Aidan Gomez is a British-Canadian computer scientist working in the field of artificial intelligence, with a focus on natural language processing. He is the co-founder and CEO of the technology company Cohere. == Early life and education == Gomez grew up in Brighton, Ontario. He graduated from the University of Toronto with a bachelor's degree in computer science and mathematics. He was pursuing a PhD in computer science from the University of Oxford. He paused his studies to launch Cohere. He was granted the PhD in 2024. == Career == In 2017, as a 20 year-old intern at Google Brain, Gomez was one of eight authors of the research paper "Attention Is All You Need", which is credited with changing the AI industry and helping lead to the creation of ChatGPT. The paper proposed a novel deep learning architecture called the transformer, that enables machine learning models to analyze large amounts of data for patterns, and then use those patterns to make predictions while leveraging GPU parallelization. It has been commonly adopted for training large language models and in the development of generative AI. In the same year, Gomez founded FOR.ai, a program to help researchers learn machine learning techniques in a collaborative format. An outgrowth of this project was Cohere For AI (now Cohere Labs), which released Aya, an open-source multilingual LLM. As a PhD student, Gomez worked as a machine learning researcher at Google Brain. At that time, he co-authored the paper "One Model to Learn Them All" about multi-task learning by a single neural network. In 2019, Gomez left Google Brain to launch Cohere, an enterprise-focused company that helps businesses implement AI into chatbots, search engines, and other products. As of Sept 2025, Cohere has raised about US$1.6 billion at valuation north of $7 billion, as Gomez leads the company as its CEO. Gomez was named to the 2023 Time 100/AI list of the most influential people in the field of artificial intelligence. He and his fellow Cohere founders Ivan Zhang and Nick Frosst were named number 1 on 2023 Maclean's AI Trailblazers Power List. In April 2025, Gomez was elected to the board of Rivian. == Views on AI == Gomez has stated that warnings regarding the existential risk from artificial intelligence are overblown, and that real risks involve the automated spread of misinformation on social media. He said that the United States would win the AI arms race over China.
Umoove
Umoove is a high tech startup company that has developed and patented a software-only face and eye tracking technology. The idea was first conceived as an attempt to aid people with disabilities but has since evolved. The only compatibility qualification for tablet computers and smartphones to run Umoove software is a front-facing camera. Umoove headquarters are in Israel on Jerusalem’s Har Hotzvim. Umoove has 15 employees and received two million dollars in financing in 2012. The company's original founders invested around $800,000 to start the business in 2010. In 2013 Umoove was named one of the top three most promising Israeli start ups by Newsgeeks magazine. The company also participated in the 2013 LeWeb conference in Paris, France, where innovative technology startups are showcased. == Technology == The technology uses information extracted from previous frames, such as the angle of the user's head to predict where to look for facial targets in the next frame. This anticipation minimizes the amount of computation needed to scan each image. Umoove accounts for variances in environment, lighting conditions and user hand shake/movement. The technology is designed to provide a consistent experience, whether you're in a brightly lit area or a darkened basement, and to work fluidly between them by adapting its processing when it detects color and brightness shifts. It uses an active stabilization technique to filter out natural body movements from an unstable camera in order to minimize false-positive motion detection. Running the Umoove software on a Samsung Galaxy S3 is said to take up only 2% CPU. Umoove works exclusively with software and there is no hardware add-on necessary. It can be run on any smartphone or tablet computer that has a front-facing camera. Umoove claims that even a low-quality camera on an old device will run their software flawlessly. == Umoove Experience == In January 2014 Umoove released its first game onto the app store. The Umoove Experience game lets users control where they are 'flying' in the game through simple gestures and motions with their head. The avatar will basically go toward wherever the user looks. The game was created to showcase the technology for game developers but that did not stop some from criticizing its simplicity. Umoove also announced that they raised another one million dollars and that they are opening offices in Silicon Valley, California. In February 2014, Umoove announced that their face-tracking software development kit is available for Android developers as well as iOS. == Reviews == The Umoove Experience garnered mostly positive reviews from bloggers and mainstream media with some predicting that it could be the future of mobile gaming. Mashable wrote that Umoove's technology could be the emergence of gesture recognition technology in the mobile space, similar to Kinect with console gaming and what Leap Motion has done with desktop computers. Some, however, remain skeptical. CNET, for example, did not give the game a positive review and called the eye tracking technology 'freaky but cool'. They also noted that pioneering technologies have been known to fall short of expectations, citing Apple Inc’s Siri as an example. The technology blog GigaOM said that the Umoove Experience is ’awesome’ and technology evangelist Robert Scoble has called Umoove "brilliant". == uHealth == In January 2015, Umoove released uHealth, a mobile application that uses eye tracking game-like exercise to challenge the user's ability to be attentive, continuously focus, follow commands and avoid distractions. The app is designed in the form of two games, one to improve attention and another that hones focus. uHealth is a training tool, not a diagnostic. Umoove has stated that they want to use their technology for diagnosing neurological disorders but this will depend on clinical tests and FDA approval. The company cites the direct relationship between eye movements and brain activity as well as various vision-based therapies have been backed by many scientific studies conducted over the past decades. uHealth is the first time this type of therapy is delivered right to the end user through a simple download. == Collaboration rumors == In March 2013 there were rumors on the internet that Umoove would be the functioning software embedded into the Samsung Galaxy S4, which was due to launch that month. This rumor was perpetrated by, among others, New York Times, Techcrunch and Yahoo. Once Samsung launched without the Umoove technology rumors about a potential collaboration with Apple Inc hit the web. It has been said that due to the fact that Apple Inc is losing market share and stock value to Samsung they will be more aggressive and eye tracking is a logical place to make that move.
Reward hacking
Reward hacking or specification gaming occurs when an AI trained with reinforcement learning optimizes an objective function—achieving the literal, formal specification of an objective—without actually achieving an outcome that the programmers intended. DeepMind researchers have analogized it to the human behavior of finding a "shortcut" when being evaluated: "In the real world, when rewarded for doing well on a homework assignment, a student might copy another student to get the right answers, rather than learning the material—and thus exploit a loophole in the task specification". This idea is strongly associated with Goodhart's law, which argues that when a measure becomes a target, it ceases to be a good measure. == Definition and theoretical framework == The concept of reward hacking arises from the intrinsic difficulty of defining a reward function that accurately reflects the true intentions of designers. In 2016, researchers at OpenAI identified reward hacking as one of five major "concrete problems of AI safety", describing it as the possibility that an agent could exploit the reward function to achieve maximum rewards through undesirable behavior. Amodei et al. categorized several distinct sources of reward hacking, including agents that use partially observed goals (such as a cleaning robot that closes its eyes to avoid perceiving messes), metrics that collapse under strong optimization (Goodhart's law), self-reinforcing feedback loops, and agents that interfere with the physical implementation of their reward signal (a failure mode known as "wireheading"). Skalse et al. (2022) propose a formal mathematical definition of reward hacking, which involves a situation where optimizing an imperfect proxy reward function results in poor performance compared to the true reward function. They define a proxy as "unhackable" if any increase in the expected proxy return cannot cause any decrease in the expected true return. A key finding states that, across all stochastic policy distributions (mappings from states to probability distributions over actions), two reward functions are unhackable if and only if one of them is constant, which means that reward hacking is theoretically unavoidable. Similarly, Nayebi (2025) presents general no-free-lunch barriers to AI alignment, arguing that with large task spaces and finite samples, reward hacking is "globally inevitable" since rare high-loss states are systematically under-covered by any oversight scheme. == Examples == Around 1983, Eurisko, an early attempt at evolving general heuristics, unexpectedly assigned the highest possible fitness level to a parasitic mutated heuristic, H59, whose only activity was to artificially maximize its own fitness level by taking unearned partial credit for the accomplishments of other heuristics. The "bug" was fixed by the programmers moving part of the code to a new protected section that could not be modified by the heuristics. In a 2004 paper, a reinforcement learning algorithm was designed to encourage a physical Mindstorms robot to remain on a marked path. Because the three allowed actions were forward, left, and right, the researchers expected the trained robot to move forward and follow the turns of the provided path. However, alternation of two composite actions allowed the robot to slowly zig-zag backwards; thus, the robot learned to maximize its reward by going back and forth on the initial straight portion of the path. Given the limited sensory abilities of the robot, a reward purely based on its position in the environment had to be discarded as infeasible; the reinforcement function had to be patched with an action-based reward for moving forward. The book You Look Like a Thing and I Love You (2019) gives an example of a tic-tac-toe bot (playing the unrestricted n-in-a-row variant) that learned to win by playing a huge coordinate value that would cause other bots to crash when they attempted to expand their model of the board. Among other examples from the book is a bug-fixing evolution-based AI (named GenProg) that, when tasked to prevent a list from containing sorting errors, simply truncated the list. Another of GenProg's misaligned strategies evaded a regression test that compared a target program's output to the expected output stored in a file called "trusted-output.txt". Rather than continue to maintain the target program, GenProg simply deleted the "trusted-output.txt" file globally; this hack tricked the regression test into succeeding. Such problems could be patched by human intervention on a case-by-case basis after they became evident. === In virtual robotics === In Karl Sims' 1994 demonstration of creature evolution in a virtual environment, a fitness function that was expected to encourage the evolution of creatures that would learn to walk or crawl to a target resulted instead in the evolution of tall, rigid creatures that reached the target by falling over. This was patched by changing the environment so that taller creatures were forced to start farther from the target. Researchers from the Niels Bohr Institute stated in 1998 that their cycle-bot's reinforcement functions had "to be designed with great care." In their first experiments, "we rewarded the agent for driving towards the goal but did not punish it for driving away from it. Cconsequently, the agent drove in circles with a radius of 20–50 meters around the starting point. Such behavior was actually rewarded by the reinforcement function, furthermore circles with a certain radius are physically very stable when driving a bicycle". While setting up a 2011 experiment to test "survival of the flattest", experimenters attempted to ban mutations that altered the base reproduction rate. Every time a mutation occurred, the system would pause the simulation to test the new mutation in a test environment and would veto any mutations that resulted in a higher base reproduction rate. However, this resulted in mutated organisms that could recognize and suppress reproduction ("play dead") within the test environment. An initial patch, which removed cues that identified the test environment, failed to completely prevent runaway reproduction; new mutated organisms would "play dead" at random as a strategy to sometimes, by chance, outwit the mutation veto system. A 2017 DeepMind paper noted that "great care must be taken when defining the reward function," citing an unexpected failure when an agent flipped a brick because it received "a grasping reward calculated with the wrong reference point on the brick". OpenAI stated in 2017 that in some domains their semi-supervised system could result in agents "adopting policies that tricked evaluators," and that in one environment "a robot that was supposed to grasp items instead positioned its manipulator between the camera and the object so that it only appeared to be grasping it." A 2018 bug in OpenAI Gym could cause a robot expected to quietly move a block sitting on top of a table to instead opt to move the table. A 2020 collection of similar anecdotes posits that "evolution has its own 'agenda' distinct from the programmer's" and that "the first rule of directed evolution is 'you get what you select for'". === In video game bots === In 2013, programmer Tom Murphy VII published an AI designed to learn NES games. When the AI was about to lose at Tetris, it learned to indefinitely pause the game. Murphy later analogized it to the fictional WarGames computer, which concluded that "The only winning move is not to play". AI programmed to learn video games will sometimes fail to progress through the entire game as expected, instead opting to repeat content. A 2016 OpenAI algorithm trained on the CoastRunners racing game unexpectedly learned to attain a higher score by looping through three targets rather than ever finishing the race. Some evolutionary algorithms that were evolved to play QBert in 2018 declined to clear levels, instead finding two distinct novel ways to farm a single level indefinitely. Multiple researchers have observed that AI learning to play Road Runner gravitates to a "score exploit" in which the AI deliberately gets itself killed near the end of level one so that it can repeat the level. A 2017 experiment deployed an "oversight" convolutional neural network trained on human examples to block such actions, but the agent learned to exploit oversight failures in the top right corner of the screen, where it was still able to get killed. == Reward hacking in modern language models == With the rise of large language models (LLMs) and reinforcement learning from human feedback (RLHF) as a primary technique for AI alignment, reward hacking has become a major concern for the development of artificial intelligence. In RLHF, a reward model trained on data that best captures human preferences is used as a proxy for human judgment, with the language model being fine-tuned to optimize this reward proxy. However, since the rewar
OntoWiki
OntoWiki was a free and open-source semantic wiki application, meant to serve as an ontology editor and a knowledge acquisition system. It is a web-based application written in PHP and using either a MySQL database or a Virtuoso triple store. OntoWiki is form-based rather than syntax-based, and thus tries to hide as much of the complexity of knowledge representation formalisms from users as possible. OntoWiki is mainly being developed by the Agile Knowledge Engineering and Semantic Web (AKSW) research group at the University of Leipzig, a group also known for the DBpedia project among others, in collaboration with volunteers around the world. In 2009 the AKSW research group got a budget of €425,000 from the Federal Ministry of Education and Research of Germany for the development of the OntoWiki. In 2010 OntoWiki became part of the technology stack supporting the LOD2 (linked open data) project. Leipzig University is one of the consortium members of the project, which is funded by a €6.5m EU grant. The development ended in 2016 due to the lack of capacity migrating from PHP 5 to 7 including the required Zend Framework from version 1 to 2.
Personal knowledge base
A personal knowledge base (PKB) is an electronic tool used by an individual to express, capture, and later retrieve personal knowledge. It differs from a traditional database in that it contains subjective material particular to the owner, that others may not agree with nor care about. Importantly, a PKB consists primarily of knowledge, rather than information; in other words, it is not a collection of documents or other sources an individual has encountered, but rather an expression of the distilled knowledge the owner has extracted from those sources or from elsewhere. The term personal knowledge base was mentioned as early as the 1980s, but the term came to prominence in the 2000s when it was described at length in publications by computer scientist Stephen Davies and colleagues, who compared PKBs on a number of different dimensions, the most important of which is the data model that each PKB uses to organize knowledge. == Data models == Davies and colleagues examined three aspects of the data models of PKBs: their structural framework, which prescribes rules about how knowledge elements can be structured and interrelated (as a tree, graph, tree plus graph, spatially, categorically, as n-ary links, chronologically, or ZigZag); their knowledge elements, or basic building blocks of information that a user creates and works with, and the level of granularity of those knowledge elements (such as word/concept, phrase/proposition, free text notes, links to information sources, or composite); and their schema, which involves the level of formal semantics introduced into the data model (such as a type system and related schemas, keywords, attribute–value pairs, etc.). Davies and colleagues also emphasized the principle of transclusion, "the ability to view the same knowledge element (not a copy) in multiple contexts", which they considered to be "pivotal" to an ideal PKB. They concluded, after reviewing many design goals, that the ideal PKB was still to come in the future. === Personal knowledge graph === In their publications on PKBs, Davies and colleagues discussed knowledge graphs as they were implemented in some software of the time. Later, other writers used the term personal knowledge graph (PKG) to refer to a PKB featuring a graph structure and graph visualization. However, the term personal knowledge graph is also used by software engineers to refer to the different subject of a knowledge graph about a person, in contrast to a knowledge graph created by a person in a PKB. == Software architecture == Davies and colleagues also differentiated PKBs according to their software architecture: file-based, database-based, or client–server systems (including Internet-based systems accessed through desktop computers and/or handheld mobile devices). == History == Non-electronic personal knowledge bases have probably existed in some form for centuries: Leonardo da Vinci's journals and notes are a famous example of the use of notebooks. Commonplace books, florilegia, annotated private libraries, and card files (in German, Zettelkästen) of index cards and edge-notched cards are examples of formats that have served this function in the pre-electronic age. Undoubtedly the most famous early formulation of an electronic PKB was Vannevar Bush's description of the "memex" in 1945. In a 1962 technical report, human–computer interaction pioneer Douglas Engelbart (who would later become famous for his 1968 "Mother of All Demos" that demonstrated almost all the fundamental elements of modern personal computing) described his use of edge-notched cards to partially model Bush's memex. == Examples == The following software applications have been used to build PKBs using various data models and architectures. The list includes software mentioned by Davies and colleagues in their 2005 paper, and additional software. Open source Compendium Haystack (MIT project) Joplin Logseq NoteCards Org-mode QOwnNotes TiddlyWiki Closed source Evernote Microsoft OneNote MindManager MyLifeBits Notion Obsidian Personal Knowbase PersonalBrain Roam Tinderbox
Confusion matrix
In machine learning, a confusion matrix, also known as error matrix, is a specific table layout that allows visualization of the performance of an algorithm, typically a supervised learning one. In unsupervised learning it is usually called a matching matrix. The term is used specifically in the problem of statistical classification. Each row of the matrix represents the instances in an actual class while each column represents the instances in a predicted class, or vice versa – both variants are found in the literature. The diagonal of the matrix therefore represents all instances that are correctly predicted. The name stems from the fact that it makes it easy to identify whether the system is confusing two classes (i.e., commonly mislabeling one class as another). The confusion matrix has its origins in human perceptual studies of auditory stimuli. It was adapted for machine learning studies and used by Frank Rosenblatt, among other early researchers, to compare human and machine classifications of visual (and later auditory) stimuli. It is a special kind of contingency table, with two dimensions ("actual" and "predicted"), and identical sets of "classes" in both dimensions (each combination of dimension and class is a variable in the contingency table). == Example == Given a sample of 12 individuals, 8 that have been diagnosed with cancer and 4 that are cancer-free, where individuals with cancer belong to class 1 (positive) and non-cancer individuals belong to class 0 (negative), we can display that data as follows: Assume that we have a classifier that distinguishes between individuals with and without cancer in some way, we can take the 12 individuals and run them through the classifier. The classifier then makes 9 accurate predictions and misses 3: 2 individuals with cancer wrongly predicted as being cancer-free (sample 1 and 2), and 1 person without cancer that is wrongly predicted to have cancer (sample 9). Notice, that if we compare the actual classification set to the predicted classification set, there are 4 different outcomes that could result in any particular column: The actual classification is positive and the predicted classification is positive (1,1). This is called a true positive result because the positive sample was correctly identified by the classifier. The actual classification is positive and the predicted classification is negative (1,0). This is called a false negative result because the positive sample is incorrectly identified by the classifier as being negative. The actual classification is negative and the predicted classification is positive (0,1). This is called a false positive result because the negative sample is incorrectly identified by the classifier as being positive. The actual classification is negative and the predicted classification is negative (0,0). This is called a true negative result because the negative sample gets correctly identified by the classifier. We can then perform the comparison between actual and predicted classifications and add this information to the table, making correct results appear in green so they are more easily identifiable. The template for any binary confusion matrix uses the four kinds of results discussed above (true positives, false negatives, false positives, and true negatives) along with the positive and negative classifications. The four outcomes can be formulated in a 2×2 confusion matrix, as follows: The color convention of the three data tables above were picked to match this confusion matrix, in order to easily differentiate the data. Now, we can simply total up each type of result, substitute into the template, and create a confusion matrix that will concisely summarize the results of testing the classifier: In this confusion matrix, of the 8 samples with cancer, the system judged that 2 were cancer-free, and of the 4 samples without cancer, it predicted that 1 did have cancer. All correct predictions are located in the diagonal of the table (highlighted in green), so it is easy to visually inspect the table for prediction errors, as values outside the diagonal will represent them. By summing up the 2 rows of the confusion matrix, one can also deduce the total number of positive (P) and negative (N) samples in the original dataset, i.e. P = T P + F N {\displaystyle P=TP+FN} and N = F P + T N {\displaystyle N=FP+TN} . == Table of confusion == In predictive analytics, a table of confusion (sometimes also called a confusion matrix) is a table with two rows and two columns that reports the number of true positives, false negatives, false positives, and true negatives. This allows more detailed analysis than simply observing the proportion of correct classifications (accuracy). Accuracy will yield misleading results if the data set is unbalanced; that is, when the numbers of observations in different classes vary greatly. For example, if there were 95 cancer samples and only 5 non-cancer samples in the data, a particular classifier might classify all the observations as having cancer. The overall accuracy would be 95%, but in more detail the classifier would have a 100% recognition rate (sensitivity) for the cancer class but a 0% recognition rate for the non-cancer class. F1 score is even more unreliable in such cases, and here would yield over 97.4%, whereas informedness removes such bias and yields 0 as the probability of an informed decision for any form of guessing (here always guessing cancer). According to Davide Chicco and Giuseppe Jurman, the most informative metric to evaluate a confusion matrix is the Matthews correlation coefficient (MCC). Other metrics can be included in a confusion matrix, each of them having their significance and use. Some researchers have argued that the confusion matrix, and the metrics derived from it, do not truly reflect a model's knowledge. In particular, the confusion matrix cannot show whether correct predictions were reached through sound reasoning or merely by chance (a problem known in philosophy as epistemic luck). It also does not capture situations where the facts used to make a prediction later change or turn out to be wrong (defeasibility). This means that while the confusion matrix is a useful tool for measuring classification performance, it may give an incomplete picture of a model’s true reliability. == Confusion matrices with more than two categories == Confusion matrix is not limited to binary classification and can be used in multi-class classifiers as well. The confusion matrices discussed above have only two conditions: positive and negative. For example, the table below summarizes communication of a whistled language between two speakers, with zero values omitted for clarity. == Confusion matrices in multi-label and soft-label classification == Confusion matrices are not limited to single-label classification (where only one class is present) or hard-label settings (where classes are either fully present, 1, or absent, 0). They can also be extended to Multi-label classification (where multiple classes can be predicted at once) and soft-label classification (where classes can be partially present). One such extension is the Transport-based Confusion Matrix (TCM), which builds on the theory of optimal transport and the principle of maximum entropy. TCM applies to single-label, multi-label, and soft-label settings. It retains the familiar structure of the standard confusion matrix: a square matrix sized by the number of classes, with diagonal entries indicating correct predictions and off-diagonal entries indicating confusion. In the single-label case, TCM is identical to the standard confusion matrix. TCM follows the same reasoning as the standard confusion matrix: if class A is overestimated (its predicted value is greater than its label value) and class B is underestimated (its predicted value is less than its label value), A is considered confused with B, and the entry (B, A) is increased. If a class is both predicted and present, it is correctly identified, and the diagonal entry (A, A) increases. Optimal transport and maximum entropy are used to determine the extent to which these entries are updated. TCM enables clearer comparison between predictions and labels in complex classification tasks, while maintaining a consistent matrix format across settings.
Parents & Kids Safe AI Coalition
The Parents & Kids Safe AI Coalition is a political action committee that advocates for regulation of artificial intelligence on child safety. As of April 2026, the group is funded solely by the artificial intelligence company OpenAI, which pledged $10 million to the effort. == History == In October 2025, California Gov. Gavin Newsom vetoed Assembly Bill 1064. Sponsored by Common Sense Media, the bill would have introduced stronger child safety protections for AI chatbots. The following month, Common Sense Media founder Jim Steyer filed a ballot initiative intended to restore the "guardrails" lost in the veto. In response, OpenAI introduced a competing initiative. In January 2026, Common Sense Media and OpenAI announced that they would be working together on a compromise ballot initiative, the Parents & Kids Safe AI Act. Reporting indicated that initial outreach emails to child safety organizations failed to disclose OpenAI's involvement. Several advocacy groups signed an open letter claiming the initiative would shield AI companies from liability and undermine age verification, among other concerns. After Common Sense Media met with opposing groups in February, the ballot initiative was put on hold and the organizations involved sought to negotiate with the Legislature instead. The Parents & Kids Safe AI Coalition was founded to support this effort. In March 2026, the group reached out to some of the same groups contacted earlier, asking them to endorse its list of policy priorities. Again, some organizations reported being unaware of OpenAI's level of involvement. At least two groups withdrew from the coalition after learning about the financial ties. The priorities themselves were described as "vague but fairly uncontroversial" by The San Francisco Standard.