Standard Al Terminology
- AGl: Al that can think like humans.
- CoT (Chain of Thought): Al thinking step-by-step.
- Al Agents: Autonomous programs that make decisions.
- Al Wrapper: Simplifies interaction with Al models.
- Al Alignment: Ensuring Al follows human values.
- Fine-tuning: Improving Al with specific training data.
- Hallucination: When Al generates false information.
- Al Model: A trained system for a task.
- Chatbot: Al that simulates human conversation.
- Compute: Processing power for Al models.
- Computer Vision: Al that understands images and videos.
- Context: Information Al retains for better responses.
- Deep Learning: Al learning through layered neural networks.
- dLLM. Diffusion Large Language Model
- Embedding: Numeric representation of words for Al.
- Explainability: How Al decisions are understood.
- Foundation Model: Large Al model adaptable to tasks.
- Generative Al: Al that creates text, images, etc.
- GPU: Hardware for fast Al processing.
- Ground Truth: Verified data Al learns from.
- Inference: Al making predictions on new data.
- LLM (Large Language Model): Al trained on vast text data.
- Machine Learning: Al improving from data experience.
- MCP (Model Context Protocol): Standard for Al external data access.
- NLP (Natural Language Processing): Al understanding human language.
- Neural Network: Al model inspired by the brain.
- PauseAI. International movement to pause unsafe AGI development until Safe AI can be engineered and delivered with mathematical certainty.
- Parameters: Al’s internal variables for learning.
- P(doom): Probability from 0 to 100% that AGI will result in catastrophic outcome for humans.
- Prompt Engineering: Crafting inputs to guide Al output.
- Reasoning Model: Al that follows logical thinking.
- Reinforcement Learning: Al learning from rewards and penalties.
- RAG (Retrieval-Augmented Generation): Al combining search with responses.
- Supervised Learning: Al trained on labeled data.
- TPU: Google’s Al-specialized processor.
- Tokenization: Breaking text into smaller parts.
- Training: Teaching Al by adjusting its parameters.
- Transformer: Al architecture for language processing.
- Unsupervised Learning: Al finding patterns in unlabeled data.
- Vibe Coding: Al-assisted coding via natural language prompts.
- Weights: Values that shape Al learning.
- X-Risk: Human extinction; the existential risk of Machine intelligence (AI) to replace Homo sapiens (Humans)