AI safety org
2021-12-03 23:04 applied to them in ~June 2021, they said I could apply in a year.
AI
Reinforcement learning
Deep deterministic policy gradient
Twin Delayed DDPG
Offline reinforcement learning
Rainbow
Implicit quantile networks
Double Q learning
Intrinsic motivation
AIXI-like RL algorithm
Offworld Gym
Boltzmann exploration
Soft Q learning
Reward learning
Soft Actor-Critic
Proximal policy optimization
Proximal Policy Optimization Algorithms
1707.06347.pdf
Generalized advantage estimation
Trust Region Policy Optimization (algorithm)
Exploration
Multi-agent reinforcement learning
Asynchronous advantage actor critic
Reinforcement learning from human feedback
Model predictive control
Doubly robust estimator
Saliency maps
Monte Carlo tree search
MuJoCo
Maybe try dm_control
dm_control
NEAT algorithm
Center for Human-Compatible Artificial Intelligence
DeepMind
Anthropic
Imitation learning
Soft Q imitation learning
SQIL: Imitation Learning via Reinforcement Learning with Sparse Rewards
sqil_imitation_learning_via_re-Original Pdf.pdf
Generative adversarial imitation learning
1606.03476.pdf
Reinforcement learning and control as probabilistic inference: Tutorial and review
1805.00909.pdf
Why does L1 penalty encourage sparsity?
Deep reinforcement learning that matters
Discriminator-actor-critic: Addressing sample inefficiency and reward bias in adversarial imitation learning
Sample efficient imitation learning for continuous control
Random expert distillation: Imitation learning via expert policy support estimation
Behavior cloning
MineRL Basalt competition
Inverse soft Q-learning
DAgger algorithm
Stein variational gradient descent
Overcooked-AI
Sequence modeling
Transformers
Language modeling
Language model fine-tuning
Nucleus sampling
Redwood Research
Iterated distillation and amplification
AI safety
Attainable Utility Preservation
Assistance game framework
AI Safety Camp
AI safety resources
Learning human preferences: black-box, white-box, and structured white-box access
Model splintering: moving from one imperfect model to another
DeepMind Safety Medium
Selection Theorems reading list
Late 2021 MIRI Conversations
Avoiding side-effects
Impact regularization
Relative reachability
Reward modeling
Myopic reinforcement learning
Debate
Eliciting latent knowledge
Critique
Causal influence diagrams
Inverse reinforcement learning
Value learning
Function optimization
Cross-entropy method
Interpretability
Architecture
Variational autoencoders
Graph neural networks
Recursive models
Activation functions
Convolutions
Attention
Diffusion
GLUE benchmark
AI governance
Transfer learning
Calibration
Adversarial approaches
Benchmarks
Fairness
Imitative generalization
Unrestricted Adversarial Examples Challenge
Anomaly detection
Algorithm learning
Distillation
Generative models
Image synthesis
Expert iteration
Multi-step reasoning
Low-rank structure
Bootstrap aggregating
Active learning
Dataset
Retrieval
Scaling
Robotics
Reward tampering
Sparsity
Self-improvement
Code generation
Out-of-distribution robustness
Optimized inference/training
Quantization
Multi-objective reinforcement learning
Sample efficiency
Reward hacking
Goodhart's law