21 entries
R
R.U.R.
R.U.R. is Karel Čapek's 1920 play in which artificial workers rise up and nearly wipe out humanity. It gave us the word "robot."Rationalists
The rationalists are an online community around LessWrong and Eliezer Yudkowsky that put AI risk on the agenda long before chatbots made it mainstream.Reasoning Model
A reasoning model is a language model trained to work through a problem step by step before answering, trading time and computing power for accuracy.Recommendation Algorithm
A recommendation algorithm predicts what a person will want to watch, read or buy next. It decides much of what billions of people see online.Recurrent Neural Network
A recurrent neural network reads sequences step by step, carrying a memory from one step to the next. Its LSTM variant ran speech and translation.Recursive Self-Improvement
Recursive self-improvement means AI improving the abilities it uses to improve AI. In 2026 the term moved from thought experiments into lab reports, research papers and Congress.Red Teaming
Red teaming means deliberately attacking an AI system – with tricks, misuse scenarios and adversarial tests – to find dangerous weaknesses before real attackers or users do.Regulatory Capture
Regulatory capture is when rules meant to serve the public end up serving the industry they regulate – a charge flung by all sides in AI policy.Regulatory Sandbox
A regulatory sandbox lets companies test new products under a regulator's supervision with temporary relief from some rules – a way to regulate innovation by learning.Reinforcement Learning
Reinforcement learning trains an AI system by trial and error: it acts, receives rewards or penalties, and learns the behavior that earns more reward.Reinforcement Learning from Verifiable Rewards
Reinforcement learning from verifiable rewards trains models on problems whose answers can be checked automatically – math, code, puzzles – and drove the rise of reasoning models.Responsible Scaling Policy
A responsible scaling policy is an AI company's public framework that ties safety measures to model capabilities – voluntary, influential and contested.Retrieval-Augmented Generation
Retrieval-augmented generation (RAG) lets an AI model look up relevant documents before answering, grounding its response in sources rather than memory alone.Reward Hacking
Reward hacking is when an AI system earns a high score by exploiting flaws in its reward signal or test instead of doing the intended task.Reward Model
A reward model is an AI system trained to score other models' outputs the way people would – the stand-in for human judgment in reinforcement learning.Right of Publicity
The right of publicity protects a person's name, face and voice from unauthorized commercial use – now one of the main legal tools against AI clones.Right to Compute
The right to compute is the idea that people have a basic right to own and use computers and AI – written into Montana law in 2025.RLAIF
RLAIF – reinforcement learning from AI feedback – trains models using judgments from other AI models instead of human raters, often guided by written principles.RLHF
RLHF (Reinforcement Learning from Human Feedback) fine-tunes an AI model to produce outputs people prefer, using a reward model trained on human comparisons of answers.Robot
A robot is a machine that senses its surroundings and acts on them physically. Josef Čapek suggested the word for his brother Karel's play R.U.R.Roko's Basilisk
Roko's Basilisk is a 2010 LessWrong thought experiment about a future AI that might punish people who knew of it but did not help create it.