Decoding AI: A Glossary from AGI to Hallucinations

Artificial intelligence is not only reshaping the world but also creating a new vocabulary to describe its impact. In meetings, pitches, and panels, terms like LLMs, RAG, and RLHF are tossed around, often confusing even tech experts. To address this, a regularly updated AI glossary has been compiled to define key concepts clearly.

Core Concepts

AGI (Artificial General Intelligence) is a vague term generally referring to AI that outperforms humans on most tasks. OpenAI CEO Sam Altman describes it as the “equivalent of a median human that you could hire as a co-worker,” while Google DeepMind views AGI as AI at least as capable as humans in most cognitive tasks.

AI agents are tools that use AI to perform tasks on your behalf, such as filing expenses, booking tickets, or writing code. The concept can vary, and infrastructure is still evolving.

Hallucination is the industry term for AI generating false information. This poses serious risks, from misleading advice to dangerous medical suggestions, often caused by gaps in training data.

Large Language Models (LLMs) power assistants like ChatGPT, Claude, and Gemini. These deep neural networks with billions of parameters learn relationships between words from massive text datasets, generating the most likely response to a prompt.

Other Key Terms

Fine-tuning involves additional training of a model for a specific task using specialized data. Many startups start with an LLM and fine-tune it for their domain.

Inference is the process of running a model to make predictions or decisions based on learned data. Training must occur first.

Memory caching is an optimization technique that stores previous computations to speed up inference, reducing power usage and response time.

This glossary is a living document, updated regularly as the AI field evolves, helping everyone stay informed.