{ "$format": "contextdoc/nested-v1", "$legend": { "learn_more": "More six-word ideas, each published with its full subtext, live at https://sixwordidea.com. Every card on the index links to the idea's raw subtext JSON, fetchable at https://sixwordidea.com/ideas/.json.", "how_to_read": [ "The six words are sentence s-0001. Its text is the entire human-facing artifact.", "sentence.meta.backstory is the full idea the six words compress — claim, mechanism, stakes.", "sentence.meta.word_choices explains, word by word, why each word earned its slot and what alternatives were rejected.", "sentence.meta.drafts is the chronological evolution of the six words, each entry stamped with who wrote it (human or agent) and why.", "metadata.interview is the question/answer exchange between the writing assistant and the author that pinned the idea down.", "metadata.candidates_considered lists every candidate the assistant proposed, including the ones that were not chosen.", "Sentence provenance (authored_by, author_type, authorship_context) and metadata.edit_log answer who wrote the final words, when, and why." ], "what_this_is": "A concept packaged into exactly six words, plus the full subtext of how they were written. The human-facing text is the six words in the single sentence; the full argument, iterations, and human/AI collaboration that produced them are recorded on the sentence's meta and the document metadata." }, "sources": {}, "metadata": { "title": "The Unread 100%", "status": "final", "purpose": "A concept packaged into six words; the full argument and the process that produced them are the subtext.", "version": 2, "edit_log": [ { "at": "2026-08-06T15:32:24.334Z", "by": "claude-sonnet-5", "action": "create_story", "reason": "A plain, first-person-forcing question that makes every engineer in the room silently answer 'not me' before you've said a word.", "target_id": "s-0001", "requested_by": "adamconway" }, { "at": "2026-08-06T15:32:42.525Z", "by": "adamconway", "action": "finalize_story", "target_id": "s-0001" } ], "interview": [ { "answer": "I am giving a conference talk about the idea of compressing information into smaller packages like we're doing here. I would likike to work on compressing the first slide into six words. The idea I'm trying to compress is a question I'll pose to the audience - how many people read 100% of the LLM generated content they create?", "question": "What's the idea you're trying to compress? Give me the roughest, longest version of it first — I'll help sharpen it after." }, { "answer": "It's meant to be the question itself", "question": "Good hook. A few questions to pin this down:\n\n1. Is the six-word slide meant to *be* the question itself (posed directly to the audience), or is it a statement that sets up the question you'll ask verbally right after?" }, { "answer": "Yes, I strongly suggest that almost nobody in the room will say that read 100% of the LLM generated content they create/consume (code, planning documents, PRs).", "question": "Got it. Next: when you ask this, do you already know (or strongly suspect) the answer is basically \"nobody\"? I'm asking because a rhetorical question that indicts the room lands very differently than an open, genuinely curious one." }, { "answer": "It is primarily engineers and ai researchers. I don't want people to hear it as you're lazy/bad. I want people to start thinking about how to solve this problem as I'm going to be laying out a solution later in the talk", "question": "One more layer: who's in the room (engineers? execs? mixed?), and what's the wrong takeaway you want to head off — e.g., people hearing this as \"you're lazy/bad\" rather than \"the nature of authorship/verification has fundamentally changed\"?" } ], "created_at": "2026-08-06T15:32:24.334Z", "last_edited_at": "2026-08-06T15:32:42.525Z", "candidates_considered": [ { "story": "Who reads everything the AI writes?", "appeal": "A plain, first-person-forcing question that makes every engineer in the room silently answer 'not me' before you've said a word.", "chosen": true }, { "story": "Do you trust unread AI output?", "appeal": "Reframes the same fact as a trust problem, which is the frame you need for the solution you're about to pitch.", "chosen": false }, { "story": "Who really reads 100% of everything?", "appeal": "Borrows your own exact number back at the audience — sharper and harder to dodge than a paraphrase.", "chosen": false } ] }, "sections": [ { "level": 1, "blocks": [ { "type": "paragraph", "sentences": [ { "id": "s-0001", "meta": { "drafts": [ { "at": "2026-08-06T15:32:24.334Z", "by": "claude-sonnet-5", "text": "Who reads everything the AI writes?", "rationale": "A plain, first-person-forcing question that makes every engineer in the room silently answer 'not me' before you've said a word.", "author_type": "agent" } ], "backstory": "The opening slide of a conference talk for engineers and AI researchers poses a direct question rather than a claim: does anyone in this room actually read all of the LLM-generated content they produce or rely on — every line of code, every planning doc, every PR? The speaker already knows the honest answer is almost universally no. The question is designed to be answered silently and personally by each audience member before any data or argument is presented, so the room arrives at the insight through self-recognition rather than being told.\n\nThe mechanism is confrontation through specificity: by naming the artifacts (code, planning documents, PRs) and demanding a literal accounting — everything, not most of it — the question closes off the comfortable, vague self-assessment most people default to (\"I review my AI output pretty carefully\"). It forces a binary, embarrassing-if-honest reckoning with how much unread, unverified machine-generated material is already flowing through real engineering work.\n\nCrucially, the point is not to shame the audience for laziness — it's to establish, as a shared and undeniable premise, that verification at scale has already broken down as a practice. That premise is the setup: the talk goes on to propose a solution to exactly this problem, so the opening question needs to land as diagnosis, not accusation — the audience should walk away from the first slide thinking \"yes, this is a real gap,\" not \"I've been caught.\"\n\nThe stakes are practical and immediate for this audience: engineers and AI researchers are the ones shipping this unread content into production systems, codebases, and decision-making documents. The six words need to make that gap undeniable in one breath, so that everything that follows in the talk is received as necessary rather than hypothetical.", "word_choices": [ { "word": "Who", "reason": "Opens as a direct interrogative aimed at the room itself, not a hypothetical third party — forces first-person self-audit.", "alternatives": [ "Does anyone", "How many" ] }, { "word": "reads", "reason": "Concrete, checkable verb tied to the exact behavior being audited, rather than a vaguer term like 'reviews' or 'checks.'", "alternatives": [ "reviews", "checks", "verifies" ] }, { "word": "everything", "reason": "Plain-English stand-in for the speaker's own '100%' framing — insists on totality, not majority or 'most.'", "alternatives": [ "100%", "all of", "every line" ] }, { "word": "the", "reason": "Necessary connective that keeps the line readable as a spoken question rather than a clipped slogan.", "alternatives": [ "that" ] }, { "word": "AI", "reason": "Names the source plainly and matches the vocabulary of an audience of engineers and AI researchers.", "alternatives": [ "LLM", "model", "machine" ] }, { "word": "writes", "reason": "General enough to cover code, planning documents, and PRs at once without over-specifying to just one artifact type.", "alternatives": [ "generates", "produces", "outputs" ] } ] }, "text": "Who reads everything the AI writes?", "citations": [], "author_type": "agent", "authored_at": "2026-08-06T15:32:24.334Z", "authored_by": "claude-sonnet-5" } ] } ], "heading": "The Unread 100%", "subsections": [] } ] }