breach of contract over the direct-to-streaming rollout of her Marvel film Black Widow, a settlement widely speculated to have netted her around $40M. Also on May 21, Shira Ovide at The Washington Post shared her list of "most bone-headed self-owns" by technology companies, with the decision to go ahead with a Johansson sound-alike voice despite her opposition and then denying the similarities ranking 6th. On May 24, Derek Robertson at Politico wrote about the "massive backlash", concluding that "appropriating the voice of one of the world's most famous movie stars — in reference [...] to a film that serves as a cautionary tale about over-reliance on AI — is unlikely to help shift the public back into [Sam Altman's] corner anytime soon." === Sycophancy === In April 2025, OpenAI rolled back an update of GPT-4o due to excessive sycophancy, after widespread reports that it had become flattering and agreeable to the point of supporting clearly delusional or dangerous ideas. In the United States, at least nine lawsuits have alleged that GPT-4o has encouraged teens to end their lives. The model was still described as sycophancy-prone when it was removed from ChatGPT in February 2026. === Removal with GPT-5 === On August 7, 2025, OpenAI released GPT-5. Its release was criticized as, with it, legacy GPT models were no longer available via ChatGPT, including GPT-4o, except for Pro users. Some users were particularly frustrated over this removal without prior warning because they used different GPT models for distinct purposes and found that GPT-5's router system left them with less control. In addition, some users preferred GPT-4o's warmer and more personal tone over that of GPT-5, which they described as "flat", "uncreative" and "lobotomized", and resembling an "overworked secretary". As a response, in a post on X, Sam Altman said that OpenAI would bring back the option to select GPT-4o to Plus users as well, and "[w]e [OpenAI] will watch usage as we think about how long to offer legacy models for." He also stated: "We for sure underestimated how much some of the things that people like in GPT-4o matter to them, even if GPT-5 performs better in most ways". "Long-term, this has reinforced that we really need good ways for different users to customize things (we understand that there isn't one model that works for everyone, and we have been investing in steerability research and launched a research preview of different personalities)". On August 13, 2025, Altman wrote on X that OpenAI is working on GPT-5's personality to make the model "feel warmer". The model was removed from ChatGPT on February 13, 2026. This caused new backlash from users that had grown attached to its personality and felt its creative writing abilities and understanding of nuance were irreplaceable. On social media, some users launched the movement "#Keep4o". A research paper highlighted the plea "Please, don’t kill the only model that still feels human". The model was removed the day before Valentine's Day, and some users had romantic relationships with GPT-4o.

  Read more →
• 

  Resilience (mathematics)

  In mathematical modeling, resilience refers to the ability of a dynamical system to recover from perturbations and return to its original stable steady state. It is a measure of the stability and robustness of a system in the face of changes or disturbances. If a system is not resilient enough, it is more susceptible to perturbations and can more easily undergo a critical transition. A common analogy used to explain the concept of resilience of an equilibrium is one of a ball in a valley. A resilient steady state corresponds to a ball in a deep valley, so any push or perturbation will very quickly lead the ball to return to the resting point where it started. On the other hand, a less resilient steady state corresponds to a ball in a shallow valley, so the ball will take a much longer time to return to the equilibrium after a perturbation. The concept of resilience is particularly useful in systems that exhibit tipping points, whose study has a long history that can be traced back to catastrophe theory. While this theory was initially overhyped and fell out of favor, its mathematical foundation remains strong and is now recognized as relevant to many different systems. == History == In 1973, Canadian ecologist C. S. Holling proposed a definition of resilience in the context of ecological systems. According to Holling, resilience is "a measure of the persistence of systems and of their ability to absorb change and disturbance and still maintain the same relationships between populations or state variables". Holling distinguished two types of resilience: engineering resilience and ecological resilience. Engineering resilience refers to the ability of a system to return to its original state after a disturbance, such as a bridge that can be repaired after an earthquake. Ecological resilience, on the other hand, refers to the ability of a system to maintain its identity and function despite a disturbance, such as a forest that can regenerate after a wildfire while maintaining its biodiversity and ecosystem services. With time, the once well-defined and unambiguous concept of resilience has experienced a gradual erosion of its clarity, becoming more vague and closer to an umbrella term than a specific concrete measure. == Definition == Mathematically, resilience can be approximated by the inverse of the return time to an equilibrium given by resilience ≡ − Re ( λ 1 ( A ) ) {\displaystyle {\text{resilience}}\equiv -{\text{Re}}(\lambda _{1}({\textbf {A}}))} where λ 1 {\textstyle \lambda _{1}} is the maximum eigenvalue of matrix A {\textstyle {\textbf {A}}} . The largest this value is, the faster a system returns to the original stable steady state, or in other words, the faster the perturbations decay. == Applications and examples == In ecology, resilience might refer to the ability of the ecosystem to recover from disturbances such as fires, droughts, or the introduction of invasive species. A resilient ecosystem would be one that is able to adapt to these changes and continue functioning, while a less resilient ecosystem might experience irreversible damage or collapse. The exact definition of resilience has remained vague for practical matters, which has led to a slow and proper application of its insights for management of ecosystems. In epidemiology, resilience may refer to the ability of a healthy community to recover from the introduction of infected individuals. That is, a resilient system is more likely to remain at the disease-free equilibrium after the invasion of a new infection. Some stable systems exhibit critical slowing down where, as they approach a basic reproduction number of 1, their resilience decreases, hence taking a longer time to return to the disease-free steady state. Resilience is an important concept in the study of complex systems, where there are many interacting components that can affect each other in unpredictable ways. Mathematical models can be used to explore the resilience of such systems and to identify strategies for improving their resilience in the face of environmental or other changes. For example, when modelling networks it is often important to be able to quantify network resilience, or network robustness, to the loss of nodes. Scale-free networks are particularly resilient since most of their nodes have few links. This means that if some nodes are randomly removed, it is more likely that the nodes with fewer connections are taken out, thus preserving the key properties of the network.

  Read more →
• 

  General-Purpose AI Code of Practice

  The General-Purpose AI Code of Practice (GPAI CoP) is a compliance tool released by the European Commission on 10 July 2025 to support compliance with the European Union Artificial Intelligence Act (AI Act). It provides operational guidance for providers of general-purpose AI models, particularly in relation to Articles 53 and 55 of the AI Act, which entered into application on 2 August 2025. The Code is organised into three chapters (Transparency, Copyright, and Safety and Security) and outlines how providers can meet the Act's relevant obligations. Although non-binding, providers can rely on adherence to the Code, meaning that EU regulators will assume that providers following the Code meet the corresponding legal requirements of the AI Act. As such, signatories to the Code will benefit from reduced administrative burdens and increased legal certainty compared to providers that prove compliance in other ways. While adherence to the Code is voluntary, compliance with the AI Act is not. == Background == The EU AI Act, adopted in 2024, established a risk-based regulatory regime for artificial intelligence in the European Union. The rationale for the GPAI CoP stems from Article 56 of the AI Act, which empowers the EU AI Office to develop a voluntary rulebook to guide how AI model providers can meet their legal obligations – specifically those found in Articles 53 and 55. Under Articles 53 and 55, developers of general-purpose AI models whose training compute exceeds 1023 floating-point operations (FLOPs) and that are placed on the EU market must meet transparency obligations and put in place a policy for EU copyright law. Models trained with more than 1025 FLOPs are classified as presenting systemic risk and are subject to enhanced safety requirements. The Commission may also designate a model as presenting systemic risk if it has equivalent impact or capabilities (Annex XIII criteria), even below that compute figure. Because the AI Act is relatively vague on how model providers should implement these requirements, the Code is meant to help by detailing processes and practices for compliance. == Drafting process == The development of the GPAI CoP was drawn up by 13 independent experts and involved four thematic working groups: Transparency & Copyright, Risk assessment for systemic risk, Technical risk mitigation for systemic risk, and Governance risk mitigation for systemic risk. Each group was coordinated by the European Union Artificial Intelligence Office (EU AI Office), drawing on contributions from nearly 1,000 stakeholders, including AI developers, academics, civil society organisations, national authorities, and international observers. The Code underwent three earlier iterations in November 2024, December 2024, and March 2025, before the final version was published on 10 July 2025, more than two months later than initially planned. The GPAI CoP will likely be updated continuously by the EU AI Office, alongside other tools such as the training data summary template. == Signatories == Among U.S.-based technology companies, Amazon, Anthropic, Google, IBM, Microsoft, and OpenAI have signed the GPAI CoP. xAI, founded by Elon Musk, has signed only one of the three chapters, namely the safety and security chapter. Prominent European AI companies that have signed include Aleph Alpha and Mistral AI. The European Commission maintains an updated list of signatories. As of January 2026, Meta is the most notable company that has declined to sign the Code. Major Chinese AI companies, such as Alibaba, Baidu or Deepseek, have also not signed. Providers that do not sign the GPAI CoP will still have to adhere to the binding requirements of the EU AI Act. The European Commission has indicated that it may take tougher action against companies that didn't sign the Code. == Transparency and Copyright chapters == The first two chapters of the GPAI CoP address transparency and copyright compliance and apply to all GPAI providers. They offer a way to demonstrate compliance with their obligations under Article 53 AI Act. The Transparency chapter addresses the documentation of a model's capabilities, limitations, and points of contact, and expects providers to make key documentation available to downstream providers. Signatories must also publish summaries of the content used to train their models. In the Copyright chapter, Signatories commit to follow a policy that aligns with EU copyright law. For example, they commit to mitigating the risk of copyright-infringing output. == Safety and Security chapter == The Safety and Security chapter is the most extensive chapter of the Code, and it applies to GPAI models with systemic risk, meaning it's only relevant to the small number of providers of the most advanced models. It specifies how Signatories commit to meeting Article 55(1) obligations to: Conduct model evaluations to identify systemic risks Assess and mitigate those risks Track and report serious incidents Ensure the cyber and physical security of their models The chapter outlines a comprehensive risk management process that must be applied before major deployment decisions, such as releasing a new systemic-risk GPAI model in the EU market, or substantially updating an existing one. Signatories commit to identifying systemic risks of their model, analysing and evaluating them, determining whether risk levels are acceptable, and implementing mitigation measures if necessary. This process should be repeated until models achieve an acceptable level of risk across all identified risks. === Risk identification === Signatories commit to analysing and evaluating at least four “specified” categories of systemic risk: CBRN (chemical, biological, radiological, and nuclear) Loss of control Cyber offence Harmful manipulation They are also expected to identify other systemic risks to public health, safety, and fundamental rights. The Code instructs providers to consider model capabilities, propensities, and affordances in this identification. Signatories commit to developing risk scenarios illustrating how identified risks could materialise in real-world conditions. === Risk analysis and risk evaluation === After identifying potential systemic risks, Signatories commit to analysing and evaluating the risks in order to determine whether they are acceptable or not, drawing on scientific literature, training data analysis, incident databases, expert consultation, and other sources. They also commit to conducting state-of-the-art model evaluations such as benchmarking, red teaming, and human uplift studies, targeting each risk. The risk analysis process is interconnected: insights from risk modelling should inform model evaluation design, while post-market monitoring should feed back into ongoing analysis. Signatories commit to ultimately estimating the likelihood and severity of each systemic risk. ==== Independent external model evaluations ==== Appendix 3.5 of the Safety and Security chapter requires signatories to ensure that independent external evaluators conduct model evaluations. Signatories may claim an exemption from this requirement only if they can demonstrate that their model is “similarly safe” to another model that has already been shown to comply with the Code, or if they are unable to appoint an appropriately qualified evaluator. The determination of “similarly safe” is based on comparable performance on benchmarks and the similarity of other model characteristics, such as their architecture. The CoP acknowledges that this kind of information is typically available only for models by the same provider, or potentially for open-weights or open-source models. === Risk acceptance criteria === The Code requires providers to compare estimated risks against predefined acceptance criteria, which must be measurable, based on model capabilities, and defined preemptively. While providers get to determine the level of risk they deem acceptable themselves, the pre-defined criteria and acceptance thresholds ensure providers cannot adjust their level of tolerance flexibly ahead of deployment decisions. Only if all risks are below acceptable levels should a model be deployed. === Continuous risk management and governance === The Code mandates ongoing risk management throughout the model lifecycle, including light-touch evaluations, continuous mitigation, post-market monitoring, and incident tracking and reporting. It further requires organisational governance structures assigning responsibility for risk management and expects providers to promote a “healthy risk culture,” including informing employees about the whistleblower protection policy, allowing internal challenges of decisions concerning systemic risk management, and committing to not retaliating against employees who disclose concerns about systemic risks to oversight authorities. === Documentation and transparency === Signatories commit to creating two types of documentation: Safety and Security Frame

  Read more →