2010s
487 words first used in this period, under the year each was born.
2010
2011
2012
- AlexNet
- Brussels effect
- ByteDance
- deep learning revolution
- dropout
- gradient clipping
- instrumental convergence
- knowledge graph
- model parallelism
- NPU
- orthogonality thesis
- random search
- RMSProp
- robotic process automation
- Wikidata
- Arcade Learning Environment
- disentangled representation
- MuJoCo
- oracle AI
- semantic textual similarity
- tool AI
- WordPiece
2013
- adversarial example
- lethal autonomous weapons system
- Meta AI
- algorithmic progress
- deep Q-network
- deep reinforcement learning
- high-bandwidth memory
- ICLR
- Lean
- predictive processing
- R-CNN
- saliency map
- techno-solutionism
- two-tower model
- variational autoencoder
- word2vec
- Caffe
- Her (film)
- high-level machine intelligence
- individual conditional expectation
- knowledge graph embedding
- leaky ReLU
- normalizing flow
- occlusion sensitivity
- OpenReview
- quantum machine learning
- reparameterization trick
- straight-through estimator
2014
- AI alignment
- alignment problem
- attention
- knowledge distillation
- levels of driving automation
- Adam
- adversarial training
- AI audit
- algorithmic accountability
- Allen Institute for AI
- AutoML
- COCO
- control problem
- cross-attention
- Cyberspace Administration of China
- encoder–decoder architecture
- Future of Life Institute
- gated recurrent unit
- generative adversarial network
- instance segmentation
- Jupyter notebook
- Polanyi's paradox
- sequence-to-sequence model
- surveillance capitalism
- treacherous turn
- value alignment
- wake word
- XGBoost
- AI Impacts
- attention pattern
- attention-based explanation
- capability control
- decisive strategic advantage
- depthwise separable convolution
- dilated convolution
- fast gradient sign method
- fully convolutional network
- GloVe
- guided backpropagation
- implicit regularization
- Inception
- machine learning technical debt
- model hub
- model inversion
- Moloch
- motivation selection
- multipolar scenario
- neural Turing machine
- NVLink
- Open Philanthropy
- perverse instantiation
- VGGNet
2015
- diffusion model
- OpenAI
- text-to-image model
- algorithmic management
- batch normalization
- byte-pair encoding
- corrigibility
- Keras
- learning rate warmup
- machine unlearning
- normalization layer
- residual connection
- ResNet
- style transfer
- TensorFlow
- U-Net
- visual question answering
- YOLO
- agent foundations
- algorithm aversion
- anchor box
- BookCorpus
- Cerebras
- deep deterministic policy gradient
- DeepDream
- expected calibration error
- exposure bias
- generalized advantage estimation
- highway network
- label smoothing
- layer-wise relevance propagation
- LibriSpeech
- noise schedule
- Open Letter on Artificial Intelligence
- quality diversity
- shutdown problem
- target network
- trust region policy optimization
2016
- AlphaGo
- Hugging Face
- reward hacking
- cosine annealing
- exact match
- fairness impossibility theorem
- federated learning
- GELU
- General Data Protection Regulation
- Grad-CAM
- gradient checkpointing
- HNSW
- image-to-image translation
- kill switch
- layer normalization
- LightGBM
- LIME
- linear probe
- mode collapse
- model extraction
- model serving
- Move 37
- post hoc explanation
- probing
- right to explanation
- Robodebt
- scalable oversight
- Scale AI
- self-attention
- SiLU
- SQuAD
- Tay
- TPU
- transformative AI
- WaveNet
- A3C
- accumulated local effects
- Algorithmic Justice League
- autoregressive image generation
- Carlini–Wagner attack
- Center for Human-Compatible AI
- Centre for the Governance of AI
- cooperative inverse reinforcement learning
- DeepLIFT
- DenseNet
- Distill
- DP-SGD
- early exit
- elastic weight consolidation
- equal opportunity
- equalized odds
- equivariance
- fairness through unawareness
- gated linear unit
- geometric deep learning
- gradient masking
- Gumbel-softmax
- HCH
- Hyperband
- hypernetwork
- illusionism
- impact measure
- Inception score
- KL penalty
- logical induction
- mild optimization
- model-agnostic
- moral crumple zone
- MS MARCO
- NCCL
- negative side effects
- neural architecture search
- off-switch game
- Partnership on AI
- PATE
- perceptual loss
- predictive parity
- prosaic alignment
- quantilizer
- ROCm
- safe interruptibility
- suffering risk
- TensorRT
- training–serving skew
- universal adversarial perturbation
2017
- AI governance
- deepfake
- reinforcement learning from human feedback
- text-to-video model
- transformer
- AdamW
- adapter
- adversarial patch
- AI for good
- AI supply chain security
- AlphaGo Zero
- AlphaZero
- attention head
- Attention Is All You Need
- attention mask
- backdoor attack
- bfloat16
- browser agent
- CatBoost
- causal mask
- confidential computing
- constrained decoding
- counterfactual explanation
- FAISS
- feed-forward network
- focal loss
- Fréchet inception distance
- integrated gradients
- KV cache
- longtermism
- membership inference attack
- mixed-precision training
- multi-head attention
- positional encoding
- preference data
- proximal policy optimization
- PyTorch
- query, key and value
- scaled dot-product attention
- SHAP
- Software 2.0
- tensor core
- transformer block
- AI Now Institute
- Asilomar AI Principles
- capsule network
- causal abstraction
- CERN for AI
- Clearview AI
- Core ML
- CoreWeave
- CoRL
- counterfactual fairness
- critical batch size
- data versioning
- domain randomization
- expert iteration
- expert parallelism
- feature store
- inverse reward design
- ISO/IEC JTC 1/SC 42
- Libratus
- LiteRT
- Mathlib
- MobileNet
- model fingerprinting
- model-agnostic meta-learning
- network dissection
- neuralese
- ONNX
- physics-informed neural network
- projected gradient descent attack
- quantization-aware training
- Ray
- secure aggregation
- SmoothGrad
- squeeze-and-excitation network
- steganography
- testing with concept activation vectors
- Universal Paperclips
- Vector Institute
- VQ-VAE
- Wasserstein GAN
- Weights & Biases
- X is all you need
- XLA
2018
- GPT
- misuse
- prompt
- red teaming
- voice cloning
- algorithmic impact assessment
- BERT
- decoder-only transformer
- deepfake detection
- encoder-only transformer
- ethics washing
- Gender Shades
- GLUE
- liar's dividend
- masked language model
- MLOps
- model card
- post-training quantization
- reward model
- special token
- specification gaming
- StyleGAN
- task exposure
- top-k sampling
- training compute
- UMAP
- Adafactor
- AI debate
- Anchors
- canary string
- certified robustness
- clean-label poisoning
- datasheet for datasets
- deadly triad
- double descent
- ELMo
- embedded agency
- FAccT
- factored cognition
- goal-directedness
- Huang's law
- intent alignment
- iterated amplification
- JAX
- liquid neural network
- local SGD
- lottery ticket hypothesis
- MLflow
- MLPerf
- mode connectivity
- neural ordinary differential equation
- neural tangent kernel
- OpenAI Five
- panoptic segmentation
- Papers with Code
- petaflop/s-day
- pipeline parallelism
- prototypical part network
- recursive reward modeling
- relative reachability
- Runway
- SentencePiece
- Shampoo
- soft actor–critic
- tensor parallelism
- Transformers
- vulnerable world hypothesis
2019
- ARC-AGI
- environmental impact of AI
- AI detector
- bi-encoder
- causal language model
- Cohere
- continued pretraining
- cross-encoder
- deceptive alignment
- ghost work
- godfathers of AI
- GPT-2
- human oversight
- inner alignment
- OECD AI Principles
- outer alignment
- parameter-efficient fine-tuning
- power-seeking
- repetition penalty
- Sentence-BERT
- speech-to-speech model
- Stability AI
- staged deployment
- Stanford HAI
- steerability
- the bitter lesson
- top-p sampling
- trustworthy AI
- vector database
- vision-language model
- algorithmic recourse
- alignment tax
- AlphaStar
- attainable utility preservation
- benign overfitting
- BERTScore
- C4
- Center for Security and Emerging Technology
- centered kernel alignment
- comprehensive AI services
- EfficientNet
- Ethics Guidelines for Trustworthy AI
- explainable boosting machine
- gradient hacking
- gradient inversion attack
- Gradio
- Green AI
- HellaSwag
- IMO Grand Challenge
- implicit neural representation
- latent adversarial training
- Megatron-LM
- mesa-optimization
- multi-query attention
- MuZero
- myopia
- neural operator
- overthinking
- plug and play language model
- quality filtering
- randomized smoothing
- REINFORCE leave-one-out
- reward tampering
- RMSNorm
- score-based generative model
- sparse attention
- SuperGLUE
- T5
- test-time training
- TinyML
- Transformer-XL
- Triton
- wafer-scale engine
- wav2vec
- WebText
- ZeRO
- Zhipu AI