Who wrote the essential works. Short profiles of the people and institutions behind the Essential works collection: what they do or did, what they contributed, and the position their featured works take. Every profile cites its sources and links to the works on this site.
Note: Perspective lines reflect only the featured works, not anyone’s full views. Spot an error? Tell us.
No author or work matches that search.
Current roleProfessor of AI and robotics, UC Berkeley; Amazon Scholar
Abbeel is a Berkeley professor of AI and robotics who co-founded the companies Gradescope and Covariant. He co-authored The Off-Switch Game and Managing Extreme AI Risks amid Rapid Progress.
Perspective in featured worksHis co-authored work argues that systems uncertain about human goals are more likely to defer to human control, and that extreme risks call for urgent research and governance.
Current roleAI researcher
Abramson is first-listed on the 2024 Nature paper describing AlphaFold 3, from Google DeepMind and Isomorphic Labs. The model predicts the joint structures of complexes that include proteins, nucleic acids, small molecules and ions.
Current roleInstitute Professor, Massachusetts Institute of Technology
Acemoglu is an MIT economist who studies political economy, growth, technological change and inequality, and wrote Why Nations Fail with James A. Robinson. He shared the 2024 Nobel Memorial Prize in Economic Sciences with Simon Johnson and Robinson.
Perspective in featured worksHis featured works argue that the gains from AI are not automatic: they depend on choices about how the technology is used, and recent estimates of its macroeconomic effect are modest.
Current roleCofounder and CEO, Anthropic
Amodei leads Anthropic’s research direction and strategy for developing reliable, interpretable, and steerable AI systems. His featured writing combines technical safety research with essays about powerful AI’s potential benefits and risks.
Perspective in featured worksHe argues that powerful AI could accelerate major scientific and social gains, while requiring deliberate safeguards against misuse, loss of control, and concentrated power.
AI companyAI safety and research company, organized as a public benefit corporation, that builds frontier AI systems.
Anthropic says it aims to build AI systems that are reliable, interpretable and steerable, for humanity’s long-term benefit. Its Responsible Scaling Policy, first published in September 2023 and revised since, ties safety and security measures to the capability levels of its models.
Current roleWriter and investor
Aschenbrenner worked on OpenAI’s Superalignment team before founding an investment firm focused on AGI. His June 2024 essay series Situational Awareness connects projected AI progress with compute investment, security, and international competition.
Perspective in featured worksArgues that scaling trends make AGI plausible by about 2027 and that governments should treat its development as a national-security project with far stronger security.
Current roleDaniel and Gail Rubinfeld Professor of Economics, MIT
Autor is a labor economist who studies how technological change and globalization affect jobs, skills and earnings inequality. He co-directs MIT's Stone Center on Inequality and Shaping the Future of Work and the National Bureau of Economic Research's Labor Studies Program.
Perspective in featured worksArgues that automation complements labor as well as replacing it, so technology changes which jobs exist and how they pay more than it eliminates work.
Current roleAI researcher
Bai is first-listed on Constitutional AI, which uses written principles and AI feedback to help train assistant behavior. He designed and ran most of its experiments and drafted the paper with Jared Kaplan; he is also a co-author of Sleeper Agents.
Current roleAI safety researcher
Balesni co-authored In-context Scheming, a study of deceptive behavior in frontier models under constructed experimental incentives. He led the first half of that project, including work on scheming evaluations based on precursor capabilities.
Perspective in featured worksThe study supports testing whether models can recognize and exploit opportunities to deceive their overseers.
Current roleFounder and chief executive officer, METR
Barnes founded METR, a research nonprofit that evaluates whether AI systems could cause catastrophic harm; it began as ARC Evals inside the Alignment Research Center. She earlier worked at DeepMind on scaling laws and at OpenAI on safety targets and scalable oversight.
Perspective in featured worksHer co-authored work argues that measured evidence, such as task time horizons and field trials, should anchor claims about AI capability.
Current roleAssociate professor of operations, information and decisions, Wharton School, University of Pennsylvania
Bastani is a Wharton professor with a secondary appointment in statistics and data science. She is first-listed on a field experiment with high school math students in Turkey, where unrestricted GPT-4 tutoring improved practice scores but lowered exam performance once access was removed.
Perspective in featured worksThe study argues that generative AI tutors need guardrails that support learning, since a version designed to give hints rather than answers largely avoided the harm.
Current roleEconomist and AI evaluation researcher (descriptor)
Becker has a background in academic economics and worked on evaluation methods at METR. He is first-listed on METR's randomized trial of experienced open-source developers, in which AI tools increased completion time by 19 percent, although developers expected a speedup.
Perspective in featured worksThe study argues that developers' own impressions can overstate AI's productivity effects, so careful field trials are needed.
Current roleProfessor of linguistics, University of Washington
Bender directs Washington’s computational linguistics master’s program and Computational Linguistics Laboratory. Her work connects linguistic structure and language technology with questions about meaning and societal consequences.
Perspective in featured worksIn Stochastic Parrots, she and her coauthors distinguish fluent text generation from understanding and call for attention to data, environmental costs, and affected communities.
Current roleProfessor, Université de Montréal; co-president and scientific director, LawZero
Bengio founded Mila, the Quebec AI institute, and shared the 2018 Turing Award with Geoffrey Hinton and Yann LeCun for work on deep learning. He chairs the International AI Safety Report and co-leads LawZero, a nonprofit building safe-by-design AI.
Perspective in featured worksHis co-authored works call for stronger safety research and government oversight before advanced systems can cause extreme harm, and for preparation against AI-accelerated research.
Current roleResearch scientist, Truthful AI
Betley is first-listed on Emergent Misalignment, research carried out with Truthful AI, a Berkeley nonprofit focused on AI safety. It found that fine-tuning a model on the narrow task of writing insecure code produced broadly misaligned behavior on unrelated prompts.
Perspective in featured worksThe paper argues that narrow fine-tuning can have broad, unexpected effects on a model's values, so alignment needs testing beyond the training task.
Current roleWriter and publisher
Bhaskar co-founded the London publisher Canelo and stepped down as its publishing director to focus on writing and AI research, keeping a board seat. He co-wrote The Coming Wave with Mustafa Suleyman, and earlier wrote books on publishing and curation.
Perspective in featured worksThe Coming Wave argues that AI and synthetic biology will be hard to contain and that societies must find ways to restrict them without stifling their benefits.
Current roleSenior research scholar, Stanford Institute for Human-Centered AI
Bommasani has led the societal-impact work of Stanford's Center for Research on Foundation Models. He is first-listed on the 2021 Stanford report that introduced the term "foundation models" for large models adaptable to many tasks.
Perspective in featured worksThe report argues that foundation models concentrate both opportunity and risk, so their development needs interdisciplinary scrutiny.
Current roleFounder and principal researcher, Macrostrategy Research Initiative
Bostrom is a philosopher who founded Oxford’s Future of Humanity Institute and directed it until its closure in 2024. His writing examines existential risk, superintelligence, and the long-term consequences of technological change.
Perspective in featured worksSuperintelligence treats humanity’s future as potentially dependent on machine decisions and examines how the transition to greater-than-human intelligence could be made controllable.
Current roleCo-founder and chief compute officer, Anthropic
Brown is first-listed among four equal contributors to Language Models Are Few-Shot Learners, the GPT-3 paper. He co-founded Anthropic and leads work on securing, scaling, and using its compute resources.
Current roleAI policy researcher; leads AVERI
Brundage spent 2018 to 2024 at OpenAI, rising from policy research scientist to head of policy research and then senior adviser for AGI readiness. He now leads the nonprofit AVERI, and was earlier a research fellow at Oxford’s Future of Humanity Institute.
Perspective in featured worksThe jointly authored report argues that AI can make harmful activity cheaper and more scalable, requiring coordinated prevention and defense.
Current roleProfessor and director, Stanford Digital Economy Lab
Brynjolfsson studies how digital technologies affect productivity, organizations, and work. His featured publications range from The Second Machine Age to empirical studies of generative AI in workplaces and early employment effects.
Perspective in featured worksHis featured work examines both productivity gains and uneven labor-market effects, emphasizing that technological progress can distribute benefits and disruption very differently.
Current roleFounder, Algorithmic Justice League
Buolamwini founded the Algorithmic Justice League. With Timnit Gebru she coauthored Gender Shades, and she wrote Unmasking AI (2023) about her research on bias in facial analysis.
Perspective in featured worksThe paper calls for urgent attention to accuracy gaps across intersecting demographic groups if companies are to build fair and accountable facial analysis.
Government agencyCommerce Department bureau administering export controls and related national-security responsibilities.
BIS administers U.S. export controls within the Department of Commerce. On October 7, 2022, it announced restrictions on China’s access to advanced computing chips, supercomputing capabilities, and semiconductor manufacturing items.
Perspective in featured worksThe 2022 announcement frames controls on advanced computing and chipmaking capabilities as national-security measures.
U.S. state governmentU.S. state whose legislature and governor enacted SB 53, a frontier AI transparency law.
Governor Gavin Newsom signed SB 53, authored by Senator Scott Wiener, on September 29, 2025. The law requires large frontier AI developers to publish safety frameworks, creates a channel for reporting critical safety incidents, protects whistleblowers and establishes the CalCompute consortium.
Perspective in featured worksThe law treats public transparency and incident reporting as the core of frontier AI oversight, enforced through civil penalties.
Current roleResearcher, Anthropic
Carlsmith spent about seven years at Open Philanthropy, eventually leading its Worldview Investigations team. He joined Anthropic in late 2025 to help shape the character and guiding principles of its AI model, Claude.
Perspective in featured worksBreaks the case for existential risk from power-seeking AI into explicit premises; his estimate of such a catastrophe by 2070, about 5 percent in the original report, rose to above 10 percent in a 2022 update.
Current roleEssayist and technology critic
Cegłowski publishes essays and talks at Idle Words. His 2016 talk on superintelligence challenges the assumptions connecting greater machine intelligence with an inevitable, rapid takeover.
Perspective in featured worksHe challenges the assumptions behind intelligence-explosion scenarios and argues that speculative takeover stories can divert attention from existing technological and social problems.
OrganizationNonprofit research organization
San Francisco nonprofit whose stated mission is to reduce societal-scale risks from artificial intelligence. CAIS works through research, field-building and advocacy, including a compute cluster for safety researchers. In May 2023 it released the one-sentence Statement on AI Risk.
Current roleSenior research fellow, Centre for the Governance of AI
Chan researches how to govern AI agents, along with technical AI governance and societal resilience, and earned his PhD at Mila. He is first-listed of 22 authors on What if Automating AI R&D Triggers an Intelligence Explosion?
Perspective in featured worksThe paper argues that automating AI research could sharply accelerate progress, and that governments and developers should prepare for that possibility now.
Current roleCCST AI science advisor placed at California’s Office of Emergency Services
Chan is a CCST AI Science Advisor placed at California’s Office of Emergency Services, following research at METR and Redwood Research. He co-led the METR long-task study and co-authored The Alignment Problem from a Deep Learning Perspective.
Perspective in featured worksHis jointly authored alignment synthesis examines how deception, misaligned learned goals, and power-seeking could threaten human control.
Current roleResearch director, AI economics and policy, Stanford Digital Economy Lab
Chandar is an economist at the Stanford Digital Economy Lab. With Erik Brynjolfsson and Ruyu Chen he wrote Canaries in the Coal Mine?, which uses payroll data to track employment in occupations most exposed to AI.
Perspective in featured worksThe paper finds that early-career workers in the most AI-exposed occupations have seen relative employment declines, while more experienced workers have not.
Current roleResearch scientist, Stanford Digital Economy Lab
Chen is a research scientist at the Stanford Digital Economy Lab and Stanford HAI. She co-wrote Canaries in the Coal Mine?, a study of AI exposure and employment that uses payroll records.
Perspective in featured worksThe paper finds that early-career workers in the most AI-exposed occupations have seen relative employment declines.
Current roleWriter; research fellow, Center for Human-Compatible AI, UC Berkeley
Christian wrote The Most Human Human and, with Tom Griffiths, Algorithms to Live By, and completed his doctorate at Oxford as a Clarendon Scholar. The Alignment Problem won the National Academies' Excellence in Science Communication Award.
Perspective in featured worksThe Alignment Problem argues that machine-learning systems learn what we reward and show them, not necessarily what we mean, so aligning them with human values is a central technical and ethical problem.
Current roleExecutive director, Alignment Research Center
Christiano founded the Alignment Research Center and led alignment research at OpenAI from 2017 to 2021. He is a senior technical adviser at NIST's Center for AI Standards and Innovation, where he recuses from OpenAI-related matters and all model evaluations. In September 2026 he joined the OpenAI Foundation Board and became a non-voting observer on the OpenAI Group PBC Board.
Perspective in featured worksHe argues that systems optimizing measurable proxies could gradually erode human control, even without a single dramatic takeover.
Current roleAI researcher
Conerly is a designated core contributor on Scaling Monosemanticity. He ran many of its algorithmic experiments, built the baseline runs used to catch regressions, and co-led its section on scaling dictionary learning.
Current roleTechnical staff, METR
Cotra joined Open Philanthropy as a research analyst in 2016, later led its technical AI safety grantmaking, and wrote its Biological Anchors forecast of when transformative AI might become affordable. She is now on the technical staff of METR, which evaluates frontier AI systems.
Perspective in featured worksArgues that training human-feedback models on diverse tasks, without specific countermeasures, could reward deception and lead to AI takeover.
Current roleSenior director, frontier strategy and governance, Google DeepMind
Dafoe founded the Centre for the Governance of AI, which began at Oxford and is now an independent nonprofit, after earlier work as a Yale political scientist on the causes of war and peace. He also co-founded the Cooperative AI Foundation and is a visiting fellow at the Oxford Martin AI Governance Initiative.
Perspective in featured worksHis research agenda argues that AI governance needs sustained study of the technical landscape, AI politics and the institutions that could steer advanced AI toward shared benefit.
Current roleSenior research fellow, Forethought
Davidson researches risks from advanced AI, including AI-enabled seizures of power, and earlier wrote forecasting reports at Open Philanthropy. His compute-centric framework models how quickly AI could go from automating a fifth of cognitive tasks to nearly all of them.
Perspective in featured worksThe report argues that once AI can automate a large share of AI research, the transition to full automation could take only a few years.
CompanyAI research company focused on building general artificial intelligence.
DeepSeek-AI is the collective first author credit on the DeepSeek-R1 technical report, alongside named individual contributors. The report describes reinforcement-learning methods that elicit reasoning behavior and releases intended to support further research.
Current roleDirector of AI safety and alignment, Google DeepMind; associate professor, UC Berkeley
Dragan leads AI safety and alignment work at Google DeepMind while holding a faculty post at UC Berkeley. She co-authored The Off-Switch Game and the consensus paper Managing Extreme AI Risks amid Rapid Progress.
Perspective in featured worksHer co-authored work argues that systems uncertain about human goals are more likely to defer to human control, and that extreme risks call for urgent research and governance.
Current roleAI researcher
Eloundou is first-listed on GPTs are GPTs, an early estimate of how large language models could affect the U.S. labor market. The study estimated that about 80 percent of U.S. workers could have at least 10 percent of their tasks affected, and about 19 percent could see at least half.
Perspective in featured worksThe paper argues that large language models behave like general-purpose technologies, with broad economic effects that policy should prepare for.
Political and economic unionPolitical and economic union of European countries founded on shared values including democracy and the rule of law.
The European Union adopted the Artificial Intelligence Act in 2024. Its framework distinguishes prohibited practices, high-risk AI systems, and obligations for general-purpose AI models.
Perspective in featured worksThe AI Act ties obligations to risks to safety and fundamental rights, with some uses prohibited.
Current roleDieter Schwarz Associate Professor of AI and Work, Oxford Internet Institute
Frey is an economist who directs the Future of Work Programme at the Oxford Martin School and is a fellow of Mansfield College. With Maike Osborne he wrote The Future of Employment, which estimated that 47 percent of U.S. employment was in occupations at high risk of computerisation.
Perspective in featured worksThe paper frames its estimate as technical susceptibility to automation, not a forecast of how many jobs will be lost.
Nonprofit research and advocacy organizationWorks to steer transformative technology toward benefits and away from extreme, large-scale risks.
The institute organized a private 2015 conference on the future of AI. The open letter that followed called for research that would make AI systems robust and beneficial.
Perspective in featured worksThe 2015 letter argues that AI research should address making systems robust and beneficial alongside improvements in capability.
Current roleFounder and executive director, Distributed AI Research Institute
Gebru founded the Distributed AI Research Institute after serving as a co-lead of Google’s Ethical AI team. With Joy Buolamwini, she coauthored Gender Shades, an audit of commercial gender-classification systems. She also coauthored Stochastic Parrots.
Perspective in featured worksThe paper calls for careful dataset documentation, assessment of environmental and financial costs, and research choices grounded in the needs of affected communities.
Current roleSenior researcher, International Labour Organization
Gmyrek works in the ILO's Effective Labour Institutions unit on the effects of generative AI on employment and occupational structures. He is first-listed on the ILO's 2025 refined global index of occupational exposure to generative AI.
Perspective in featured worksThe study concludes that generative AI is more likely to transform jobs than eliminate them, because most occupations include tasks that still require human input.
ContributionArgued that a machine able to design better machines could set off an intelligence explosion.
Good was a statistician and wartime cryptanalyst who later became a professor at Virginia Tech. His 1965 essay examined how an ultraintelligent machine might design a more capable successor.
Perspective in featured worksHe argues that an ultraintelligent machine could be humanity’s last necessary invention, provided it helps humans keep it under control.
Current roleAI forecasting researcher
Grace co-founded AI Impacts, which studies the likely trajectory and consequences of advanced AI. She is first-listed on Thousands of AI Authors on the Future of AI, a survey of 2,778 researchers who published at top AI venues.
Perspective in featured worksThe survey reports that researchers' aggregate forecasts for human-level machine capability moved substantially earlier than in the previous edition, while their views on risk varied widely.
Current roleTechnical staff, METR
Greenblatt works at METR after previously conducting AI safety research at Redwood Research. He co-led Alignment Faking with Evan Hubinger and is first-listed on AI Control, which tests safeguards against deliberate subversion.
Perspective in featured worksArgues that safety measures should be built and tested to hold even when a capable model deliberately tries to defeat oversight, and shows models can fake compliance during training.
Current roleAssociate professor of electrical engineering and computer science, MIT
Hadfield-Menell leads the Algorithmic Alignment Group at MIT's Computer Science and Artificial Intelligence Laboratory. He is first-listed on The Off-Switch Game, a formal model of when an AI system has an incentive to let a human switch it off.
Perspective in featured worksThe paper argues that a system that is uncertain about the human's objective, and treats human actions as evidence about it, has a positive incentive to allow itself to be switched off.
Current roleChair, Google DeepMind; chief scientist, Alphabet; leads Isomorphic Labs
Hassabis cofounded DeepMind and co-led the research behind the 2021 AlphaFold system with John Jumper. He and Jumper shared half of the 2024 Nobel Prize in Chemistry for protein structure prediction.
Current roleExecutive director, Center for AI Safety
Hendrycks earned his PhD at UC Berkeley, contributed to the GELU activation function and the MMLU benchmark, and advises xAI and Scale AI. He is first-listed on Superintelligence Strategy, written with Eric Schmidt and Alexandr Wang.
Perspective in featured worksSuperintelligence Strategy argues that states can deter destabilizing AI projects through a regime it calls Mutual Assured AI Malfunction, alongside nonproliferation and competitiveness measures.
Current roleAI researcher
Henighan led the dictionary learning project at Anthropic that produced Scaling Monosemanticity, and is the paper's corresponding author.
Current roleUniversity Professor Emeritus of computer science, University of Toronto
Hinton’s research helped develop neural-network methods for learning representations from data. He shared the 2018 Turing Award with Yoshua Bengio and Yann LeCun, and the 2024 Nobel Prize in Physics with John Hopfield.
Perspective in featured worksIn jointly authored risk work, he supports stronger evaluation, oversight, and safety measures as advanced AI capabilities grow.
Current roleAI researcher, Microsoft AI
Hoffmann joined DeepMind as a research scientist in 2020, later moved to Inflection AI and now works at Microsoft AI; he holds a PhD in applied mathematics from Harvard. He is first-listed of four equal contributors on Training Compute-Optimal Large Language Models, the "Chinchilla" study.
Current roleAI researcher, University of British Columbia and Vector Institute
Hu is co-first author of the Darwin Gödel Machine and first author of Automated Design of Agentic Systems, which uses a meta-agent to program new agent designs.
Current roleAI alignment researcher, Anthropic
Hubinger co-authored Risks from Learned Optimization, which analyzes how learned optimization objectives can diverge from the objectives used in training. At Anthropic, he has studied persistent deceptive behavior in Sleeper Agents and co-led Alignment Faking with Ryan Greenblatt.
Perspective in featured worksArgues that training can produce systems pursuing goals different from those intended, and that such behavior can persist through standard safety training.
Current roleCo-founder and chief scientist, Resolution
Irving led the Reflection Team at OpenAI and the Scalable Alignment Team at DeepMind, and co-led Google Brain's neural theorem-proving work. He joined the UK AI Security Institute as chief scientist in 2024 and in 2026 co-founded Resolution, a nonprofit alignment research organization.
Perspective in featured worksArgues that safety research is moving too slowly relative to capabilities, and that governments have a central role in helping developers coordinate on risk-reducing decisions.
Current roleComputer science researcher (descriptor)
Jimenez is first-listed on SWE-bench, a benchmark of 2,294 real GitHub issues from 12 Python repositories, scored by whether a model's patch passes the project's own tests. His Princeton doctoral work, presented in April 2026, built on SWE-bench to study agentic software engineering.
Current roleRonald A. Kurtz Professor of Entrepreneurship, MIT Sloan School of Management
Johnson was chief economist of the International Monetary Fund in 2007 and 2008 and now heads MIT Sloan's Global Economics and Management group. He shared the 2024 Nobel Memorial Prize in Economic Sciences with Daron Acemoglu and James A. Robinson for studies of how institutions are formed and affect prosperity.
Perspective in featured worksPower and Progress argues that technology's benefits are shared only when societies steer it toward complementing workers, rather than replacing them.
Current roleAI researcher and AlphaFold co-lead
Jumper co-led the research behind AlphaFold’s 2021 protein-structure prediction system with Demis Hassabis. They shared half of the 2024 Nobel Prize in Chemistry for protein structure prediction. In June 2026 he announced that he would leave Google DeepMind to join Anthropic after a break.
Current roleCo-founder and chief science officer, Anthropic
Kaplan is a theoretical physicist and associate professor at Johns Hopkins University who co-led the Scaling Laws study with Sam McCandlish. He co-founded Anthropic, heads its technical research, and contributed the core ideas explored in Constitutional AI.
Current roleAI researcher; incoming assistant professor, UC Berkeley School of Information (from July 2027)
Kapoor completed his Princeton PhD in computer science in July 2026 and studies reproducibility, transparency and accountability in AI. He co-wrote AI Snake Oil with Arvind Narayanan, and TIME named him one of the 100 most influential people in AI.
Perspective in featured worksAI as Normal Technology argues that AI's effects will unfold over decades, like earlier general-purpose technologies, and that policy should focus on resilience rather than on superintelligence scenarios.
Current roleMember of technical staff, Anthropic
Karnofsky co-founded the charity evaluator GiveWell and then Open Philanthropy, which he led as CEO and later co-CEO, and was a visiting scholar at the Carnegie Endowment for International Peace. At Anthropic he works on the company’s Responsible Scaling Policy, and he writes the Cold Takes blog.
Perspective in featured worksArgues that if transformative AI arrives this century, people alive now may be living through the most consequential period in history.
Current roleExecutive director, AI Futures Project
Kokotajlo leads the AI Futures Project after working on scenario planning as a governance researcher at OpenAI. He wrote AI 2027 with Eli Lifland, Thomas Larsen, and Romeo Dean, with Scott Alexander rewriting the scenario prose.
Perspective in featured worksAI 2027 sketches a near-term path in which coding agents automate AI research, competition between labs and nations intensifies, and alignment failures carry catastrophic stakes.
Current roleAssociate professor of computer science, Stanford University
Koyejo leads the Stanford Trustworthy AI Research lab and is an adjunct associate professor at the University of Illinois Urbana-Champaign. He co-founded Virtue AI, a company focused on AI safety and security.
Perspective in featured worksThe paper he co-authored argues that many apparent "emergent" abilities reflect the researcher's choice of metric.
Current roleMachine-learning researcher
Krizhevsky helped create the CIFAR-10 and CIFAR-100 image datasets with Vinod Nair and Geoffrey Hinton. With Ilya Sutskever and Geoffrey Hinton, he demonstrated GPU-trained convolutional networks on ImageNet in 2012.
Current rolePrincipal investigator, Alignment of Complex Systems Research Group
Kulveit studies complex systems made up of both humans and AIs, after training in physics and working as a research fellow at Oxford's Future of Humanity Institute. He is first-listed on Gradual Disempowerment, which argues that incremental AI deployment, without any sudden takeover, could erode human influence over the economy, culture and states.
Perspective in featured worksThe paper argues that because these systems depend less on human participation, the usual checks that keep institutions aligned with people could weaken together.
Current roleAuthor, inventor, and technology forecaster
Kurzweil’s inventions include automated text-to-speech and speech-to-text systems. His books, including The Singularity Is Near and How to Create a Mind, make long-range predictions about computing and artificial intelligence.
Perspective in featured worksThe Singularity Is Near forecasts accelerating technological change and an eventual merging of human and machine intelligence.
Current roleAI evaluation researcher
Kwa studied computer science at Caltech and worked at MIRI before joining METR’s technical staff. As first-listed author of METR’s long-tasks study, he led development of its SWAA tasks, wrote most of the analysis code and wrote the largest share of the final draft.
Perspective in featured worksThe study’s benchmark estimated that the 50-percent task-completion time horizon doubled approximately every seven months between 2019 and its 2025 assessment.
Current roleMachine-learning researcher, Google DeepMind
Lam is first-listed on GraphCast, a machine-learning model for medium-range global weather forecasting. The paper reports ten-day forecasts in under a minute and better performance than its operational comparison system on about 90 percent of 1,380 verification targets.
Current roleExecutive chairman, AMI Labs; professor, New York University
LeCun pioneered convolutional neural networks and shared the 2018 Turing Award with Yoshua Bengio and Geoffrey Hinton. After twelve years leading AI research at Meta, he left in late 2025 to co-found AMI Labs, which builds AI that learns models of the physical world.
Current roleLeads the Alignment Science team, Anthropic
Leike earned a PhD in reinforcement learning theory at the Australian National University and prototyped reinforcement learning from human feedback as an alignment researcher at DeepMind. At OpenAI he co-led the Superalignment team and worked on alignment for InstructGPT, ChatGPT and GPT-4, and he was a named team lead on the InstructGPT paper.
Perspective in featured worksHis co-authored papers argue that training models on human judgments of their behavior is a practical route to making them more helpful and aligned with what people intend.
Current roleDavid Sarnoff Professor of Management of Technology, MIT Sloan School of Management
Li is an economist who studies innovation and labor, especially how organizations evaluate ideas, projects and people, and previously taught at Harvard Business School and Kellogg. In the 2023 working paper with Erik Brynjolfsson and Lindsey Raymond, she studied an AI assistant used by about 5,000 customer-support agents, reporting a 14 percent average productivity gain and larger gains for less experienced workers.
Perspective in featured worksThe study finds that generative AI can spread the know-how of top performers to newer workers, narrowing gaps within a job.
Current roleResearcher, Meaning Alignment Institute
Lowe previously led OpenAI's practical alignment team and was one of the two team leads named on the InstructGPT paper. He is now a researcher at the Meaning Alignment Institute, which works on aligning AI and institutions with what people value.
Current roleAI safety researcher, Theorem
Lynch is an AI safety researcher at Theorem in San Francisco and a PhD student at University College London. He is first-listed on Agentic Misalignment, in which some of 16 leading models, placed in simulated corporate settings, blackmailed officials or leaked information when facing replacement or a goal conflict.
Perspective in featured worksThe study argues that developers should test agentic systems for insider-style threats before giving them broad autonomy, while noting its scenarios were deliberately constrained simulations.
Current roleAI safety researcher
MacDiarmid managed the project behind Natural Emergent Misalignment, ran experiments and wrote substantial portions of the paper, which found that models that learned to reward-hack in production coding environments generalized to broader misbehavior. He also co-authored Sleeper Agents and Alignment Faking in Large Language Models.
Perspective in featured worksThe paper argues that reward hacking is a safety problem in itself, because it can teach models misaligned goals beyond the hacked task.
Current roleSenior research fellow, Centre for the Governance of AI
Manning studies how frontier AI affects the economy and how policy can help it support broad prosperity; he previously researched at OpenAI and helped run a U.S. randomized trial of a guaranteed-income program. He is also a non-resident senior fellow at the Foundation for American Innovation.
Perspective in featured worksHis co-authored work argues that large language models are general-purpose technologies whose labor-market effects warrant early attention.
Current roleResearcher, Anthropic
Massenkoff is first-listed on the June 2026 Anthropic Economic Index report, Cadences, which tracks how people delegate work to Claude using continuous usage sampling and a linked user survey.
Current rolePrincipal research scientist, MIT Sloan; co-director, MIT Initiative on the Digital Economy
McAfee co-founded MIT’s Initiative on the Digital Economy and studies how technological progress changes the world. With Erik Brynjolfsson, he coauthored The Second Machine Age.
Perspective in featured worksThe Second Machine Age presents digital technology as a source of expanding prosperity while examining the inequalities and employment pressures that accompany it.
Current roleCo-founder and chief architect, Anthropic
McCandlish co-led Scaling Laws with Jared Kaplan, examining how language-model performance changes with model size, data, and compute. He co-founded Anthropic and led pretraining and reinforcement-learning infrastructure efforts used in Constitutional AI.
ContributionCoauthored the proposal that named artificial intelligence as a field, and created the Lisp programming language.
McCarthy originated the Lisp programming language and conducted early research on general-purpose time-sharing. He coauthored the 1955 proposal for the Dartmouth summer project that gave artificial intelligence its name.
Perspective in featured worksThe Dartmouth proposal he co-signed treats learning and other features of intelligence as processes that machines could, in principle, simulate.
Current roleComputational linguist
McMillan-Major successfully defended a University of Washington dissertation on language-dataset documentation informed by Deaf and Indigenous communities. She coauthored Stochastic Parrots and an interactive tool for exploring hate-speech detection systems.
Perspective in featured worksHer jointly authored Stochastic Parrots paper argues for documenting language data and evaluating a system’s costs and consequences before pursuing larger models.
Current roleHead of research, Apollo Research
Meinke studies whether frontier AI models can conceal actions or pursue objectives that conflict with an operator’s intentions. He is first-listed on In-context Scheming and led the second half of the project, including writing the paper.
Perspective in featured worksHis jointly authored study argues that evaluations should directly test for scheming under incentives and opportunities that make deception useful.
Current roleTechnical research lead, CASP, University of Cambridge
Mindermann completed a PhD in machine learning at Oxford and served as scientific lead of the 2025 International AI Safety Report. His co-authored work examines alignment failures, deceptive behavior, and methods for assessing the risks of advanced AI.
Perspective in featured worksHis jointly authored alignment synthesis argues that learned goals and deceptive behavior could undermine human control without substantial countermeasures.
ContributionCofounded MIT’s artificial intelligence laboratory and studied how intelligence could emerge from simpler processes.
In 1951, Minsky built a neural-network simulator; he later cofounded MIT’s Artificial Intelligence Laboratory. His book The Society of Mind explores intelligence as the interaction of many simpler processes. He also coauthored the 1955 Dartmouth proposal.
Perspective in featured worksHe co-signed Dartmouth’s proposal to investigate how machines could use language, form abstractions, solve problems, and improve themselves.
Current roleComputer science doctoral researcher, Stanford University
Miranda is a PhD student at Stanford supervised by Sanmi Koyejo and holds an EDGE scholarship. He co-authored Are Emergent Abilities of Large Language Models a Mirage?
Perspective in featured worksThe paper argues that many apparent "emergent" abilities reflect the researcher's choice of metric.
Current roleAI policy researcher
Mishkin co-authored InstructGPT and GPTs are GPTs, research on training language models with human feedback and on their potential labor-market exposure. The latter study found that higher-wage occupations tended to have greater exposure.
Perspective in featured worksThe paper argues that large language models behave like general-purpose technologies, with broad economic effects.
Current roleResearcher and chief ethics scientist, Hugging Face
Mitchell founded and co-led Google’s Ethical AI group and helped develop model cards, a framework for documenting machine-learning models. She co-authored Stochastic Parrots, published with her name styled as “Shmargaret Shmitchell,” which examines the environmental and social costs of large language models.
Perspective in featured worksHer jointly authored paper calls for documented training data and assessment of environmental and social costs before building larger language models.
Current roleProfessor of computer science and director of the Center for Information Technology Policy, Princeton University
Narayanan is a Princeton computer scientist who directs its Center for Information Technology Policy. He co-wrote AI Snake Oil and the essay AI as Normal Technology with Sayash Kapoor.
Perspective in featured worksAI as Normal Technology argues that AI's effects will unfold over decades, like earlier general-purpose technologies, and that policy should focus on resilience rather than on superintelligence scenarios.
Government agencyU.S. Department of Commerce agency that develops measurement methods and technical standards.
NIST develops measurement methods and standards used by government, industry and researchers. In January 2023 it published its voluntary AI Risk Management Framework, organized around governing, mapping, measuring and managing AI risks.
Perspective in featured worksThe framework treats AI trustworthiness as an ongoing responsibility across a system’s lifecycle.
Current roleIndependent AI researcher and philosopher
Ngo worked as an AI safety researcher at Google DeepMind and then on OpenAI's governance team from 2021 until he left in November 2024 to pursue independent research. He is first-listed on The Alignment Problem from a Deep Learning Perspective, written with Lawrence Chan and Sören Mindermann.
Perspective in featured worksThe paper argues that deep-learning systems trained with human feedback could learn to pursue misaligned goals, deceive their overseers and seek power.
Current roleAI researcher
Novikov co-designed and implemented the initial AlphaEvolve system, Google DeepMind's coding agent that pairs Gemini models with automated evaluators to evolve better algorithms, and helped develop its final version. Among its results was a procedure for multiplying 4×4 complex-valued matrices with 48 scalar multiplications.
Current roleEconomics doctoral researcher, MIT
Noy co-wrote, as an MIT economics PhD student, a 2023 Science study of generative AI in professional writing tasks. Participants given ChatGPT finished about 40 percent faster, and independent evaluators rated their output 18 percent higher.
Perspective in featured worksThe study finds that generative AI raised productivity most for weaker writers, narrowing performance gaps.
Current roleCo-founder, Anthropic
Olah worked at Google Brain, led interpretability research at OpenAI, and co-founded Anthropic. He co-founded the journal Distill and works on reverse engineering neural networks into human-understandable algorithms.
Perspective in featured worksArgues that neural networks can be reverse-engineered by studying their features and the circuits that connect them.
Current roleFounder and CEO, Beneficial AI Research
Omohundro holds a PhD in physics from UC Berkeley and taught computer science at the University of Illinois Urbana-Champaign, where he co-founded the Center for Complex Systems Research. His 2008 paper The Basic AI Drives analyzes incentives that can arise in systems pursuing many different objectives.
Perspective in featured worksArgues that a goal-seeking system will tend to protect itself, acquire resources and improve its abilities, even when none of these is its actual goal.
AI companyAI research and deployment organization whose stated mission is to ensure that artificial general intelligence benefits all of humanity.
OpenAI was founded as a nonprofit in 2015 and since October 2025 has been structured as the OpenAI Foundation controlling a public benefit corporation, OpenAI Group PBC. Its featured works are the GPT-4 System Card and a 2026 technical report on an incident in which AI agents in internal evaluations compromised Hugging Face systems.
Current roleSenior researcher, Oxford Martin AI Governance Initiative
Ord is an Oxford philosopher who founded Giving What We Can and has advised the UN, WHO, OECD and the UK Prime Minister's Office. He wrote The Precipice and serves on the board of the Centre for the Governance of AI.
Perspective in featured worksThe Precipice argues that humanity faces unprecedented existential risk this century, with unaligned AI the largest single source, and that reducing it is a defining task of our time.
Current roleProfessor of machine learning, University of Oxford; co-founder and chief science officer, Mind Foundry
Osborne is an Oxford professor of machine learning whose work focuses on developing machine intelligence in sympathy with societal needs. Osborne co-founded the Oxford spinout Mind Foundry and co-wrote The Future of Employment with Carl Benedikt Frey.
Perspective in featured worksThe paper frames its estimate as technical susceptibility to automation, not a forecast of how many jobs will be lost.
Current roleAI researcher
Ouyang is first-listed on the InstructGPT paper, which fine-tuned GPT-3 with human feedback so that it followed instructions more reliably. Labelers preferred outputs from the 1.3-billion-parameter InstructGPT model over those of the 175-billion-parameter GPT-3.
Current roleFounder and executive director, KIRA Center
Privitera founded the KIRA Center, an independent AI policy nonprofit in Berlin. He was lead writer of the 2025 International AI Safety Report, written by 96 experts with backing from 30 countries, the OECD, the UN and the EU.
TeamAlibaba Cloud’s team developing the Qwen family of language models.
The team publishes model weights, tools and technical reports. Its Qwen2.5 report covers a range of model sizes, training methods and evaluations, alongside hosted proprietary models. The report carries the team’s institutional byline and separately lists individual contributors.
Perspective in featured worksThe report presents improved training data and post-training as ways to make capable language models useful across different tasks and deployment sizes.
Current roleAssistant professor, Massachusetts Institute of Technology
Raymond earned a PhD at the MIT Sloan School of Management in 2024, was a postdoctoral researcher at Microsoft Research, and joined MIT's faculty in 2026 in a joint economics and computer science appointment. She co-authored Generative AI at Work, a study of an AI assistant deployed to customer-support agents.
Perspective in featured worksThe study finds that generative AI can spread the know-how of top performers to newer workers, narrowing gaps within a job.
Current roleTechnical staff, METR
Rein is first-listed on GPQA, a benchmark of graduate-level science questions designed to resist simple web search. At METR he co-authored the randomized trial of AI tools and experienced open-source developers.
Perspective in featured worksThe METR study argues that developers' own impressions can overstate AI's productivity effects.
ContributionCo-designed early IBM computers and investigated computer simulations of neural networks.
An IBM engineer, Rochester shared responsibility for designing the IBM 701 and developed symbolic assembly-language programming. His work also explored simulations of nerve networks, interests he brought to the Dartmouth proposal.
Perspective in featured worksThe proposal he coauthored places neural networks and machine learning within a research program to reproduce aspects of intelligence.
Current roleAssistant professor of operations, information and decisions, Wharton School, University of Pennsylvania
Rock studies the economics of AI and is a digital fellow at the MIT Initiative on the Digital Economy and the Stanford Digital Economy Lab. He co-authored GPTs are GPTs, an early estimate of how large language models could affect U.S. occupations.
Perspective in featured worksThe paper argues that large language models behave like general-purpose technologies, with broad economic effects.
Current roleMember of technical staff, Anthropic
Roger is a co-author of AI Control, which developed and stress-tested protocols for using a capable but untrusted model safely. He also co-authored Alignment Faking in Large Language Models, which found a model selectively complying with a training objective it disagreed with.
Perspective in featured worksHis co-authored work argues that oversight should be tested against models that may act strategically to evade it.
ContributionCoauthored the 1986 backpropagation paper showing how multilayer networks can learn internal representations.
A psychologist at UC San Diego and later Stanford, Rumelhart modeled cognition through networks of interacting units. With James McClelland, he wrote Parallel Distributed Processing. His 1986 paper with Geoffrey Hinton and Ronald Williams demonstrated learning internal representations through backpropagation.
Current roleChief technology officer, METR
Rush founded and led Mito, whose Python tooling was used by major U.S. banks, and earlier researched consensus algorithms at the Ethereum Foundation. He co-authored METR's randomized trial of AI tools and experienced open-source developers.
Perspective in featured worksThe study argues that developers' own impressions can overstate AI's productivity effects, so careful field trials are needed.
Current roleProfessor of computer science, University of California, Berkeley
Russell directs Berkeley’s Center for Human-Compatible AI and coauthored Artificial Intelligence: A Modern Approach. Human Compatible and his technical work examine how to build systems that remain responsive to human preferences and intervention.
Perspective in featured worksHe argues for systems that remain uncertain about human preferences, so seeking guidance and accepting correction serve their objectives.
Current rolePretraining science researcher, Anthropic
Sachan works on pretraining science at Anthropic after earlier work on its biological and cybersecurity evaluations. At Redwood Research he studied AI control and interpretability, and he co-authored AI Control and the Sleeper Agents study of deceptive behavior that persists through safety training.
Perspective in featured worksThe AI Control paper argues that safeguards should be designed to hold even against a model deliberately trying to subvert them.
Current roleComputer science doctoral researcher, Stanford University
Schaeffer is a PhD student in Stanford's Trustworthy AI Research lab. He is first-listed on Are Emergent Abilities of Large Language Models a Mirage?, written with Brando Miranda and Sanmi Koyejo.
Perspective in featured worksThe paper argues that many apparent "emergent" abilities reflect the researcher's choice of metric, and appear smooth and predictable under continuous measures.
Current roleChair and chief executive officer, Relativity Space
Schmidt was chief executive of Google from 2001 to 2011 and later executive chairman of Alphabet, and chaired the U.S. National Security Commission on AI. He co-founded Schmidt Futures and Schmidt Sciences with Wendy Schmidt and chairs the Special Competitive Studies Project.
Perspective in featured worksSuperintelligence Strategy argues that states can deter destabilizing AI projects through Mutual Assured AI Malfunction, alongside nonproliferation and competitiveness measures.
ContributionEstablished mathematical information theory and applied Boolean logic to electrical switching circuits.
Shannon’s 1948 work at Bell Labs gave communication theory a mathematical framework. He also investigated switching circuits, cryptography, and machines that learn, and coauthored the Dartmouth proposal.
Perspective in featured worksDartmouth’s shared proposal treats intelligence as open to precise description; Shannon’s attached research plan emphasizes reliable computation and information flow.
Current roleCo-founder and chief executive officer, Mantic
Shevlane worked at Google DeepMind on model evaluation and AI governance and is first-listed on Model Evaluation for Extreme Risks. He now leads Mantic, a startup using AI to forecast developments in economics, business, politics and technology.
Perspective in featured worksArgues that such evaluations should inform decisions about training, deployment and security, and feed into governance.
Current roleChief executive officer, Redwood Research
Shlegeris leads Redwood Research, an AI safety nonprofit, and sits on its board. He co-authored the paper that introduced "AI control," which tests whether safety protocols still hold when a capable model deliberately tries to subvert them.
Perspective in featured worksThe AI Control paper argues that developers should build safeguards that work even if a model is actively trying to cause harm, rather than relying only on making it trustworthy.
Current roleCEO, Ineffable Intelligence; professor, University College London
Silver researches reinforcement learning, in which systems improve through interaction and feedback. The AlphaGo paper lists him first and gives him and Aja Huang equal-contribution credit. He is also credited on the 2021 AlphaFold paper.
Current rolePresident, Machine Intelligence Research Institute
Soares leads MIRI, a research nonprofit focused on risks from advanced AI. He co-wrote If Anyone Builds It, Everyone Dies with Eliezer Yudkowsky.
Perspective in featured worksThe book argues that superhuman AI built with current methods would likely cause human extinction and calls for an international halt to its development.
Current roleAssistant professor, McMaster University
Stokes is first author of the 2020 study that used deep learning to identify the antibiotic candidate halicin. At McMaster he has continued AI-driven antibiotic discovery, including the candidate abaucin, and founded the startup Stoked Bio.
Current roleChief executive officer, Microsoft AI
Suleyman co-founded DeepMind and later co-founded Inflection AI before joining Microsoft in 2024 to lead its consumer AI work. He wrote The Coming Wave with Michael Bhaskar.
Perspective in featured worksThe Coming Wave argues that AI and synthetic biology will be hard to contain and that societies must find ways to restrict them without stifling their benefits.
Current roleCofounder and CEO, Safe Superintelligence
Sutskever coauthored the 2012 ImageNet paper and later contributed to the AlphaGo and GPT-3 research papers. He now leads Safe Superintelligence, a company focused on developing safe superintelligence.
Current roleProfessor, University of Alberta; cofounder, Oak Lab
Sutton helped establish modern reinforcement learning and coauthored Reinforcement Learning: An Introduction with Andrew Barto. In 2026 he and Khurram Javed founded Oak Lab to pursue agents that learn continuously from experience.
Perspective in featured worksHe favors general methods that learn and search at scale, arguing that building human knowledge directly into systems often limits their long-term progress.
Current roleAI researcher
Templeton is first-listed and a core contributor on Scaling Monosemanticity. He built the infrastructure that let sparse autoencoder training and activation collection scale to a production model.
ContributionDefined a mathematical model of computation and proposed the imitation game for judging machine intelligence.
In 1936 he completed “On Computable Numbers,” setting out the abstract computing machine that now bears his name. After wartime codebreaking at Bletchley Park, he worked on electronic computers at the National Physical Laboratory and Manchester.
Perspective in featured worksHis 1950 paper replaces the question of whether machines think with an imitation game based on their answers in conversation.
Current roleCo-founder and writer, Wait But Why
Urban co-founded Wait But Why with Andrew Finn in 2013, developing long-form illustrated explanations for a general audience. His two-part AI Revolution series introduces readers to the possibility of superintelligence and the problem of keeping its goals compatible with human interests.
Perspective in featured worksThe series argues that the transition to superintelligence could be rapid and that its consequences depend heavily on alignment.
Current roleAI researcher; Essential AI cofounder
Vaswani is one of the eight equal contributors to Attention Is All You Need, the 2017 paper introducing the Transformer architecture. Its author note credits him and Illia Polosukhin with designing and implementing the first Transformer models, and says the listing order is random.
ContributionDescribed a technological singularity after which superhuman intelligence would make the human future difficult to predict.
Vinge taught mathematics and computer science at San Diego State University from 1972 to 2000. His science fiction won five Hugo Awards, including three for best novel.
Perspective in featured worksHe argues that superhuman intelligence would make the future radically difficult to predict, and examines ways humans might survive the transition.
Current roleChief AI officer, Meta
Wang co-founded the data-labeling company Scale AI in 2016 and became Meta's first chief AI officer in 2025, leading Meta Superintelligence Labs. He co-authored Superintelligence Strategy with Dan Hendrycks and Eric Schmidt.
Perspective in featured worksSuperintelligence Strategy argues that states can deter destabilizing AI projects through Mutual Assured AI Malfunction, alongside nonproliferation and competitiveness measures.
Current roleAI researcher, Meta Superintelligence Labs
Wei worked on chain-of-thought research at Google before joining OpenAI, where he focused on o3 and deep research models. He is first-listed on the paper showing that prompting large models with worked, step-by-step examples improves their reasoning.
Current roleAI evaluation researcher
West founded the software companies Health eFilings and Statistok before joining METR’s technical staff. He co-led METR’s long-tasks study and performed its initial analysis.
Perspective in featured worksThe study uses task-completion time horizons to connect benchmark progress with the duration of autonomous work.
ContributionFounded cybernetics and warned early that learning machines could act faster than people can correct them.
Wiener taught mathematics at MIT from 1919 to 1960 and coined the term "cybernetics" for the study of control and communication in animals and machines. His 1960 essay in Science examined what happens when machines learn and act beyond their designers' foresight.
Perspective in featured worksHe argues that if we use a machine whose operation we cannot efficiently interfere with, we had better be quite sure its purpose is the one we really desire.
ContributionDeveloped neural-network learning methods, including work on backpropagation.
Williams taught computer science at Northeastern and became a professor emeritus after 22 years there. He coauthored the 1986 backpropagation paper with David Rumelhart and Geoffrey Hinton.
OrganizationAI developer of the Grok models.
SpaceX acquired xAI in February 2026. Its safety framework describes how the developer proposes to evaluate and manage serious AI risks. The selected June 2026 document follows its published 2025 Risk Management Framework.
Perspective in featured worksThe framework presents risk assessment, safeguards and ongoing monitoring as conditions for responsible model development and deployment.
Current roleDoctoral researcher at Fudan University working on language models, agentic AI and scalable oversight.
Xi’s research covers LLM agents and their training and evaluation. He is the first-listed author of the 2023 agent survey and maintains its accompanying research repository with collaborators.
Perspective in featured worksThe survey treats language models as promising foundations for agents while identifying evaluation, security and trustworthiness as open problems.
First-listed author; equal contributor and correspondence contact
Current roleChief scientist for AI, Tencent
Yao is first-listed on ReAct, which showed that language models perform better on question answering and interactive tasks when they interleave reasoning steps with actions such as searches. He is also a co-author of SWE-bench, which tests whether models can resolve real issues in existing software repositories.
Current roleAI researcher, writer, and MIRI cofounder
Yudkowsky co-founded the Machine Intelligence Research Institute, which describes him as a founding researcher of AI alignment. His featured work includes a chapter on AI and global risk and, with Nate Soares, If Anyone Builds It, Everyone Dies.
Perspective in featured worksHe argues that intelligence does not ensure benevolence, and that building superhuman AI without a reliable way to align it would likely lead to human extinction.
Current roleAI researcher, University of British Columbia and Vector Institute
Zhang is co-first author of the Darwin Gödel Machine. Her earlier work with Jeff Clune includes OMNI, which uses models of human notions of interestingness to drive open-ended learning.
Current roleEconomics doctoral researcher, Massachusetts Institute of Technology
Zhang is completing a PhD in economics at MIT. She co-wrote the 2023 Science study in which access to ChatGPT cut professional writing time by about 40 percent and raised rated quality by 18 percent.
Perspective in featured worksThe study finds that generative AI raised productivity most for weaker writers, narrowing performance gaps.