Results for "instruction design"
A high-priority instruction layer setting overarching behavior constraints for a chat model.
Fine-tuning on (prompt, response) pairs to align a model with instruction-following behaviors.
Task instruction without examples.
Designing systems where rational agents behave as desired.
Designing efficient marketplaces.
Automated detection/prevention of disallowed outputs (toxicity, self-harm, illegal instruction, etc.).
One example included to guide output.
Multiple examples included in prompt.
Controlling robots via language.
The text (and possibly other modalities) given to an LLM to condition its output behavior.
System design where humans validate or guide model outputs, especially for high-stakes decisions.
Tendency to trust automated suggestions even when incorrect; mitigated by UI design, training, and checks.
Mathematical framework for controlling dynamic systems.
Robots made of flexible materials.
Competition arises without explicit design.
System-level design for general intelligence.
Methods to set starting weights to preserve signal/gradient scales across layers.
The set of tokens a model can represent; impacts efficiency, multilinguality, and handling of rare strings.
Maximum number of tokens the model can attend to in one forward pass; constrains long-document reasoning.
Controlled experiment comparing variants by random assignment to estimate causal effects of changes.
Logging hyperparameters, code versions, data snapshots, and results to reproduce and compare experiments.
How many requests or tokens can be processed per unit time; affects scalability and cost.
A dataset + metric suite for comparing models; can be gamed or misaligned with real-world goals.
Coordinating tools, models, and steps (retrieval, calls, validation) to deliver reliable end-to-end behavior.
Forcing predictable formats for downstream systems; reduces parsing errors and supports validation/guardrails.
A theoretical framework analyzing what classes of functions can be learned, how efficiently, and with what guarantees.
Measures how much information an observable random variable carries about unknown parameters.
The shape of the loss function over parameter space.
Limiting gradient magnitude to prevent exploding gradients.
Neural networks can approximate any continuous function under certain conditions.