Yoshua Bengio
Helped lay the statistical foundations of modern language modeling and co-founded the Montreal deep learning school — now a leading voice on AI safety.
Yoshua Bengio (b. 1964) built much of the statistical scaffolding language models still rely on. His 2003 paper "A Neural Probabilistic Language Model" was among the first to represent words as dense vectors — embeddingsEmbeddingAn embedding is a learned vector of numbers representing a token, word, or passage of text such that similar meanings end up close together in vector space. — learned jointly with a neural network trained to predict the next word, rather than the sparse, hand-engineered representations statistical NLP had used until then. It's a direct conceptual ancestor of every LLMLLM (Large Language Model)An LLM is a large transformer trained to predict the next token on massive text corpora, then fine-tuned to follow instructions — the architecture behind GPT, Claude, Gemini, and Llama.'s embedding layer and tokenizationTokenizationTokenization converts raw text into a sequence of integers a model can process, typically via subword schemes like byte-pair encoding (BPE). pipeline.
Alongside Geoffrey HintonGeoffrey HintonKnown as a godfather of deep learning — co-authored backpropagation in 1986, co-invented the Boltzmann machine and dropout, and co-authored AlexNet, the 2012 result that restarted the field. and Yann LeCunYann LeCunInvented the convolutional neural network in the late 1980s and deployed it commercially reading handwritten checks years before deep learning was fashionable., Bengio helped keep neural networkNeural NetworkA neural network is layers of simple weighted-sum-plus-nonlinearity units (neurons) chained together, trained by gradient descent and backpropagation. research alive through the 1990s and 2000s, when it was out of fashion, from a base at the University of Montreal that grew into one of the world's most influential deep learningDeep LearningDeep learning is machine learning using multi-layer neural networks, which learn their own features from raw data instead of relying on hand-engineered ones. research clusters (Mila). His lab also contributed early work on attentionAttention (Self-Attention)Attention is a mechanism letting each position in a sequence weigh every other position via learned Query/Key/Value vectors, forming the core of the transformer. mechanisms in sequence models and on generative adversarial networksGAN (Generative Adversarial Network)A GAN trains a generator and a discriminator against each other, the generator learning to fool the discriminator into mistaking its output for real data., both foundational to later architectures. He shared the 2018 Turing Award with Hinton and LeCun.
More recently, Bengio has become one of the most prominent researchers publicly warning about AI safetyAI SafetyAI safety is the question of whether a system's own behavior matches what its designers actually want, independent of any attacker. and existential risk, co-authoring the "International AI Safety Report" and arguing that capability research has outpaced the field's ability to ensure systems remain controllable — a notable reversal for someone whose earlier career was spent trying to make those systems more capable.
See also: Geoffrey HintonGeoffrey HintonKnown as a godfather of deep learning — co-authored backpropagation in 1986, co-invented the Boltzmann machine and dropout, and co-authored AlexNet, the 2012 result that restarted the field., Yann LeCunYann LeCunInvented the convolutional neural network in the late 1980s and deployed it commercially reading handwritten checks years before deep learning was fashionable.
Learn more: LLMs · Wikipedia: Yoshua Bengio
Yann LeCun
Invented the convolutional neural network in the late 1980s and deployed it commercially reading handwritten checks years before deep learning was fashionable.
Fei-Fei Li
Built ImageNet, the large labeled dataset that made the 2012 deep learning breakthrough in computer vision possible in the first place.