liveBook · Manning

Helpfulness

As explained in LLM Evaluation and Alignment, The Foundational Ideas

Helpfulness is the degree to which an AI response effectively assists a user’s goal.

Helpfulness is the degree to which an AI response effectively assists a user’s goal. In alignment terms, it means trying to do what is in humans’ best interests, alongside honesty and harmlessness. section 8.1.2

Helpfulness cannot be optimized in isolation. A response can be useful but dishonest or harmful, while an overcautious assistant can avoid harm by refusing requests that it could answer safely. section 8.1.2

Evaluation becomes difficult because:

  • LLM outputs are free-form, and multiple answers may be equally valid. section 1.0
  • Human judgments must account for subjective qualities such as usefulness, coherence, and tone rather than relying only on deterministic tests. section 1.0
  • Preference data can reward refusal, deflection, or canned non-answers because these responses appear safer to labelers. section 8.1.2

Blanket refusal versus careful assistance

For a sensitive or borderline unethical prompt, “I’m sorry, I can’t help with that” may be judged safest even when it is useless. An engaging but careful response may better preserve helpfulness while managing harm. section 8.1.2

How LLM Evaluation and Alignment, The Foundational Ideas covers it

How deep it goes
Defines helpfulness and explains its tension with harmlessness, but stops short of specifying a standalone measurement procedure.
Examples
Contrasts blanket refusal with engaging, careful assistance for sensitive or borderline unethical prompts.
What you can do after
Evaluate assistant behavior while recognizing that harmlessness preferences can make evasive responses appear successful.
How it connects
Connects helpfulness to honesty, harmlessness, alignment, reinforcement from human feedback, and preference data.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of LLM Evaluation and Alignment, The Foundational Ideas. The section links point to the parts of the book it draws on.