Sign inSign in
ConceptAI & MLTheory & Algos4 more

Hallucinations

One word, taught differently in 6 fields. Each use is given apart.

In AI & ML

Instances where AI models generate factually incorrect or fabricated information.
as Arjun Kumar Manrai defines itUC BerkeleyThe wording closest to all 6 definitions

In Theory & Algos

Instances where AI systems very confidently tell you about something that isn't true.

In Bio & Neuro

Defining features of schizophrenia, predominantly auditory in nature, where patients perceive voices and sounds that are structured rather than random.

In Systems & CS

Instances where tools like ChatGPT make up false statements with high confidence because they are probabilistically plausible.

In Medicine

Generated ideas or content that are not real.

In Psychology

A sensory experience that is not real, such as hearing voices.

Know first

Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.

Opens up

Concepts that name this one as a prerequisite.

In AI & ML

6

Each in their own terms, from the easiest level.

  • Andrej Karpathy

    Undergraduate

    When the model produces plausible-sounding outputs that are factually incorrect or fabricated based on probabilistic guesses.
    • 55:50Deep Dive into Large Language Models: Pretraining to RLHF
  • Jeremy Howard

    Undergraduate

    When a language model generates an opinionated or made-up sentence to complete a prompt with high confidence.
    • 25:50A Hacker's Guide to Language Models
  • Steven Feng

    Stanford University · Undergraduate

    A world-modeling error where a language model produces outputs that contradict the reference or ground truth world model.
    • 55:26Overview of Transformers (Stanford CS25: Transformers United V6)
  • Chelsea Finn

    Stanford University · Graduate

    Mistakes made by models even in non-safety critical scenarios where incorrect information is presented with high confidence.
    • 24:14Frontiers of Deep Reinforcement Learning
  • Douwe Kiela

    Stanford University · Graduate

    When a retrieved or generated statement does not correspond to the established ground truth.
    • 1:16:06Retrieval Augmented Language Models

In Theory & Algos

1

    In Bio & Neuro

    1

      In Systems & CS

      1

        In Medicine

        1

          In Psychology

          1

            Lectures that assume it

            3

            Connections

            24

            The lectures that define it or assume it, and the concepts next to it.

            11 lecturers. Definitions are written by AI from each lecture’s transcript. A time with a play mark starts the lecture at the moment the lecturer says it.