# Glossary

> 28 terms, in plain English.

- [AGI](https://www.superintelligence.com.au/glossary/agi.md): Artificial general intelligence: AI that can handle a wide range of tasks at roughly human level.
- [Alignment](https://www.superintelligence.com.au/glossary/alignment.md): The work of making AI systems pursue the goals their designers and users actually intend.
- [ASI](https://www.superintelligence.com.au/glossary/asi.md): Artificial superintelligence: AI that would greatly outperform the best humans at nearly all intellectual tasks.
- [Benchmark](https://www.superintelligence.com.au/glossary/benchmark.md): A standard set of test questions used to compare how well different AI models perform.
- [Benchmark saturation](https://www.superintelligence.com.au/glossary/benchmark-saturation.md): When top models score close to the maximum on a test, so it can no longer tell them apart.
- [Capability overhang](https://www.superintelligence.com.au/glossary/capability-overhang.md): When an AI system can do more than people have yet realised, or been able to draw out of it.
- [Compute](https://www.superintelligence.com.au/glossary/compute.md): The processing power used to train and run AI systems, usually measured in calculations performed.
- [Data contamination](https://www.superintelligence.com.au/glossary/data-contamination.md): When test questions leak into a model’s training data, making its scores look better than they should.
- [Emergent abilities](https://www.superintelligence.com.au/glossary/emergent-abilities.md): Skills that seem to appear suddenly in larger AI models and were not present in smaller ones.
- [Evaluation](https://www.superintelligence.com.au/glossary/evaluation.md): The process of testing what an AI system can do and how safely it behaves, often shortened to an eval.
- [Fine-tuning](https://www.superintelligence.com.au/glossary/fine-tuning.md): Further training of an existing model on a smaller, focused set of examples to change how it behaves.
- [FLOP](https://www.superintelligence.com.au/glossary/flop.md): A single basic arithmetic step by a computer, used to count how much calculation a training run needs.
- [Frontier lab](https://www.superintelligence.com.au/glossary/frontier-lab.md): An organisation building some of the most capable AI models in the world, at the leading edge.
- [Frontier model](https://www.superintelligence.com.au/glossary/frontier-model.md): One of the most capable AI models available at a given time, at the leading edge of what is possible.
- [Human baseline](https://www.superintelligence.com.au/glossary/human-baseline.md): How people score on the same test as an AI model, provided for comparison.
- [Instrumental convergence](https://www.superintelligence.com.au/glossary/instrumental-convergence.md): The idea that AI systems with very different goals may pursue similar sub-goals, like gaining resources or avoiding shutdown.
- [Intelligence explosion](https://www.superintelligence.com.au/glossary/intelligence-explosion.md): A hypothesised runaway process in which AI rapidly improves itself, leaving human ability far behind.
- [Interpretability](https://www.superintelligence.com.au/glossary/interpretability.md): Research aimed at understanding what is happening inside an AI model, and why it gives its answers.
- [Parameters](https://www.superintelligence.com.au/glossary/parameters.md): The adjustable numbers inside an AI model that are tuned during training and store what it has learned.
- [Pre-training](https://www.superintelligence.com.au/glossary/pre-training.md): The first, very large stage of building a model, where it learns general patterns from huge amounts of data.
- [Reasoning model](https://www.superintelligence.com.au/glossary/reasoning-model.md): An AI model trained to work through a problem step by step before it gives a final answer.
- [Recursive self-improvement](https://www.superintelligence.com.au/glossary/recursive-self-improvement.md): A process in which an AI system improves its own design, so each version can build a better one.
- [RLHF](https://www.superintelligence.com.au/glossary/rlhf.md): Reinforcement learning from human feedback: training a model using people’s ratings of its answers.
- [Scaling laws](https://www.superintelligence.com.au/glossary/scaling-laws.md): Patterns showing that AI models improve predictably as they get more compute, data and parameters.
- [Superintelligence](https://www.superintelligence.com.au/glossary/superintelligence.md): Intelligence that greatly exceeds the best human minds in nearly every area.
- [Takeoff](https://www.superintelligence.com.au/glossary/takeoff.md): How quickly AI could move from roughly human-level ability to far beyond it once that point is reached.
- [Test-time compute](https://www.superintelligence.com.au/glossary/test-time-compute.md): The computing effort a model uses while answering a question, as opposed to while it is being trained.
- [Transformer](https://www.superintelligence.com.au/glossary/transformer.md): The design behind most modern AI language models, built to weigh how each word relates to the others.