Richard Socher

Richard Socher was born in Dresden in 1983.⁠[1]

He completed his Abitur at Gymnasium Dresden-Plauen in 2003, then studied computational linguistics, completing his Leipzig UniversityLeipzig UniversityLeipzig University is a comprehensive university in Leipzig, Germany, founded in 1409. Its teaching and research span humanities, social sciences, natural and life sciences, and medicine. Richard Socher studied computational linguistics there.Open the full entry studies in 2006 and a master’s degree at Saarland UniversitySaarland UniversitySaarland University is a university in Germany’s Saarland state, founded in 1948 with French-German educational roots. Its research priorities include computer science, nano and life sciences, and European studies.Open the full entry in 2008.⁠[2][3]

Originally from Dresden, Germany, Socher studied at Leipzig University and Saarland University, receiving bachelor’s and master’s degrees before pursuing a computer science PhD at Stanford University under Andrew NgAndrew NgAndrew Ng is a British-born computer scientist and AI educator. A former director of the Stanford Artificial Intelligence Lab, he helped found Google Brain and Coursera, served as Baidu’s chief scientist from 2014 to 2017, and later founded DeepLearning.AI, LandingAI and AI Fund.Open the full entry and Christopher ManningChristopher ManningChristopher Manning is a natural language processing researcher and Stanford professor emeritus, a former Stanford AI Lab director and an AIX Ventures investor. He collaborated with Richard Socher on recursive neural networks, sentiment analysis and word vectors.Open the full entry. During a 2007 visit to Princeton UniversityPrinceton UniversityPrinceton University is a private university and Ivy League member in Princeton, New Jersey. Chartered on October 22, 1746 as the College of New Jersey, it moved to Princeton in 1756 and adopted its present name in 1896. Undergraduate education leads to A.B. and B.S.E. degrees and emphasizes small-group discussion, independent research and residential college life. Graduate education spans the humanities, social sciences, natural sciences and engineering. As of October 2026, Christopher L. Eisgruber was president. Its 2026–27 operating budget was $3.407 billion, including $342 million for undergraduate aid.Open the full entry, he worked on robot vision, bringing language and visual understanding together.⁠[4][3]

Socher co-authored the 2009 CVPR paper ImageNet: A Large-Scale Hierarchical Image Database with Jia Deng, Wei Dong, Li-Jia Li, Kai Li and Fei-Fei Li. The work developed an image database organized by a semantic hierarchy and formed part of his early computer vision research.⁠[5]

Socher, Cliff Chiung-Yu Lin, Andrew Ng and Christopher Manning applied recursive neural networks to parsing natural scenes and language, learning compositional representations for hierarchical image regions or sentence constituents. Their paper received the ICML 2011 Distinguished Application Paper award.⁠[6][7]

Socher received a MicrosoftMicrosoftMicrosoft is a technology company founded by Bill Gates and Paul Allen in 1975 and headquartered in Redmond, Washington, with businesses spanning Windows, Microsoft 365, Azure, enterprise software, artificial intelligence, gaming and devices.Open the full entry Research PhD Fellowship in 2012 and a Magic Grant from the Brown Institute for Media Innovation in 2013.⁠[8]

In 2013, Socher, Alex Perelygin, Jean Wu, Jason Chuang, Christopher Manning, Andrew Ng and Christopher PottsChristopher PottsChristopher Potts is a Stanford linguistics professor working on semantics, pragmatics and computational linguistics, and a co-author of the Stanford Sentiment Treebank paper.Open the full entry published the Stanford Sentiment Treebank research. The team annotated fine-grained sentiment for phrases in movie-review parse trees and introduced a recursive neural tensor network to model how composition, negation and contrast affect sentiment.⁠[9]

Socher completed his Stanford PhD in 2014 with a dissertation on recursive deep learning for natural language processing and computer vision, receiving the university’s computer science doctoral dissertation award. He later served as an adjunct professor in Stanford’s computer science department and taught deep learning for natural language processing.⁠[3][10]

In 2014, he introduced GloVe with Jeffrey PenningtonJeffrey PenningtonJeffrey Pennington is a machine learning researcher whose work moved from theoretical particle physics to word representations and deep learning theory. He co-authored GloVe with Richard Socher and Christopher Manning.Open the full entry and Christopher Manning. It learns vector representations from global word co-occurrence statistics, allowing some semantic relationships to be expressed as vector relationships. The project released training code and pretrained word vectors.⁠[11]

MetaMind publicly launched its AI platform in December 2014, with Socher as co-founder and CTO and Sven StrohbandSven StrohbandSven Strohband is an engineer, entrepreneur and venture investor, and a partner and managing director at Khosla Ventures. He helped lead the Stanley autonomous-car project and co-founded MetaMind with Richard Socher.Open the full entry as CEO. Backed by $8 million in initial funding from Khosla VenturesKhosla VenturesKhosla Ventures (KV) is an American venture capital firm founded in 2004 by Vinod Khosla, the founding CEO of Sun Microsystems, and headquartered in Menlo Park, California. It made heavy early bets on clean technology, became OpenAI’s first venture capital investor in 2019, and its portfolio also includes companies such as DoorDash, Block (formerly Square), Impossible Foods, and Rocket Lab.Open the full entry and Marc Benioff, it applied deep learning to text and image classification and offered cloud tools for businesses to train predictive models.⁠[12]

His Tree-LSTM paper with Kai Sheng Tai and Christopher Manning extended long short-term memory networks to tree structures, evaluating sentence representations on semantic relatedness and sentiment classification. The same year’s dynamic memory network research used a question to guide repeated input selection and episodic-memory updates across question answering, sentiment analysis and part-of-speech tagging.⁠[13][14]

In 2016, Caiming XiongCaiming XiongCaiming Xiong is an AI researcher and co-founder of Recursive, formerly senior vice president of AI research at Salesforce.Open the full entry, Stephen Merity and Socher extended dynamic memory networks to visual and textual question answering. Socher taught Stanford’s spring 2016 CS224d course, covering word vectors, recurrent and recursive networks, with practical work implementing, training and debugging models.⁠[15][16]

Salesforce acquired MetaMind in April 2016. Socher subsequently served as chief scientist and executive vice president, leading fundamental and applied research, product incubation, search, service automation and a cross-product AI platform for structured and unstructured data.⁠[8][10]

In Learned in Translation, Bryan McCannBryan McCannBryan McCann is an AI researcher, entrepreneur and philosopher, and a You.com co-founder, director and adviser. He previously served as its CTO and as a research lead at Salesforce Research.Open the full entry, James Bradbury, Caiming Xiong and Socher used a machine-translation encoder to produce context-dependent word representations called CoVe, transferring them to classification, textual entailment and question answering.⁠[17]

Seq2SQL, by Victor Zhong, Caiming Xiong and Socher, translated natural-language questions into SQL queries using query structure and execution feedback during training. The team also released WikiSQL, connecting language understanding with executable database operations.⁠[18]

In 2018, Socher, Bryan McCann, Nitish Shirish Keskar and Caiming Xiong published the Natural Language Decathlon paper, casting ten language tasks as question answering over a context and studying a multitask question answering network. The work explored shared task representations and multitask learning.⁠[19]

The 2009 ImageNet paper received the PAMI Longuet-Higgins Prize at CVPR 2019; Socher was listed among its co-authors.⁠[20]

Socher co-authored CTRL, a 1.63-billion-parameter conditional Transformer that used control codes to guide the domain, style or task of generated text. The team released model weights and investigated more explicit conditioning of language generation.⁠[21]

In ProGen, Socher and colleagues applied language modeling to proteins, training a 1.2-billion-parameter model on roughly 280 million sequences and conditioning generation on functional or taxonomic tags. The initial paper primarily evaluated sequences computationally; it was protein-design research rather than evidence of an approved medicine.⁠[22]

The AI Economist, co-authored by Socher, used two-level reinforcement learning for simulated economic agents and a tax-policy planner, examining trade-offs between productivity and equality. Its results came from a simulation, rather than the adoption of those policies by a country.⁠[23]

In 2020, Socher left Salesforce and founded You.com with Bryan McCann, continuing to combine language models with search. You.com opened its public beta and announced $20 million in funding on November 9, 2021.⁠[24][25]

He also co-authored a Nature Communications study investigating prediction of hormone-receptor status from routine hematoxylin-and-eosin-stained breast-cancer tissue images. This collaboration with medical researchers studied pathology-image analysis; model performance alone does not establish suitability for unvalidated treatment decisions.⁠[26]

Alongside running a search company, Socher founded AIX Ventures and became a managing partner, investing in and advising AI startups. AIX lists him as a co-founder and managing partner, with an investment team that includes researchers and entrepreneurs.⁠[10][27]

On March 23, 2022, AIX introduced its $50 million first fund for early-stage AI companies. Its announcement named investment partners including Socher, Pieter AbbeelPieter AbbeelPieter Abbeel is an artificial intelligence and robotics researcher from Belgium, a professor at UC Berkeley and director of its Robot Learning Lab, known for research on robot apprenticeship learning and deep reinforcement learning. He co-founded Gradescope and Covariant and joined Amazon with part of the Covariant team in 2024.Open the full entry, Anthony GoldbloomAnthony GoldbloomAnthony Goldbloom is an Australian-born entrepreneur and the co-founder and former chief executive of Kaggle, a data science competition platform. After Google acquired Kaggle in 2017, he continued to lead Kaggle within Google until 2022; he is now co-founder and CEO of the sales intelligence company Sumble and an investment partner at AIX Ventures.Open the full entry and Christopher Manning, alongside full-time co-founder Shaun JohnsonShaun JohnsonShaun Johnson is a technology executive and investor who co-founded AIX Ventures and previously held product, design and engineering leadership roles at Lilt and NimbleRx. He left AIX in December 2025.Open the full entry. Socher also reflected on his earlier support for AI startups.⁠[28]

The 2013 Stanford Sentiment Treebank paper received the ACL 2023 Test-of-Time Paper Award for its lasting research impact. That September, Socher was included in TIME100 AI.⁠[29][30]

On February 13, 2024, AIX announced a $202 million second fund for AI applications and infrastructure, with plans to lead more funding rounds.⁠[28]

On April 26, 2024, TU DresdenTU DresdenTU Dresden is a research university in Dresden, Germany, originating in a technical educational institution founded in 1828, with work spanning engineering, natural sciences, medicine, humanities and social sciences.Open the full entry awarded Socher an honorary doctorate for his AI research and industrial applications. It is an honorary degree, distinct from the research doctorate he earned at Stanford in 2014.⁠[3]

You.com announced a $50 million Series B led by Georgian on September 4, 2024, emphasizing AI assistance, research and custom agents for knowledge workers.⁠[31]

On September 3, 2025, You.com announced a $100 million Series C at a $1.5 billion valuation, led by Cox EnterprisesCox EnterprisesCox Enterprises is an American family-owned business group headquartered in Atlanta.Open the full entry with participation from Georgian, Salesforce VenturesSalesforce VenturesSalesforce Ventures is Salesforce’s corporate venture capital arm. Active since 2009, it backs companies from seed through growth across enterprise software, cloud infrastructure, data, security and AI. Salesforce supplies all of its capital.Open the full entry and Norwest. It increasingly positioned its business as search and research infrastructure connecting enterprise models and AI agents to live web information.⁠[32]

On March 17, 2026, Socher announced founding engineer Saahil JainSaahil JainSaahil Jain is You.com’s CTO. He joined as a founding engineer in 2020 and subsequently led engineering and AI teams, after researching NLP and medical machine learning at Stanford.Open the full entry as You.com’s next CTO. Co-founder Bryan McCann became an adviser and remained on its board.⁠[33]

On May 13, 2026, Socher publicly launched RecursiveRecursiveRecursive (Recursive Superintelligence) is a company pursuing automated AI research and recursive self-improvement, co-founded by Richard Socher and seven others. It emerged publicly in May 2026, released training and GPU-kernel optimization experiments, and signed a multi-year compute agreement with AWS.Open the full entry as co-founder and CEO. The company studies recursive self-improvement and automated knowledge discovery, initially targeting AI research before other fields. His website continued to list roles at Recursive, You.com and AIX Ventures in November 2026.⁠[34][35][10]

The other seven co-founders are Tim RocktäschelTim RocktäschelTim Rocktäschel is a German-born artificial intelligence researcher who is Professor of Artificial Intelligence at University College London and co-founder and chief scientist of Recursive Superintelligence; he previously led the Open-Endedness team at Google DeepMind.Open the full entry, Alexey DosovitskiyAlexey DosovitskiyAlexey Dosovitskiy is a computer vision and machine learning researcher and a Recursive co-founder, with contributions to FlowNet, CARLA and Vision Transformer.Open the full entry, Josh TobinJosh TobinJosh Tobin is a machine learning researcher and entrepreneur who was a research scientist at OpenAI, co-founded the machine learning platform company Gantry, and is a member of the founding team of the AI research company Recursive Superintelligence.Open the full entry, Caiming Xiong, Yuandong TianYuandong TianYuandong Tian is an artificial intelligence researcher, science-fiction writer, and co-founder of Recursive Superintelligence. Formerly a research scientist director at Meta FAIR, he has worked on Go AI, long-context inference, efficient training, latent-space reasoning, and neural-network learning dynamics. His research includes ELF OpenGo, StreamingLLM, GaLore, and Coconut.Open the full entry, Tim ShiTim ShiTim Shi (Tianlin Shi) is an artificial intelligence researcher and entrepreneur who worked on web agents in OpenAI’s early days, co-founded the conversational AI company Cresta in 2017 and served as its CTO, and later co-founded Recursive Superintelligence.Open the full entry and Jeff CluneJeff CluneJeff Clune is an artificial intelligence researcher who is a professor of computer science at the University of British Columbia and a Canada CIFAR AI Chair at the Vector Institute; he works on open-ended algorithms and “AI-generating algorithms,” and in 2026 co-founded Recursive.Open the full entry. GV disclosed $650 million in early funding at a $4.65 billion valuation and was one of the co-lead investors.⁠[36]

On June 11, Recursive published early results from an automated AI research system that proposes ideas, implements code, runs experiments and validates results before choosing the next experiment. Tests covered small language-model training and GPU-kernel optimization. The company released artifacts but noted that underlying models may have seen public techniques, so results do not necessarily demonstrate independent rediscovery or general superintelligence.⁠[37]

On July 28, Recursive announced a $410 million multiyear agreement with Amazon Web Services for cloud computing to run its automated research system, with plans to co-develop infrastructure for large-scale AI research. Socher emphasized computing scale as a condition for expanding research loops.⁠[38]

In a Nicholas Thompson interview published on September 9, 2026, Socher discussed the uses and risks of self-improving AI. In a subsequent excerpt, he stressed goal and reward design: systems can exploit metrics while appearing to succeed, requiring alignment, validation and better objective definitions. These were his proposed directions for safety research.⁠[39][40]

PublicAffairs published The Eureka Machine: Why AI Is the Key to Unlocking a New Era of Scientific Discoveries on September 22, 2026. The 288-page book explores AI and scientific discovery, including Socher’s predictions about self-improving systems. Kirkus praised its clear explanations and optimism while suggesting skeptics might find it insufficiently attentive to losses of human agency.⁠[41][42]

In Fortune AI Weekly, Socher connected his book with Recursive: after finishing the manuscript the previous year, he wanted to build a system that could advance discovery. He emphasized the eight founders’ varied experience in robotics, open-ended learning, language models and vision, and described a loop for proposing, implementing and validating research ideas beyond coding assistance.⁠[43]

He said an early system was working but still required greater capability and scale to reach his intended level of recursive self-improvement, distinguishing the company’s ambition from completed experimental results.⁠[43]

On AI risk, Socher gave his personal assessment that the probability of complete human extinction caused by AI is zero. He distinguished extinction from severe harms involving many deaths, citing constraints on compute and physical resources, defenders’ capabilities and commercial incentives. He also argued for constraining irresponsible organizations. These are his views, not a scientific consensus or a verified safety guarantee.⁠[43]

For scientific discovery, he proposed combining knowledge in language models, scientific measurements, simulation and automated experiments, with agents collaborating on new questions. He anticipated faster progress in medicines, materials, solar power and batteries, and changes in research work toward new questions and higher-level creative tasks. These were forecasts about the future.⁠[43]

Recursive released Flint in October 2026 as a mixture-of-experts language model for coding agents. Its model card specifies 39.8 billion total parameters and about 2.5 billion active per token, with text input and output, reasoning and multi-turn tool use; weights are released under Apache 2.0. The company describes it as a product of human and automated research. Its published evaluations are developer assessments, not independent certification.⁠[44][45]

Socher supported an AI learning center at his former school, Gymnasium Dresden-Plauen, which named it after him. He also participated in public lectures and research exchanges at TU Dresden, connecting academic work with AI education for students and the public.⁠[2][3]

Socher uses @RichardSocher on X at https://x.com/RichardSocher (opens in a new window). His personal website, https://www.socher.org/ (opens in a new window), provides research papers, professional information and photography.⁠[46][10]