Richard Socher

CEO & Cofounder — You.com Founder — Recursive Superintelligence Founder & Managing Director — AIX Ventures Former Chief Scientist — Salesforce

Listen — profile

0:00 / 3:49

Profile

Richard Socher is one of the people who quietly built the plumbing of modern language AI, then spent the last five years trying to turn that research pedigree into products. Before “large language model” was a household phrase, Socher was among the most-cited researchers in natural language processing — his Stanford work on recursive neural networks, sentiment analysis, and word vectors sits underneath a decade of everything that followed. If you have ever used GloVe embeddings, trained on the Stanford Sentiment Treebank, or leaned on the notion that words can be represented as dense vectors carrying meaning, you have used Socher’s fingerprints. He earned his PhD at Stanford in 2014 (a best-thesis winner), co-created GloVe with Jeffrey Pennington and his advisor Chris Manning, and later co-taught the legendary CS224n deep-learning-for-NLP course that trained a generation of practitioners.

The commercial arc started with MetaMind, the AI startup he founded in 2014 and sold to Salesforce in 2016. There he became Chief Scientist, running research, product incubation, and the company’s AI platform until 2020 — a stint that produced influential work like contextual word vectors (CoVe), the decaNLP multitask benchmark, and the CTRL controllable language model. For developers, this is the useful part of his story: Socher is one of the rare people who has lived the full pipeline from arXiv paper to shipped enterprise product, and his research reads like a preview of ideas — contextual embeddings, multitask learning, controllable generation — that the transformer era later industrialized.

Today he runs two things at once. You.com, which he co-founded in 2020, started as a consumer AI search engine positioned against Google and has since pivoted hard toward enterprise: it hit unicorn status in September 2025 with a $1.5B valuation, and its ARI research agent — which fans out across hundreds of sources to write cited reports — made TIME’s Best Inventions of 2025. His newer and far more ambitious bet is Recursive Superintelligence, which came out of stealth in May 2026 having raised roughly $650M at a $4.65B valuation from GV, Greycroft, Nvidia, and AMD. Its pitch is exactly what the name says: AI that identifies its own weaknesses, writes its own benchmarks, and rewrites its own code to get better — recursive self-improvement as an engineering program rather than a thought experiment. The founding bench is stacked (Tim Rocktäschel, Jeff Clune, Josh Tobin, Caiming Xiong, Yuandong Tian, Alexey Dosovitskiy, Tim Shi).

Read Socher with clear eyes. The research credentials are genuine and foundational — this is not a hype merchant who arrived in 2023. But a multi-billion-dollar valuation for a pre-product lab whose thesis is “self-improving superintelligence” invites healthy skepticism, and the You.com pivot from consumer search to enterprise infrastructure is a reminder that even great researchers don’t automatically win markets. He’s also an active investor: through AIX Ventures, the “by builders, for builders” fund he founded with Manning and Anthony Goldbloom, he backed early rounds of Perplexity (Aravind Srinivas), Hugging Face (Clément Delangue), and Weights & Biases. For someone learning AI, Socher is worth studying as a case of how deep NLP fundamentals translate — and sometimes don’t cleanly translate — into companies.