{"slug": "kevin-rose-ships-a-free-prompt-for-probing-what-ai-remembers-about-you", "title": "Kevin Rose ships a free prompt for probing what AI remembers about you", "summary": "Kevin Rose released Reflection Engine v1.3, a free prompt that asks an AI assistant to mine its memory of a user for blind spots, contradictions, emotional habits and likely trajectories, distributed via GitHub and X on August 3rd. The prompt, which includes 22 questions and requires confidence scores and evidence checks, depends on AI labs' memory infrastructure and aims to expose limits in AI-generated introspection. Rose, founder of Digg and currently investing at Basic Intelligence, noted that AI has been 'quietly assembling a complete behavioral record of you,' though OpenAI and Google document that their systems do not remember every detail.", "body_md": "[Kevin Rose (@kevinrose)](https://x.com/kevinrose) released [Reflection Engine](https://github.com/kropdx/reflection-engine), a free prompt that asks an AI assistant to mine its memory of a user for blind spots, contradictions, emotional habits and likely trajectories.\n\nRose distributed the file in [a two-post thread on X](https://x.com/kevinrose/status/2084354329604174037?s=20) on August 3rd, writing that AI had been \"quietly assembling a complete behavioral record of you.\" The GitHub repository and its four initial commits appeared the same day.\n\nRose is best known as the founder of Digg, the social-news site he began building in 2004. He later launched consumer products including the fasting app Zero and meditation app Oak, served as a general partner at Google Ventures and led watch publisher and retailer HODINKEE. His [current biography](https://www.kevinrose.com/about) says he is investing and building at [Basic Intelligence](https://basic.in/), the group behind the rebooted Digg.\n\nReflection Engine fits Rose's long-running interest in consumer software, personal habits and self-examination. It also depends entirely on infrastructure controlled by the major AI labs: the prompt becomes useful only when an assistant can retrieve enough prior conversations, saved memories and uploaded material to find patterns across time.\n\n### What the prompt asks\n\nThe downloadable [Reflection Engine v1.3 file](https://github.com/kropdx/reflection-engine/blob/main/Reflection-Engine-v1.3.md) contains 22 questions. They include: \"What truth about myself would I probably resist hearing?\", \"Where am I mistaking motion for progress?\", \"What contradiction in me explains the most?\" and \"What is the biggest bet I appear to be making with my life?\"\n\nRose built more than a list of provocative questions. The prompt tells the model to state which material it could access, map recurring themes before reaching conclusions and distinguish stable traits from temporary states, aspirations, isolated events and speech-to-text errors. It also requires the assistant to look for counterevidence and avoid treating a user's curiosity about a subject as proof of behavior.\n\nEach answer must carry a confidence score, a description of the breadth of its evidence and a status such as \"observed pattern,\" \"inference\" or \"insufficient evidence.\" Conclusions scoring seven or higher require at least two independent evidence points, preferably drawn from different periods or parts of the user's life. Every section ends with a practical action the user can test.\n\nThose controls target a central failure mode in AI-generated introspection: a model can produce a persuasive personality reading from thin or selectively retrieved evidence. Reflection Engine instructs the assistant to expose those limits inside the report, although the quality of that disclosure still rests on the model following the prompt and accurately understanding what it retrieved.\n\n### The record is incomplete\n\nRose's description of a \"complete behavioral record\" goes further than the platforms' own documentation. [OpenAI says](https://help.openai.com/en/articles/11146739-how-does-reference-saved-memories-work) ChatGPT can reference past conversations when chat history is enabled, while explicitly warning that it does not remember every detail. Saved memories and referenced chat history are separate systems, and users can disable either one.\n\n[Google's documentation](https://support.google.com/gemini/answer/16598623) says Gemini can personalize responses using past chats, preferences and connected Google services, although those capabilities are unavailable to some accounts and users. The resulting portrait therefore reflects what a provider retained and retrieved, rather than a neutral archive of everything a person has said or done.\n\nThe timing also puts Reflection Engine alongside an emerging class of memory-native AI features. On July 9th, Anthropic introduced [Reflect with Claude](https://www.anthropic.com/news/reflect-with-claude), a beta dashboard that summarizes topics, working patterns and usage over periods ranging from one month to one year. Anthropic's feature focuses on how people use Claude. Rose's prompt pushes the same underlying history toward broader judgments about work, relationships, habits and identity.\n\nReflection Engine adds no server, account or telemetry controlled by Rose. Users download the Markdown file and attach it inside an existing assistant conversation. That design avoids sending a second copy of the user's history to a separate startup, though the assistant provider remains inside the trust boundary.\n\nThe output creates its own privacy risk. Rose warns that the generated portrait may draw from conversations about health, money, family and relationships, and advises users to keep the report out of shared workspaces and team chats.\n\nReflection Engine's early form matters as much as its content. Rose packaged the workflow as a file rather than an application, leaving distribution, storage, inference and personal data with the model provider. That makes the release easy to copy and difficult to defend as a standalone business. It also shows where a new software layer is forming: prompts and agents designed to extract value from years of personal context already stored inside ChatGPT, Claude and Gemini.\n\nWhether that layer supports independent products will depend on how much memory the labs expose, how portable users can make it and whether people trust a third party to interpret their most sensitive conversations. Rose's release avoids the third-party data problem for now. It also demonstrates how quickly the model providers' memory features can become raw material for products they did not build.", "url": "https://wpnews.pro/news/kevin-rose-ships-a-free-prompt-for-probing-what-ai-remembers-about-you", "canonical_source": "https://runtimewire.com/article/kevin-rose-ships-a-free-prompt-for-probing-what-ai-remembers-about-you", "published_at": "2026-08-04 03:56:43+00:00", "updated_at": "2026-08-04 04:26:34.455467+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-tools", "ai-ethics"], "entities": ["Kevin Rose", "Reflection Engine", "Digg", "Basic Intelligence", "OpenAI", "Google", "Anthropic", "Reflect with Claude"], "alternates": {"html": "https://wpnews.pro/news/kevin-rose-ships-a-free-prompt-for-probing-what-ai-remembers-about-you", "markdown": "https://wpnews.pro/news/kevin-rose-ships-a-free-prompt-for-probing-what-ai-remembers-about-you.md", "text": "https://wpnews.pro/news/kevin-rose-ships-a-free-prompt-for-probing-what-ai-remembers-about-you.txt", "jsonld": "https://wpnews.pro/news/kevin-rose-ships-a-free-prompt-for-probing-what-ai-remembers-about-you.jsonld"}}