Feed it right | Master AI Automation in 4 hours Master AI Automation in 4 hours Course About Ayush Modules Sample chapter Toolbox The Microcap Minute Classroom / Module 08: Documents, Data & Research / Chapter 1 Feed it right Watch first, then read. Same lesson, your pace. What you will learn – The three doors a document takes into a model, and when each wins – NotebookLM: the grounded reading assistant – Context hygiene for big piles Three doors for one document Door 1, Paste. Under ~20 pages of mostly text, pasting is fastest. It survives the trip well, loses tables/layout, and (Module 11 preview) should never carry sensitive personal data. Door 2, Upload to chat. Claude, ChatGPT and Kimi all accept PDFs directly; layout, tables and images survive. Best for one-off Q&A on big files. Files persist in that chat, or pin them to a Project (Module 2 Chapter 5) so every conversation there inherits them. Door 3, NotebookLM. Google’s free research assistant built entirely around your sources: upload many documents and it answers only from them , citing exact passages. Hallucination collapses because the desk holds your papers and little else, the strongest anti-hallucination architecture in Module 1’s terms. Bonus: it generates study guides, FAQs and even podcast-style audio overviews of your material. Decision rule: quick question on short text → paste; heavy file or recurring project material → upload/Project; serious multi-source study → NotebookLM. Context hygiene Big piles degrade answers (attention dilutes across a crowded desk): Split by purpose , “chapter 3 only” beats “entire textbook” for specific questions Say what to ignore , “use only section 4.2” focuses the model Name your files meaningfully , 2026-physics-syllabus.pdf beats doc_final_v3.pdf ; models read filenames too Re-anchor long chats , after many exchanges, restate key constraints; early instructions fade Try it yourself Take any real PDF (syllabus, circular, chapter). Run identical questions through all three doors, paste in Kimi, upload to Claude, load in NotebookLM: “What are the three most exam-relevant points?” Compare grounding: which cites pages? Which invents? Then practice hygiene: ask the same question twice in one chat, once raw, once with “use only page 12”, and note the precision gain in learn/feed-comparison.md . Key takeaways – Three doors: paste (fast), upload (fidelity), NotebookLM (grounded multi-source). – NotebookLM’s citations-only-from-sources design minimises hallucination. – Hygiene: split by purpose, direct attention, name files, re-anchor long chats. Download the exercise sheet (PDF) Module workbook (PDF) ← Module 08 index Next: Spreadsheets on easy mode → Classroom / Module 08: Documents, Data & Research / Chapter 1 Feed it right What you will learn – The three doors a document takes into a model, and when each wins – NotebookLM: the grounded reading assistant – Context hygiene for big piles Three doors for one document Door 1, Paste. Under ~20 pages of mostly text, pasting is fastest. It survives the trip well, loses tables/layout, and (Module 11 preview) should never carry sensitive personal data. Door 2, Upload to chat. Claude, ChatGPT and Kimi all accept PDFs directly; layout, tables and images survive. Best for one-off Q&A on big files. Files persist in that chat, or pin them to a Project (Module 2 Chapter 5) so every conversation there inherits them. Door 3, NotebookLM. Google’s free research assistant built entirely around your sources: upload many documents and it answers only from them , citing exact passages. Hallucination collapses because the desk holds your papers and little else, the strongest anti-hallucination architecture in Module 1’s terms. Bonus: it generates study guides, FAQs and even podcast-style audio overviews of your material. Decision rule: quick question on short text → paste; heavy file or recurring project material → upload/Project; serious multi-source study → NotebookLM. Context hygiene Big piles degrad
Feed it right
Written by
in