Andrej Karpathy

Andrej Karpathy's indexed work centers on large-scale video classification with convolutional neural networks, a first-authored 2014 study with over 1,200 citations that established foundational methods for learning spatiotemporal features. He also developed deep visual-semantic alignments for generating image descriptions, a 2016 first-authored paper with over 1,000 citations that connected visual representations to natural language, and contributed visualizing and understanding recurrent networks, a 2015 first-authored work with over 870 citations that analyzed the internal dynamics of sequence models. These papers collectively advanced the integration of vision and language through neural architectures. Among Karpathy's additional contributions, he was a co-author on Grounded Compositional Semantics for Finding and Describing Images with Sentences, a 2014 paper with over 820 citations that linked compositional language to visual grounding. He also contributed to PixelCNN++, a 2017 work improving autoregressive image generation with discretized logistic mixture likelihoods. His more recent outputs include autoresearch, a 2026 first-authored system for AI agents running single-GPU nanochat training, and Bibby AI