Results for "causal generation"

Directed acyclic graph encoding causal relationships.

Framework for reasoning about cause-effect relationships beyond correlation, often using structural assumptions and experiments.

Models effects of interventions (do(X=x)).

Formal model linking causal mechanisms and variables.

A hidden variable influences both cause and effect, biasing naive estimates of causal impact.

What would have happened under different conditions.

Variable enabling causal inference despite confounding.

Prevents attention to future tokens during training/inference.

Expected causal effect of a treatment.

Probability of treatment assignment given covariates.

Training objective where the model predicts the next token given previous tokens (causal modeling).

AI proposing scientific hypotheses.

Controlled experiment comparing variants by random assignment to estimate causal effects of changes.

Model relies on irrelevant signals.

Trend reversal when data is aggregated improperly.

Generates sequences one token at a time, conditioning on past tokens.

Architecture that retrieves relevant documents (e.g., from a vector DB) and conditions generation on them to reduce hallucinations.

Search algorithm for generation that keeps top-k partial sequences; can improve likelihood but reduce diversity.

A high-capacity language model trained on massive corpora, exhibiting broad generalization and emergent behaviors.

Stochastic generation strategies that trade determinism for diversity; key knobs include temperature and nucleus sampling.

Models that learn to generate samples resembling training data.

Generating human-like speech from text.

Ensuring decisions can be explained and traced.

The internal space where learned representations live; operations here often correlate with semantics or generative factors.

Attention where queries/keys/values come from the same sequence, enabling token-to-token interactions.

Predicts masked tokens in a sequence, enabling bidirectional context; often used for embeddings rather than generation.

A model that assigns probabilities to sequences of tokens; often trained by next-token prediction.

Crafting prompts to elicit desired behavior, often using role, structure, constraints, and examples.

Breaking documents into pieces for retrieval; chunk size/overlap strongly affect RAG quality.

Constraining outputs to retrieved or provided sources, often with citation, to improve factual reliability.