Richard Socher

CEO & Founder — You.com

Profile

Richard Socher is one of the researchers who pulled natural language processing into the deep learning era — and now he’s trying to use what he learned to rebuild search from scratch. His 2014 Stanford PhD under Chris Manning won the department’s best thesis award and laid groundwork for how modern models understand language. He co-authored GloVe, the word vector method that sat alongside word2vec as the default representation for a generation of NLP systems, and his recursive neural network papers were among the first to show that tree-structured deep learning could actually parse and classify real sentences.

After his PhD he founded MetaMind, an NLP startup acquired by Salesforce in 2016 for roughly $32.8M. He then spent four years as Salesforce Chief Scientist and EVP, where he and Bryan McCann published The Natural Language Decathlon — the paper that introduced what we now call prompt engineering, framing every NLP task as a question in natural language years before GPT-3 made the idea famous.

In 2020 he left to start You.com, betting that search was about to get rewritten by language models. You.com was the first mainstream search engine to wire an LLM directly into the results page — beating Google, Microsoft, and Perplexity to market. In February 2025 they launched ARI (Advanced Research & Insights), a deep-research agent aimed at enterprises that claims to analyze 400+ sources in parallel to produce cited reports in minutes. The consumer search battle is crowded; the enterprise research angle is where Socher is placing his chips.

He also runs AIX Ventures, an AI-focused early-stage fund (~$200M Fund II), and still teaches occasionally at Stanford. For developers learning AI, Socher is worth following because he sits at an unusual intersection: a genuine first-wave deep-learning researcher who is now a founder building a product in the most competitive market in AI. His takes on what actually works in production tend to be grounded rather than hyped.

Key Articles & Papers

GloVe: Global Vectors for Word Representation 2014 — With Pennington and Manning — one of the two dominant pre-transformer word embedding methods. If you used NLP between 2014 and 2018, you probably used GloVe. Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank 2013 — Introduced the Stanford Sentiment Treebank and the recursive neural tensor network. An influential early demonstration that deep learning could handle compositional language. The Natural Language Decathlon: Multitask Learning as Question Answering 2018 — With Bryan McCann. The paper often credited with introducing prompt engineering — reframing every NLP task as a natural language question answered by one shared model. Learned in Translation: Contextualized Word Vectors (CoVe) 2017 — An early attempt at contextualized embeddings via a pretrained translation encoder. A direct predecessor to ELMo and BERT. Ask Me Anything: Dynamic Memory Networks for Natural Language Processing 2015 — A MetaMind-era architecture for question answering with episodic memory. Influential in early memory-augmented network research. Regularizing and Optimizing LSTM Language Models 2017 — AWD-LSTM. The language model that was state-of-the-art before transformers took over, and the foundation of early fast.ai NLP work. CTRL: A Conditional Transformer Language Model for Controllable Generation 2019 — Salesforce's 1.6B parameter controllable language model — one of the largest open LMs of its era and an early exploration of control codes for generation. Introducing ARI: The First Professional-Grade Research Agent for Business 2025 — You.com's pivot toward enterprise deep research. The clearest articulation of where Socher thinks AI search is going.

Videos

Controversies

No significant public controversies. You.com has faced the usual competitive scrutiny — debates over AI search accuracy, citation quality, and whether any startup can meaningfully challenge Google — but nothing rising to the level of a personal controversy around Socher.