ai

Whisper, an automatic speech recognition system released by OpenAI in 2022, was trained on 680,000 hours of multilingual audio.

Reinforcement learning was famously used by DeepMind to train Atari-playing agents directly from pixel input in 2013.

AlphaZero learned chess, shogi, and Go to superhuman levels purely through self-play, as published in 2018.

ResNet, introduced in 2015, used skip connections to enable training of very deep convolutional networks.

Batch normalization, introduced in 2015, accelerates training of deep neural networks by normalizing layer inputs.

Natural language processing combines computational linguistics with machine learning to enable computers to understand and generate human language.

Computer vision is the field of artificial intelligence concerned with how computers can process and interpret digital images and videos.

Google released Gemini, its multimodal large language model, in December 2023.

Stable Diffusion, released as open source in 2022, is a latent diffusion model for text-to-image generation.

OpenAI Five defeated the OG team, world champions in Dota 2, in April 2019.

AlphaStar achieved Grandmaster level in StarCraft II in 2019 against human professional players.

Anthropic, founded in 2021, has published research on Constitutional AI for training helpful and harmless assistants.

CLIP, introduced by OpenAI in 2021, learns visual concepts from natural language supervision.

The Adam optimizer, introduced in 2014, is a widely used adaptive learning rate optimization algorithm.

Dropout is a regularization technique for neural networks that randomly omits units during training to reduce overfitting.

Meta's LLaMA family of large language models has been released as openly accessible weights for research.

DALL-E, introduced by OpenAI in 2021, generates images from textual descriptions using a transformer-based model.

Anthropic released Claude as a conversational AI system in 2023.

ChatGPT was released by OpenAI on November 30, 2022, gaining 100 million users within two months.

AlphaFold, developed by DeepMind, predicted the 3D structures of nearly all known proteins by 2022.

AlphaGo defeated world Go champion Lee Sedol 4-1 in March 2016.

GPT-3 was introduced by OpenAI in 2020 and contained 175 billion parameters.

Machine learning is a subfield of artificial intelligence focused on algorithms that improve through experience and data.

Reinforcement learning is a paradigm in which agents learn to make decisions by maximizing cumulative reward through interaction with an environment.

Convolutional neural networks are particularly effective for image processing tasks due to their use of local receptive fields and weight sharing.

Long Short-Term Memory networks, introduced by Hochreiter and Schmidhuber in 1997, address the vanishing gradient problem in recurrent networks.

BERT, introduced by Google in 2018, popularized bidirectional pretraining of transformers for natural language understanding.

The 2018 Turing Award was given to Yoshua Bengio, Geoffrey Hinton, and Yann LeCun for their work on deep learning.

Tokenization splits input text into smaller units that a language model can process, often using subword approaches like Byte-Pair Encoding.

Reinforcement learning from human feedback fine-tunes language models using human preference data to make outputs more helpful and aligned.

The transformer architecture was introduced in the 2017 paper Attention Is All You Need by Vaswani et al.

ImageNet is a large visual database that played a key role in the resurgence of deep learning, with the 2012 AlexNet result demonstrating breakthrough performance.

Supervised learning trains models on labeled data, while unsupervised learning finds structure in unlabeled data.

Neural networks are computational models inspired by biological neurons, composed of layers of weighted connections.

Backpropagation is the standard algorithm for training feedforward neural networks by computing gradients of the loss function.

Recurrent neural networks are designed to process sequences by maintaining hidden state across time steps.

Generative adversarial networks, introduced by Goodfellow et al. in 2014, train two networks in competition to generate realistic data.

Word embeddings represent words as dense vectors capturing semantic relationships, popularized by word2vec in 2013.

Large language models are typically trained with a next-token prediction objective on broad text corpora.

The 2024 Nobel Prize in Physics was awarded to John Hopfield and Geoffrey Hinton for foundational discoveries enabling machine learning with artificial neural networks.

The 2024 Nobel Prize in Chemistry was awarded in part to Demis Hassabis and John Jumper for AlphaFold's protein structure prediction work.

Diffusion models are a class of generative models that learn to reverse a noising process, used in modern image generation systems.

Self-attention is a mechanism that lets a sequence model weigh different parts of its input differently when computing each output.

Mixture of Experts models route inputs to different subnetworks, enabling larger total parameter counts with manageable per-token compute.

Retrieval-augmented generation combines a language model with an external information retrieval step to ground responses in source documents.