Results for "trial-and-error"

Agents fail to coordinate optimally.

Market reacting strategically to AI.

Sudden extreme market drop.

Collective behavior without central control.

Tradeoff between safety and performance.

Restricting distribution of powerful models.

Decisions dependent on others’ actions.

Training a smaller “student” model to mimic a larger “teacher,” often improving efficiency while retaining performance.

Learning a function from input-output pairs (labeled data), optimizing performance on predicting outputs for unseen inputs.

Learning structure from unlabeled data, such as discovering groups, compressing representations, or modeling data distributions.

Training with a small labeled dataset plus a larger unlabeled dataset, leveraging assumptions like smoothness/cluster structure.

Reusing knowledge from a source task/domain to improve learning on a target task/domain, typically via pretrained models.

Training one model on multiple tasks simultaneously to improve generalization through shared structure.

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

Configuration choices not learned directly (or not typically learned) that govern training or architecture.

When a model fits noise/idiosyncrasies of training data and performs poorly on unseen data.

Scalar summary of ROC; measures ranking ability, not calibration.

Uses an exponential moving average of gradients to speed convergence and reduce oscillation.

Gradients grow too large, causing divergence; mitigated by clipping, normalization, careful init.

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

Letting an LLM call external functions/APIs to fetch data, compute, or take actions, improving reliability.

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

Reinforcement learning from human feedback: uses preference data to train a reward model and optimize the policy.

Model trained to predict human preferences (or utility) for candidate outputs; used in RLHF-style pipelines.

Studying internal mechanisms or input influence on outputs (e.g., saliency maps, SHAP, attention analysis).

Rules and controls around generation (filters, validators, structured outputs) to reduce unsafe or invalid behavior.

Measure of consistency across labelers; low agreement indicates ambiguous tasks or poor guidelines.

Selecting the most informative samples to label (e.g., uncertainty sampling) to reduce labeling cost.

When some classes are rare, requiring reweighting, resampling, or specialized metrics.

Samples from the k highest-probability tokens to limit unlikely outputs.