Citation and evidence

Jared Kaplan

27 min full readUpdated 42 references

This article's verification

Report a problem with this article

Edit article
More

Use this article

Raw MarkdownExplore connections

Improve this page

Revision historyDiscuss this article

Topics

AnthropicPeople

Cite this article

Jared Daniel Kaplan is a theoretical physicist and artificial intelligence researcher who co-founded Anthropic in 2021 and serves as its Chief Science Officer, overseeing the research behind the Claude family of language models.[1][2] He is best known as the lead author of the 2020 paper "Scaling Laws for Neural Language Models," which showed that transformer language model loss falls as a smooth power law in model size, dataset size, and training compute, an empirical finding that guided OpenAI's decision to scale GPT-3 to 175 billion parameters and reshaped frontier model development across the industry.[4][5] Before founding Anthropic, Kaplan spent roughly fifteen years as a theoretical physicist and remains an associate professor in the Department of Physics and Astronomy at Johns Hopkins University, where his academic work centered on quantum gravity, the AdS/CFT correspondence, and the conformal field theory bootstrap.[1][3] In October 2024 he took on the additional role of Anthropic's Responsible Scaling Officer, putting him in charge of the company's pre-deployment safety evaluations under its Responsible Scaling Policy.[6]

Photograph of Jared Kaplan
Jared Kaplan. TechCrunch, via Wikimedia Commons, CC BY 2.0.
FieldDetail
BornJared Daniel Kaplan
EducationIllinois Mathematics and Science Academy; B.S. physics and mathematics, Stanford (2005); Ph.D. physics, Harvard (2009)
Doctoral advisorNima Arkani-Hamed
Doctoral thesis"Aspects of Holography"
Academic postAssociate Professor, Johns Hopkins University Department of Physics and Astronomy (on the faculty since 2012)
Industry rolesResearcher at OpenAI (2019 to 2020); co-founder and Chief Science Officer, Anthropic (2021 to present); Anthropic Responsible Scaling Officer (named October 2024)
Notable works"Scaling Laws for Neural Language Models" (2020); "Language Models are Few-Shot Learners" (2020); "Constitutional AI: Harmlessness from AI Feedback" (2022)
HonorsHertz Fellow (2005), Sloan Research Fellowship (2014), NSF CAREER Award (2015), TIME100 AI (2025)
Net worthApproximately $3.7 billion (Forbes, 2026)

Expanded article table

Who is Jared Kaplan?

Jared Kaplan is an American theoretical physicist turned AI researcher who is a co-founder and the Chief Science Officer of Anthropic, the AI safety company that builds the Claude models.[1][2] He is widely credited, alongside Sam McCandlish and Dario Amodei, with introducing the neural scaling-laws framework that organizes how the field thinks about pretraining budgets.[34] He continues to hold an associate professorship in physics at Johns Hopkins University in parallel with his Anthropic role.[1][3] As of 2026, Forbes estimated his net worth at approximately $3.7 billion, reflecting his founding equity stake in Anthropic after successive funding rounds valued the company in the hundreds of billions of dollars.[1]

"Scaling and the Road to Human-Level AI | Anthropic Co-founder Jared Kaplan", video from Y Combinator, published July 29, 2025. Watch on YouTube.

What was Jared Kaplan's early life and education?

Kaplan attended the Illinois Mathematics and Science Academy, a selective public high school, before enrolling at Stanford University, where he completed a bachelor's degree in physics and mathematics in 2005.[1][7] During his senior year at Stanford he was named a Fannie and John Hertz Foundation Fellow, a competitive graduate fellowship that funds doctoral study in the applied physical, biological, and engineering sciences and counts numerous prominent scientists among its alumni.[7] The Hertz Foundation supported his doctoral studies at Harvard, where he entered the physics Ph.D. program and worked in theoretical particle physics with advisors Howard Georgi and Nima Arkani-Hamed, two of the most influential figures in modern high-energy theory.[1][8] His 2009 dissertation, titled "Aspects of Holography," addressed topics in the AdS/CFT correspondence and quantum gravity, including holographic descriptions of black holes in anti-de Sitter space and questions about whether the holographic principle extends naturally to asymptotically flat spacetimes.[8][9]

After completing his Ph.D., Kaplan held a postdoctoral fellowship jointly affiliated with the SLAC National Accelerator Laboratory and Stanford University from 2009 through 2012, working on questions at the intersection of effective field theory, the conformal bootstrap, and quantum gravity.[3][10] During the SLAC years he became a regular collaborator of A. Liam Fitzpatrick, with whom he would publish many of his most influential physics papers over the next decade.[10][17] He joined the faculty of Johns Hopkins University in 2012 as an assistant professor in the Department of Physics and Astronomy, was tenured as an associate professor a few years later, and has remained on the Hopkins faculty since while continuing his research collaborations.[1][3] At Hopkins his teaching has included graduate courses in quantum field theory, conformal field theory, and, in later years, the foundations of deep learning.[1]

What did Jared Kaplan do as a theoretical physicist?

For roughly the first fifteen years of his research career Kaplan worked exclusively as a theoretical physicist.[11] His academic publications cover effective field theory, particle physics, cosmology, scattering amplitudes, the conformal field theory bootstrap, AdS/CFT correspondence, and quantum gravity.[3][10] At Johns Hopkins he taught graduate-level quantum field theory; the lecture notes he prepared for that course, about 260 pages long and circulating informally as "QFT Lectures Notes," are used by graduate students as a free study reference and draw on standard texts such as Weinberg and Schwartz, with less conventional pedagogical choices such as an early section that builds up fields from a lattice of atoms with nearest-neighbor interactions, or "balls and springs."[12] The fall-semester notes work through Lagrangians and harmonic oscillators, special relativity and antiparticles, canonical quantization, scattering and perturbation theory, the LSZ reduction formula, Feynman rules, spinors and quantum electrodynamics, renormalization (including Wilsonian renormalization flows), and path integrals, with a spring-semester treatment of effective actions, symmetries and spin-statistics, Lie algebras, Goldstone bosons, non-abelian gauge theories and asymptotic freedom, the Higgs mechanism, QCD and deep inelastic scattering, and anomalies.[12] He also distributed lecture notes titled "Lectures on AdS/CFT from the Bottom Up," covering the conformal group, conformal partial wave expansions, and the analytic bootstrap, which serve as an accessible introduction to a technically demanding area of mathematical physics.[13]

What is the conformal bootstrap work he contributed to?

A significant strand of Kaplan's physics output examined how the conformal bootstrap, an axiomatic approach to conformal field theory that uses crossing symmetry and unitarity to constrain operator dimensions and operator product expansion coefficients, intersects with the AdS/CFT correspondence. A 2012 paper with A. Liam Fitzpatrick, David Poland, and David Simmons-Duffin, "The Analytic Bootstrap and AdS Superhorizon Locality," demonstrated that every unitary conformal field theory above two dimensions containing a scalar operator must possess an infinite tower of operators whose twists approach specific accumulation points as their spin grows; this result connected the consistency of CFT four-point functions to the locality of bulk physics in anti-de Sitter space and is one of the founding works of the modern analytic-bootstrap program.[14] The paper has been cited more than 600 times, according to Google Scholar.[22] A follow-up 2014 paper with Fitzpatrick and Matthew Walters, "Universality of Long-Distance AdS Physics from the CFT Bootstrap," extended the analysis to show that the leading interactions between widely separated objects in AdS gravity are universal consequences of the bootstrap, providing a CFT-side derivation of the long-distance behavior usually obtained from bulk effective field theory.[15]

Kaplan also co-authored a 2011 paper, "A Natural Language for AdS/CFT Correlators," with Fitzpatrick, Joao Penedones, Suvrat Raju, and Balt van Rees, which argued that Mellin space provides a particularly natural representation for holographic correlators because CFT correlators in Mellin space have poles whose residues encode the operator product expansion in a way that closely mirrors the factorization channels of bulk scattering amplitudes.[16] In the regime where correlators are computable by tree-level Witten diagrams in AdS, the authors derived explicit formulae for Mellin amplitudes and showed that they satisfy algebraic finite-difference equations, giving simple diagrammatic rules for constructing Mellin amplitudes in any bulk scalar theory.[16] A 2012 companion paper, "Unitarity and the Holographic S-Matrix," explored the analytic structure of these Mellin amplitudes and their relation to the bulk S-matrix.[17] Other physics papers addressed Virasoro conformal blocks and thermality in two-dimensional CFTs, information loss in AdS3/CFT2, and the S-matrix in twistor space.[22]

What recognition did he receive in physics?

In 2014 the Alfred P. Sloan Foundation named Kaplan a Sloan Research Fellow, an award given annually to about 126 early-career researchers in recognition of distinguished work and the potential to make substantial contributions to their fields; the fellowship carried a $50,000 two-year award.[18] In 2015 the National Science Foundation awarded him a CAREER grant (PHY-1454083) supporting research and teaching activities in theoretical particle physics; the CAREER program is the NSF's most prestigious award for junior faculty.[7] He is also a principal investigator within the Simons Foundation Collaboration on the Nonperturbative Bootstrap, a multi-institution program funding work on the conformal bootstrap that brings together researchers from leading physics departments and institutes.[3]

How did Jared Kaplan transition from physics to AI?

By the late 2010s Kaplan had grown frustrated with the pace of progress in the areas of physics he worked on, and friends, including several future Anthropic co-founders, persuaded him that AI was becoming a big deal.[34] In 2019 he joined OpenAI as a researcher while retaining his Hopkins faculty position.[1] His arrival at OpenAI placed him in close working relationships with Dario Amodei, then OpenAI's vice president of research, and Sam McCandlish, then a research scientist, both of whom co-authored his scaling-laws paper.[4][26] Kaplan brought to the engagement a physicist's intuition for scaling phenomena: in many physical systems, dimensionless observables follow power-law behavior across many orders of magnitude when no other characteristic scale is present, and Kaplan's instinct was that neural network training loss, viewed as a function of compute and data, might exhibit similar regularities if measured carefully across a wide enough range.[34]

What are the neural scaling laws (2020)?

The collaborative work that emerged from this intuition appeared on arXiv on January 23, 2020, under the title "Scaling Laws for Neural Language Models," with Kaplan as first author followed by Sam McCandlish, Tom Henighan, Tom B. Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei.[4] The paper reported that the cross-entropy test loss of decoder-only transformer language models is, to high accuracy, a smooth power-law function of three quantities considered separately: the number of non-embedding parameters N in the model, the size of the training dataset D in tokens, and the amount of compute C spent during training, with some trends spanning more than seven orders of magnitude.[4] Architectural details such as the ratio of network width to depth and the number of attention heads had only small effects within a wide range, suggesting that the scaling laws reflected a deep property of the loss landscape rather than artefacts of a particular hyperparameter choice.[4] Overfitting was shown to be governed by a simple ratio of model size to data size, and the dependence of training speed on model size was described by another simple functional form.[4]

The paper's most consequential prescriptive claim concerned compute-optimal training. Given a fixed compute budget, the authors argued that the optimal allocation pushed most of the budget into model size and trained the resulting very large model for a relatively short period on a modest number of tokens. As the abstract states, "Larger models are significantly more sample-efficient, such that optimally compute-efficient training involves training very large models on a relatively modest amount of data and stopping significantly before convergence."[4] This conclusion directly motivated OpenAI's subsequent decision to train GPT-3 at 175 billion parameters, ten times larger than any prior dense language model.[20][21] The paper became one of the most influential works in modern machine learning and has accumulated more than 10,000 direct citations on Google Scholar as of November 2026, with countless more citations through derivative works.[22] In a 2025 talk at Y Combinator's AI Startup School, Kaplan said the trends his team found were as precise as anything seen in physics or astronomy and gave them conviction that AI would keep getting smarter in a very predictable way.[34]

The compute-optimal prescription was later partially revised. In 2022 DeepMind's "Training Compute-Optimal Large Language Models" paper introduced the Chinchilla model and argued that Kaplan and colleagues had underestimated the importance of data: when the cosine learning-rate schedule was matched to each run's number of training tokens, model size and training tokens should be scaled equally, which for Chinchilla itself (70 billion parameters) worked out to about twenty training tokens per parameter.[23] Hoffmann and colleagues attributed the difference mainly to Kaplan and coauthors having used a fixed number of training tokens and a fixed learning-rate schedule for all models, and to their models being significantly smaller, many under 100 million parameters.[23] The Chinchilla scaling revision became the standard reference point for compute-optimal pretraining, but the qualitative finding of the Kaplan paper, that loss scales as predictable power laws in compute, data, and parameters, has been confirmed repeatedly and remains the central organizing principle of frontier model development.[23][34] The broader research field that grew out of these results is now referred to as the study of scaling laws for neural networks.[34]

What did Jared Kaplan contribute to GPT-3 and Codex?

While at OpenAI, Kaplan contributed to two further landmark papers. He is a co-author of "Language Models are Few-Shot Learners," posted on arXiv on May 28, 2020, which introduced GPT-3 and reported that scaling a transformer language model to 175 billion parameters produced strong in-context learning across a wide range of natural language tasks without parameter updates.[24] The paper's contributions list attributes the demonstration that larger models learn more quickly from in-context examples specifically to Kaplan and Sam McCandlish, an observation that linked the scaling-laws framework directly to the few-shot learning phenomenon that made GPT-3 commercially attractive.[24] The paper has accumulated more than 86,000 Google Scholar citations as of November 2026.[22]

Kaplan is also a co-author of the 2021 paper "Evaluating Large Language Models Trained on Code," which introduced OpenAI Codex (the model that initially powered GitHub Copilot) and the HumanEval benchmark for assessing functional correctness in code generation.[25] Codex was the first widely deployed application of GPT-style models to code, and HumanEval has since become the canonical entry-level coding benchmark cited by virtually every subsequent code-generation system.[25]

How did Jared Kaplan co-found Anthropic?

In late 2020 and early 2021 a group of senior OpenAI researchers and leaders, including Dario Amodei (then vice president of research), Daniela Amodei (then vice president of safety and policy), Kaplan, Sam McCandlish, Tom Brown, Chris Olah, Jack Clark, and Ben Mann, departed to found Anthropic as a public benefit corporation focused on AI safety research.[26][27] Anthropic launched publicly in 2021 and raised an initial $124 million Series A round led by Jaan Tallinn, with investors including Dustin Moskovitz and Eric Schmidt.[41] Kaplan took the title of Chief Science Officer.[1][2]

What is Jared Kaplan's role at Anthropic?

As Chief Science Officer, Kaplan is one of Anthropic's senior research leaders, alongside CEO Dario Amodei and President Daniela Amodei; since October 2025 the technical leadership has also included CTO Rahul Patil and co-founder Sam McCandlish, who moved from CTO to Chief Architect and leads pretraining.[1][42] In interviews he has described the founding scientific bet of the company as a continuation of the scaling story: that AI capabilities will continue to improve predictably with compute and data, that this trajectory points smoothly toward broadly human-level AI, with an impact comparable to the industrial and scientific revolutions that could begin to arrive within the coming decade, and that responsible development requires preparing safety techniques and policies that scale with capability rather than waiting until capability arrives.[34][35]

Kaplan is a co-author of several foundational Anthropic alignment papers. He is a listed author on "Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback" (April 2022), an early study using RLHF to train large language models toward helpfulness and harmlessness simultaneously, which introduced the now-standard "HH" preference framework used throughout the industry.[29] He is also a listed author on "Constitutional AI: Harmlessness from AI Feedback" (December 2022), which introduced the technique of training a model to critique and revise its own outputs against a written list of principles, a "constitution," and then using AI-generated preference labels rather than human ones in the reinforcement-learning stage; this method is now known as RLAIF (reinforcement learning from AI feedback).[30] Constitutional AI became the public name for Anthropic's alignment methodology and was used in training successive generations of Claude, including Claude 1, Claude 2, Claude 3 Opus, and later models.[30][31] In a 2024 episode of the Life with Machines podcast Kaplan discussed Constitutional AI as a way of shaping a language model's behavior with a written constitution, contrasting the inspectable, principle-based approach with the harder-to-audit alternative of relying solely on human labels for adversarial behaviors.[32]

Anthropic's research program also includes mechanistic interpretability, the project of reverse-engineering trained neural networks into human-understandable algorithmic components. In his 2025 Y Combinator talk Kaplan compared interpretability to neuroscience, noting that, unlike a brain, everything inside an AI model can be measured.[34] Anthropic's work in this area has used dictionary learning and sparse autoencoders, and in May 2024 the company released work mapping millions of interpretable features inside the Claude 3 Sonnet model.[33]

What does the Responsible Scaling Officer do?

On October 15, 2024, Anthropic announced an updated version of its Responsible Scaling Policy and named Kaplan the company's Responsible Scaling Officer, succeeding co-founder and CTO Sam McCandlish, who had held the position during the policy's first year of implementation.[6] In that role Kaplan determines the safety assessments and precautions Anthropic adopts before a model is released.[1] The updated policy defined two key Capability Thresholds that would require upgraded safeguards under its graduated AI Safety Level (ASL) Standards: autonomous AI research and development, and chemical, biological, radiological, and nuclear (CBRN) weapons.[6] Anthropic announced at the same time that it was hiring a Head of Responsible Scaling to coordinate the various teams involved in implementing the updated policy under Kaplan's overall direction.[6] In February 2026 Anthropic replaced that framework with version 3.0 of the policy, which dropped the company's earlier pledge not to train or release models unless adequate risk mitigations could be guaranteed in advance; Kaplan told TIME that it would not "actually help anyone for us to stop training AI models" while competitors pressed ahead.[39]

What is Jared Kaplan's role in public policy and advocacy?

Kaplan represents Anthropic in many policy and public-affairs settings. In 2023 he submitted a written statement to the U.S. Senate AI Insight Forum on "Risk, Alignment, and Guarding Against Doomsday Scenarios," describing the responsible-scaling approach and Anthropic's view of catastrophic AI risk.[35] He has appeared on the TechCrunch "Equity" podcast and at TechCrunch Sessions: AI, given a Y Combinator Startup Library interview on "Scaling and the Road to Human-Level AI," spoken on the Life with Machines podcast about Constitutional AI, and participated in industry conversations at venues including Salesforce TrailblazerDX.[32][34] In March 2026 he filed a declaration in Anthropic PBC v. U.S. Department of War, the company's federal lawsuit in the Northern District of California.[38]

What are Jared Kaplan's main research contributions?

Scaling laws

Kaplan's central named scientific contribution to AI is the establishment of empirical scaling laws for neural language models, encapsulated in the Scaling Laws for Neural Language Models paper. The result, that loss decreases as a smooth power law in model size, dataset size, and compute, supplied a quantitative basis for treating language-model development as a budgeted scaling exercise rather than an architectural search, and is widely credited with shifting both research culture and industry investment toward ever-larger pretraining runs.[4][21][34] The general framework is now referred to simply as "Kaplan scaling laws" in distinction to the later "Chinchilla scaling laws" revision; both are part of the broader field of scaling laws.[23] The framework has since been extended in several directions, including a 2021 OpenAI paper on "Scaling Laws for Transfer" co-authored by Kaplan and collaborators, which examined how scaling continues to behave when models trained on one distribution are fine-tuned on a related distribution.[37]

The Kaplan paper's empirical findings have been treated as design rules in industry practice. Frontier models have been planned with reference to projected loss curves derived from scaling-law fits (the GPT-3 paper credits Kaplan and McCandlish with applying scaling laws to guide its model and data scaling decisions), and frontier labs routinely run small "ladder" experiments to fit scaling exponents before committing to a large training run.[22][34] The economic implication, that returns to scale are smooth and predictable, has also influenced AI investment patterns and has been cited in policy discussions of compute governance.[35]

Constitutional AI and RLAIF

Within Anthropic, Kaplan is a co-author of and frequent public spokesperson for Constitutional AI, the alignment training method that uses a model to revise and re-rank its own outputs against a written set of principles, replacing some or all of the human preference labels conventionally used in reinforcement learning from human feedback.[30] The technique substantially reduces the volume of human-labeled adversarial data required for harmlessness training and provides a written specification of the model's intended values that can be inspected and updated as policies evolve.[30] In 2023 Anthropic published "Claude's Constitution," a public document describing the specific principles, drawn from sources including the Universal Declaration of Human Rights and other widely recognized normative documents, that the company uses to guide Claude's training.[31] Constitutional AI has been replicated and extended by researchers at other organizations, and a sizable academic literature has emerged comparing its outcomes with conventional RLHF on safety and helpfulness benchmarks.[30]

Mechanistic interpretability and safety

Anthropic's mechanistic interpretability research, part of the scientific program Kaplan helps lead as Chief Science Officer, aims to identify interpretable circuits inside trained models; Kaplan has compared it to neuroscience, with the advantage that everything inside an AI model can be measured.[34] Anthropic has framed the work as a long-term safety bet: if researchers can read out what a trained model is computing in human-interpretable terms, they can verify whether a model's actual reasoning matches its stated reasoning, an audit that pure black-box behavioral evaluations cannot provide.[33] The interpretability team's milestones during his tenure include the 2024 release of "Scaling Monosemanticity," which used sparse autoencoders to extract millions of human-interpretable features from a frontier-scale model.[33]

He has spoken on Capitol Hill about AI risks and policy, including the 2023 written statement to the U.S. Senate AI Insight Forum on "Risk, Alignment, and Guarding Against Doomsday Scenarios" mentioned above, and has consistently emphasized that responsible-scaling commitments and pre-deployment evaluations are complements to, rather than substitutes for, government oversight of frontier AI.[35]

Other AI research

Beyond the headline scaling-laws and Constitutional AI work, Kaplan is a listed co-author on numerous other Anthropic and OpenAI research outputs, including studies of red-teaming and adversarial robustness, alignment evaluations, model behavior under reward hacking, and dataset curation effects on capability. His Google Scholar page lists more than 150 entries spanning physics and machine learning, with an h-index of 83 as of November 2026.[22]

What are Jared Kaplan's selected publications?

Kaplan's Google Scholar profile lists more than 150 research outputs across theoretical physics and machine learning and more than 180,000 total citations as of November 2026, with an h-index of 83.[22] A representative selection of his most cited or otherwise significant works follows.

Machine learning

  • Kaplan, J.; McCandlish, S.; Henighan, T.; Brown, T. B.; Chess, B.; Child, R.; Gray, S.; Radford, A.; Wu, J.; Amodei, D. "Scaling Laws for Neural Language Models." arXiv:2001.08361 (2020).[4]
  • Brown, T. B.; Mann, B.; Ryder, N.; Subbiah, M.; Kaplan, J.; Dhariwal, P.; et al. "Language Models are Few-Shot Learners." Advances in Neural Information Processing Systems 33 (2020).[24]
  • Chen, M.; Tworek, J.; Jun, H.; Yuan, Q.; Pinto, H. P. de O.; Kaplan, J.; et al. "Evaluating Large Language Models Trained on Code." arXiv:2107.03374 (2021).[25]
  • Bai, Y.; Jones, A.; Ndousse, K.; ...; Kaplan, J. "Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback." arXiv:2204.05862 (2022).[29]
  • Bai, Y.; Kadavath, S.; Kundu, S.; ...; Kaplan, J. "Constitutional AI: Harmlessness from AI Feedback." arXiv:2212.08073 (2022).[30]

Physics

  • Kaplan, J. D. "Aspects of Holography." Ph.D. thesis, Harvard University (2009).[9]
  • Fitzpatrick, A. L.; Kaplan, J.; Poland, D.; Simmons-Duffin, D. "The Analytic Bootstrap and AdS Superhorizon Locality." JHEP 12 (2013) 004; arXiv:1212.3616.[14]
  • Fitzpatrick, A. L.; Kaplan, J.; Walters, M. T. "Universality of Long-Distance AdS Physics from the CFT Bootstrap." JHEP 08 (2014) 145; arXiv:1403.6829.[15]
  • Fitzpatrick, A. L.; Kaplan, J. "Unitarity and the Holographic S-Matrix." JHEP 10 (2012) 032; arXiv:1112.4845.[17]
  • Kaplan, J. "Lectures on AdS/CFT from the Bottom Up." Lecture notes, Johns Hopkins University.[13]
  • Kaplan, J. "QFT Lectures Notes." Graduate quantum field theory course notes, Johns Hopkins University.[12]

What awards has Jared Kaplan received?

  • 2005 Hertz Fellow.[7]
  • 2014 Alfred P. Sloan Research Fellow in Physics.[18]
  • 2015 NSF CAREER Award (PHY-1454083).[7]
  • Principal Investigator, Simons Foundation Collaboration on the Nonperturbative Bootstrap.[3]
  • 2025 TIME100 AI list.[40]

How is Jared Kaplan viewed publicly?

Kaplan is among the more publicly visible Anthropic co-founders. Y Combinator hosted him in a 2025 AI Startup School talk and conversation titled "Scaling and the Road to Human-Level AI," in which he discussed the company's research roadmap and his projections about when transformative AI might arrive.[34] He appeared in a Salesforce conversation series during TrailblazerDX 2024 and on the TechCrunch "Equity" podcast in 2025, and Dwarkesh Patel's 2025 book "The Scaling Era: An Oral History of AI, 2019 to 2025" includes a previously unpublished interview with Kaplan.[28] Within the AI research community he is often credited, alongside Sam McCandlish and Dario Amodei, with introducing the scaling-laws framework that organizes how the field thinks about pretraining budgets.[34]

What is known about Jared Kaplan's personal life?

Kaplan resides in Pacifica, California, and has a son.[1] As of 2026, Forbes estimated his net worth at approximately $3.7 billion, reflecting his founding equity stake in Anthropic after successive funding rounds valued the company in the hundreds of billions of dollars.[1] In January 2026 Kaplan was among the seven Anthropic co-founders, alongside Dario Amodei, Daniela Amodei, Tom Brown, Jack Clark, Sam McCandlish, and Chris Olah, who collectively pledged to commit approximately eighty percent of their personal fortunes to addressing AI-driven inequality and to philanthropic efforts aligned with the company's stated mission.[36] Kaplan has continued to hold his Johns Hopkins faculty appointment in parallel with his Anthropic role, and he has continued to advise graduate students in physics at Hopkins while leading research at the company.[1][3]

ELI5: Who is Jared Kaplan?

Imagine you want to build a really smart robot brain, but building one is very expensive, so you want to know ahead of time how smart it will get before you spend the money. Jared Kaplan is a scientist who found a simple rule for this: if you make the brain bigger, give it more to read, and let it practice longer, it gets predictably better, like a recipe where doubling the ingredients gives you a known result.[4] He used to study black holes and the deepest math of physics, then switched to studying AI, and he helped start a company called Anthropic that makes the Claude chatbot and tries hard to keep AI safe.[1][2] Today he is one of the bosses of the science at that company, and he also helps decide whether a new AI is safe enough to release.[6]

Continue reading

More from See also

Topics:AnthropicPeople

References

  1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17Wikipedia contributors, "Jared Kaplan", Wikipedia, 2026. en.wikipedia.org/...Jared_Kaplan. Accessed 2026-05-25.
  2. ^1 ^2 ^3 ^4Anthropic, "About Anthropic", Anthropic, 2024. anthropic.com/company. Accessed 2026-05-25.
  3. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8Simons Foundation, "Jared Kaplan", Simons Foundation, n.d. simonsfoundation.org/...jared-kaplan. Accessed 2026-05-25.
  4. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10Kaplan, Jared; McCandlish, Sam; Henighan, Tom; Brown, Tom B.; Chess, Benjamin; Child, Rewon; Gray, Scott; Radford, Alec; Wu, Jeffrey; Amodei, Dario, "Scaling Laws for Neural Language Models", arXiv:2001.08361, 2020-01-23. arxiv.org/...2001.08361. Accessed 2026-05-25.
  5. ^OpenAI, "Scaling laws for neural language models", OpenAI, 2020-01-23. openai.com/...ling-laws-for-neural-language-models. Accessed 2026-05-25.
  6. ^1 ^2 ^3 ^4 ^5Anthropic, "Announcing our updated Responsible Scaling Policy", Anthropic, 2024-10-15. anthropic.com/...pdated-responsible-scaling-policy. Accessed 2026-05-25.
  7. ^1 ^2 ^3 ^4 ^5Hertz Foundation, "Jared Kaplan", Hertz Foundation, n.d. hertzfoundation.org/...jared-kaplan. Accessed 2026-05-25.
  8. ^1 ^2Harvard Department of Physics, "Harvard PhD Theses in Physics, 2001 to present", Harvard University, n.d. physics.harvard.edu/...phds. Accessed 2026-05-25.
  9. ^1 ^2Kaplan, Jared Daniel, "Aspects of Holography", Ph.D. dissertation, Harvard University, 2009. ui.adsabs.harvard.edu/...abstract. Accessed 2026-05-25.
  10. ^1 ^2 ^3Johns Hopkins University Department of Physics and Astronomy, "Jared Kaplan", Johns Hopkins University, n.d. physics-astronomy.jhu.edu/...jared-kaplan. Accessed 2026-05-25.
  11. ^Hertz Foundation, "Jared Kaplan biography and career summary", Hertz Foundation, n.d. hertzfoundation.org/...jared-kaplan. Accessed 2026-05-25.
  12. ^1 ^2 ^3Kaplan, Jared, "QFT Lectures Notes", Johns Hopkins University, 2016. sites.krieger.jhu.edu/...QFTNotes.pdf. Accessed 2026-05-25.
  13. ^1 ^2Kaplan, Jared, "Lectures on AdS/CFT from the Bottom Up", Johns Hopkins University, n.d. stat.ucla.edu/...AdSCFT.pdf. Accessed 2026-05-25.
  14. ^1 ^2Fitzpatrick, A. Liam; Kaplan, Jared; Poland, David; Simmons-Duffin, David, "The Analytic Bootstrap and AdS Superhorizon Locality", arXiv:1212.3616, Journal of High Energy Physics 12 (2013) 004, 2013. arxiv.org/...1212.3616. Accessed 2026-05-25.
  15. ^1 ^2Fitzpatrick, A. Liam; Kaplan, Jared; Walters, Matthew T., "Universality of Long-Distance AdS Physics from the CFT Bootstrap", arXiv:1403.6829, Journal of High Energy Physics 08 (2014) 145, 2014. arxiv.org/...1403.6829. Accessed 2026-05-25.
  16. ^1 ^2Fitzpatrick, A. Liam; Kaplan, Jared; Penedones, Joao; Raju, Suvrat; van Rees, Balt C., "A Natural Language for AdS/CFT Correlators", arXiv:1107.1499, Journal of High Energy Physics 11 (2011) 095, 2011. arxiv.org/...1107.1499. Accessed 2026-05-25.
  17. ^1 ^2 ^3Fitzpatrick, A. Liam; Kaplan, Jared, "Unitarity and the Holographic S-Matrix", arXiv:1112.4845, Journal of High Energy Physics 10 (2012) 032, 2012. arxiv.org/...1112.4845. Accessed 2026-05-25.
  18. ^1 ^2Johns Hopkins University Department of Physics and Astronomy, "Jared Kaplan and Tyrel McQueen Selected for Sloan Research Fellowship", Johns Hopkins University, 2014-02-18. physics-astronomy.jhu.edu/...n-research-fellowship. Accessed 2026-05-25.
  19. Wikipedia contributors, "Anthropic", Wikipedia, 2026. en.wikipedia.org/...Anthropic. Accessed 2026-05-25.
  20. ^Brown, Tom B.; et al., "Language Models are Few-Shot Learners", arXiv:2005.14165, 2020-05-28. arxiv.org/...2005.14165. Accessed 2026-05-25.
  21. ^1 ^2OpenAI, "Language models are few-shot learners", OpenAI, 2020-05-28. openai.com/...language-models-are-few-shot-learners. Accessed 2026-05-25.
  22. ^1 ^2 ^3 ^4 ^5 ^6 ^7Google Scholar, "Jared Kaplan citation profile", Google Scholar, n.d. scholar.google.com/citations. Accessed 2026-05-25.
  23. ^1 ^2 ^3 ^4Hoffmann, Jordan; Borgeaud, Sebastian; et al., "Training Compute-Optimal Large Language Models", arXiv:2203.15556, 2022-03-29. arxiv.org/...2203.15556. Accessed 2026-05-25.
  24. ^1 ^2 ^3Brown, Tom B.; Mann, Benjamin; Ryder, Nick; Subbiah, Melanie; Kaplan, Jared; Dhariwal, Prafulla; et al., "Language Models are Few-Shot Learners", arXiv:2005.14165, NeurIPS 2020, 2020. arxiv.org/...2005.14165. Accessed 2026-05-25.
  25. ^1 ^2 ^3Chen, Mark; Tworek, Jerry; Jun, Heewoo; Yuan, Qiming; Pinto, Henrique Ponde de Oliveira; Kaplan, Jared; et al., "Evaluating Large Language Models Trained on Code", arXiv:2107.03374, 2021-07-07. arxiv.org/...2107.03374. Accessed 2026-05-25.
  26. ^1 ^2Wikipedia contributors, "Dario Amodei", Wikipedia, 2026. en.wikipedia.org/...Dario_Amodei. Accessed 2026-05-25.
  27. ^Contrary Research, "Anthropic: Business Breakdown and Founding Story", Contrary Research, 2024. research.contrary.com/...anthropic. Accessed 2026-05-25.
  28. ^Patel, Dwarkesh; Leech, Gavin, "The Scaling Era: An Oral History of AI, 2019 to 2025", Stripe Press, 2025. harvard.com/...9781953953551. Accessed 2026-05-25.
  29. ^1 ^2Bai, Yuntao; Jones, Andy; Ndousse, Kamal; et al., "Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback", arXiv:2204.05862, 2022-04-12. arxiv.org/...2204.05862. Accessed 2026-05-25.
  30. ^1 ^2 ^3 ^4 ^5 ^6Bai, Yuntao; Kadavath, Saurav; Kundu, Sandipan; et al., "Constitutional AI: Harmlessness from AI Feedback", arXiv:2212.08073, 2022-12-15. arxiv.org/...2212.08073. Accessed 2026-05-25.
  31. ^1 ^2Anthropic, "Claude's Constitution", Anthropic, 2023-05-09. anthropic.com/...claudes-constitution. Accessed 2026-05-25.
  32. ^1 ^2Life with Machines, "From Scaling Laws to Safe AI: Anthropic's Jared Kaplan in conversation", Life with Machines, 2024. lifewithmachines.media/...ws-to-safe-ai-anthropics. Accessed 2026-05-25.
  33. ^1 ^2 ^3Anthropic, "Mapping the Mind of a Large Language Model", Anthropic, 2024-05-21. anthropic.com/...mapping-mind-language-model. Accessed 2026-05-25.
  34. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14Y Combinator, "Scaling and the Road to Human-Level AI: Anthropic Co-founder Jared Kaplan", Y Combinator Startup Library, 2025-07-29. ycombinator.com/...thropic-co-founder-jared-kaplan. Accessed 2026-05-25.
  35. ^1 ^2 ^3 ^4Kaplan, Jared, "Written Statement for AI Insight Forum: Risk, Alignment, and Guarding Against Doomsday Scenarios", United States Senate, 2023. schumer.senate.gov/...20Kaplan%20-%20Statement.pdf. Accessed 2026-05-25.
  36. ^Lifestyles Magazine, "Anthropic's seven cofounders commit 80% of their fortunes to combat AI-driven inequality", Lifestyles Magazine, 2026-01-28. lifestylesmagazine.com/...ortunes-to-combat-ai-dri. Accessed 2026-05-25.
  37. ^Hernandez, Danny; Kaplan, Jared; Henighan, Tom; McCandlish, Sam, "Scaling Laws for Transfer", arXiv:2102.01293, 2021-02-02. arxiv.org/...2102.01293. Accessed 2026-05-25.
  38. ^Kaplan, Jared, "Declaration of Jared Kaplan", United States District Court for the Northern District of California, Case No. 3:26-cv-01996, 2026. storage.courtlistener.com/...s.cand.465515.6.1.pdf. Accessed 2026-05-25.
  39. ^Perrigo, Billy, "Exclusive: Anthropic Drops Flagship Safety Pledge", TIME, 2026-02-24. time.com/...anthropic-drops-flagship-safety-pledge. Accessed 2026-11-04.
  40. ^Perrigo, Billy, "Jared Kaplan: The 100 Most Influential People in AI 2025", TIME, 2025. time.com/...jared-kaplan. Accessed 2026-11-04.
  41. ^Coldewey, Devin, "Anthropic is the new AI research outfit from OpenAI's Dario Amodei, and it has $124M to burn", TechCrunch, 2021-05-28. techcrunch.com/...o-amodei-and-it-has-124m-to-burn. Accessed 2026-11-04.
  42. ^Anthropic, "Rahul Patil joins Anthropic as Chief Technology Officer", Anthropic, 2025-10-07. anthropic.com/...rahul-patil-joins-anthropic. Accessed 2026-11-04.

Improve this article

Add missing citations, update stale details, or write a clearer explanation. Anyone can edit: changes go live when you publish, and every edit is checked by an editor.

8 revisions · v8 · 5,423 words · full history

Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify

Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here

Reviewer note: Independent fact-check of the full article against its cited sources (150 claims, 38 sources fetched); 13 material + 27 minor fixed

Cite this page: AI Wiki. "Jared Kaplan." aiwiki.ai, updated 28 Oct 2026, fact-checked 4 Nov 2026. CC BY 4.0. https://aiwiki.ai/wiki/jared_kaplan

Edit article