HOW LLMs WORK FOR THE NON-SPECIALIST

ARTIFICIAL INTELLIGENCE ASSISTANT PROMPTS [START OF PROMPT] Estimate the token footprint of the following prompt and expected completion before I send it to a paid model Application Programming Interface (API), so I can budget cost for this harness: [PASTE PROMPT TEXT / TEMPLATE HERE]. Break the estimate into system-prompt tokens, injected-context tokens, and expected-output tokens, flag any single component that is disproportionately large, and suggest at least two concrete edits to shrink the total token count without losing meaning. Output format: a short table (component, estimated tokens, notes) followed by a bulleted list of cut suggestions. [END OF PROMPT] [START OF PROMPT] Audit the following agent output log for hallucination risk before we ship this feature: [PASTE AGENT OUTPUTS / TRANSCRIPT HERE]. For each claim the agent makes, mark it as (a) grounded in a cited source actually present in context, (b) plausible but unverifiable from the given context, or (c) likely hallucinated. For every item marked (b) or (c), recommend a specific harness fix (add a retrieval step, lower temperature, add a verification tool call (a request that makes the harness run a checking tool), or require a citation) that would have prevented it. Output format: a numbered list of claims with their classification and recommended fix. [END OF PROMPT]