How AI got here,
in one library.

The first comprehensive horizontal AI law, introducing risk tiers and obligations with phased enforcement.

A wide-ranging executive order setting US federal direction on AI safety, security, and governance.

A large multimodal model with markedly stronger reasoning and exam performance than its predecessors.

A family of capable language models with openly available weights, catalyzing the open-weights ecosystem.

A conversational interface to a large language model that reached mass adoption and brought AI into mainstream use.

An open speech-recognition model trained on large-scale weak supervision, robust across many languages.

An openly released text-to-image diffusion model that brought image generation to a wide audience.

Showed that many large models were undertrained, reframing how compute should be split between model size and data.

Achieved a breakthrough in predicting protein 3D structure, a decades-old grand challenge in biology.

A multitask knowledge benchmark across 57 subjects, widely used to compare language-model capabilities.

A key paper establishing diffusion models as a powerful approach to high-quality image generation.

Showed that scaling language models to 175B parameters produced strong few-shot abilities, popularizing large-scale LLMs.

A bidirectional Transformer pretraining method that set a new standard for many language-understanding tasks.

Introduced the Transformer architecture, which became the foundation of nearly all modern large language models.

AlphaGo beat a top human Go player, a landmark for reinforcement learning and search in a game long thought hard for AI.

Introduced residual connections, enabling much deeper networks and influencing architectures across AI.

Introduced GANs, training two networks against each other to generate realistic data, a foundational idea for generative models.

A deep convolutional neural network dramatically cut the ImageNet error rate, widely seen as the spark of the modern deep-learning era.