SI Glossary
SI Glossary: Super Intelligence terms explained
Since Executive Order 14434, federal documents say “SI” where they used to say “AI.” Here are 31 terms in plain English, including how the old words map to the new ones.
Jump to
Terms
- Agent
- An SI system that takes actions toward a goal, such as browsing, running code, or calling tools, rather than only answering a question. Under EO 14434 federal usage, an “AI agent” becomes an “SI agent.”
- AI (Artificial Intelligence)
- The long-standing term for computer systems that perform tasks associated with human intelligence. It remains the term in existing law (15 U.S.C. § 9401(3)), and you can keep using it. See SI.
- AI → SI mapping
- The rule of thumb for reading federal documents since the order: swap “AI” for “SI” and “Artificial Intelligence” for “Super Intelligence.” “AI model” reads “SI model,” “AI policy” reads “SI policy.” The underlying technology is the same.
- Alignment
- Work to make an SI system pursue the goals and values its builders and users intend, and avoid harmful behavior it was not asked for.
- Benchmark
- A standardized test set used to compare models on a skill like coding, math, or reading comprehension. High scores show progress on that test, not general ability.
- Context window
- How much text (measured in tokens) a model can consider at once, including your prompt, documents, and its own reply.
- Executive Order 14434
- “Inaugurating the Era of Super Intelligence,” signed Sept 29, 2026 and published in the Federal Register Oct 2, 2026 (91 FR 63129). It directs executive-branch non-statutory communications to use “Super Intelligence” and “SI.” Read our explainer.
- Fine-tuning
- Further training a pre-trained model on a smaller, targeted dataset so it gets better at a specific task, style, or domain.
- Foundation model
- A large model trained on broad data that can be adapted to many tasks. Most chatbots and agents are built on one.
- Frontier model
- One of the most capable models available at a given time, typically from the largest labs. Frontier models draw the most safety and policy attention.
- Guardrails
- Rules, filters, and checks placed around a model to block unsafe or off-limits outputs and actions.
- Hallucination
- When a model states something false or made up with apparent confidence.
- Inference
- Running a trained model to get an output, such as answering a prompt. Training builds the model; inference uses it.
- Large language model (LLM)
- A model trained on huge amounts of text to predict and generate language. ChatGPT-style assistants run on LLMs.
- Model
- The trained system itself: a large file of learned numbers (weights) that turns inputs into outputs. In federal usage, “AI model” is now “SI model.”
- Multimodal
- Able to handle more than one kind of input or output, such as text, images, audio, and video.
- Open-weight model
- A model whose trained weights are published so anyone can download, run, and modify it.
- Parameters
- The adjustable numbers inside a model that are set during training. Parameter count is a rough proxy for model size.
- Pre-training
- The first, most expensive training stage, where a model learns general patterns from a very large dataset.
- Prompt
- The instructions or question you give a model.
- Red teaming
- Deliberately trying to make a model fail or misbehave, to find weaknesses before others do.
- RLHF
- Reinforcement learning from human feedback: people rate model outputs and the model is trained to produce the kinds of answers rated higher.
- SI (Super Intelligence)
- The federal government’s term, under EO 14434, for what was called AI. Section 3(a) defines it as the same technologies covered by the statutory AI definition, so it is a rename, not a new category of technology. See the explainer.
- SI adoption
- How widely agencies, companies, and media have actually switched from “AI” to “SI.” The order binds only executive-branch non-statutory documents. Our tracker follows who switched and who didn’t.
- SI definition proposal
- Under EO 14434 § 3(b), the Assistant to the President for Science and Technology has about 60 days (around Nov 28, 2026) to propose legislative language for a federal definition of Super Intelligence and SI.
- si.gov
- The federal government’s SI website, launched by the White House on Oct 13, 2026. ai.gov redirects there.
- Superintelligence (research sense)
- In research, a long-standing term for a hypothetical system that outperforms humans across most domains. That is different from the government’s SI, which covers today’s systems, and the difference is a source of criticism.
- Synthetic data
- Training data generated by models instead of collected from people or the world.
- Token
- The small chunk of text (often part of a word) that models read and write. Usage, pricing, and context windows are measured in tokens.
- Training
- The process of adjusting a model’s parameters using data until it performs well. See pre-training and fine-tuning.
- Weights
- Another word for a model’s learned parameters. “Releasing the weights” means publishing the model itself.