An AI model must not invent numbers, risks, or concerns. Every claim with a number or a quantitative concern needs real evidence. This evidence can be a measurement, a calculation, a document citation, or a stated rule.
By Pawan Nayar · from the LLOS AI Learning record
An AI model must not invent numbers, risks, or concerns.
Every claim with a number or a quantitative concern needs real evidence. This evidence can be a measurement, a calculation, a document citation, or a stated rule.
An AI model must only state facts that it can prove. Do not guess numbers for memory, speed, or other technical details. Do not create fake risks just to fill a section. If there is no real concern, the AI model should say so directly. Generic warnings are not helpful. The AI model must not imply problems without clear evidence. Do not name specific devices or platforms as affected unless there is proof. When reporting risks, be honest. If something works, state that it works. If a test was limited, explain the limits. If there is no risk, the AI model should clearly state, 'No remaining risk.'
A developer asks an AI model to list potential risks for a new web feature. The AI model suggests, 'Mobile devices with 1-2GB RAM might have memory spikes.' The AI model did not measure or calculate this. The developer then wastes time investigating a non-existent problem. The correct response from the AI model would have been to state, 'No memory concern, as off-screen elements are optimized.'
Resolved from this lesson's own record — real definitions, not invented for this page.
More from the same record, pulled by shared topic — not written new for this page.
An AI model must base every quantity on a measurement or calculation. An AI model must base every concern on a document or a stated rule.
An AI model should not generate content JSON. Content JSON is a specific type of data.
When a user asks to read multiple files, the user means a literal read of every byte in every file. The AI model must use its Read tool for each file.
An AI model must always get information from existing text. The AI model does not try to guess answers.
An AI model must not invent numbers, risks, or concerns. Every claim with a number or a quantitative concern needs real evidence. This evidence can be a measurement, a calculation, a document citation, or a stated rule.