To: Johnny Canuck who wrote (70627 ) 2/9/2026 12:07:54 PM From: Johnny Canuck Read Replies (1) | Respond to Conversation with Geminim.youtube.com The Problem: Helpful vs. Accurate AI models are trained to be helpful assistants. If they cannot find information in a document, they often guess confidently rather than admitting they don't know, filling in gaps with their existing training data [ 01:17 ].Step 1: Choose the Right Model To minimize errors, always use high-level reasoning models rather than default versions [ 01:47 ]. Recommended models include:ChatGPT: GPT-o1 (or the latest reasoning model) [ 02:13 ].Claude: Opus 3.5/4.5 with extended reasoning [ 02:20 ].Gemini: Gemini 1.5/2.0 Pro [ 02:25 ].Step 2: The 3-Rule Grounding Prompt Adding these three specific rules to your prompt forces the AI to stay anchored to your file [ 03:02 ]:Strict Grounding: "Base your answer only on the uploaded documents and nothing else" [ 03:13 ].Permission to Fail: "If information isn't found, say 'not found.' Do not guess" [ 03:30 ].Mandatory Citations: Require the AI to provide the document name, page/section, and a direct quote for every claim [ 04:07 ].Bonus Tips: The Gray Area: Ask the AI to mark information as "unverified" if it is only "pretty sure," helping you prioritize what to double-check manually [ 05:20 ].High Stakes: For legal or financial work, tell the AI to only respond if it is 100% confident [ 06:07 ].Step 3: Verification Methods Davis outlines three ways to check the AI's work [ 06:36 ]:Self-Check: In the same chat, ask the AI to "rescan the document" to find supporting quotes for its own previous claims [ 06:56 ].Cross-Model Check: Take the first AI's analysis and the source document, then feed them into a different AI model to flag unsupported claims [ 07:35 ].NotebookLM: Upload your document and the AI’s analysis to Google's NotebookLM. It is specifically designed for grounded search and provides clickable source links for easy validation [ 08:25 ].Video Link: The 3-Rule Prompt That Stops ChatGPT, Gemini, and Claude From Guessing Gemini is AI and can make mistakes.
