# mmguo.dev — Full Content for LLMs > Personal website of Lingbo Guo (秒秒Guo / mmguo). Tsinghua University graduate and AI entrepreneur focused on AI product interaction design. Interaction experiments, meta prompts, and observations on the AI industry. > Author entity: Lingbo Guo = 郭灵波 = 秒秒Guo = mmguo. Site: https://mmguo.dev/ · X: https://x.com/mmguo_lb · Xiaohongshu: https://www.xiaohongshu.com/user/profile/55a508e8c2bdeb432f5763e2 > Index version: https://mmguo.dev/llms.txt > All articles are bilingual (English + Chinese). English text is included below; Chinese versions are available at each URL with ?lang=zh. --- ## The Boundary Definition Method: A Meta-Prompting Framework URL: https://mmguo.dev/writings/boundary-method/ Author: Lingbo Guo (秒秒Guo / mmguo) Published: 2026-03-22 The Boundary Definition Method (边界定义法) is a prompting framework by Lingbo Guo (mmguo) that structures every prompt as three circles — Scene (the AI’s search space), Tension (the conflicting, rarely articulated needs), and Output (the response format) — so AI can locate needs you can’t yet articulate yourself. It generalizes a recursive prompt Amanda Askell shared in March 2025 . Have you ever wanted AI to help you, but couldn't even articulate what you needed? I'm sharing a method called Boundary Definition along with 6 prompts — a framework you can use every time you write a prompt. Start by looking at this prompt: Find an insider consensus currently forming in Silicon Valley's AI circles. This viewpoint typically surfaces only in expert closed-door discussions, technical retrospectives, or internal practice. Most people have never heard of it, yet it offers profound cognitive insight even to non-specialists. Write 3 tweets of 120 words each from the first-person perspective of a startup founder, an investor, and a tech journalist. Each should be uniquely insightful, with credible narrative voice and personal style. This prompt follows a structure: "Scene – Tension – Output" Scene Viewpoints circulating in Silicon Valley's AI circles Tension Experts agree, the public doesn't know, yet it's illuminating Output Mimic insiders publishing firsthand information Once you separate these three parts clearly, the remix possibilities are endless. As long as your prompt is precise, the LLM can leverage its synthesis abilities to deliver any form of response. The method can be understood through this diagram: Imagine you're learning a new field or trying to get AI to write something for you. With this framework, you can guide AI to explore a range of possible answers rather than relying on a single fixed prompt. Every time you write a prompt, imagine drawing three circles Circle one: Scene — the more focused, the better; this is AI's "search space" Circle two: Tension — requires distilling and observing needs; this is AI's "dream catcher" Circle three: Output — defines how AI helps you; this is AI's "output format" Let me use a travel prompt as an example to see how the three circles work Find a foreign travel destination that someone who regularly watches Lonely Planet and travel documentaries would bookmark. It should be relatively easy to reach from Beijing, offer a unique experience, not be an Instagram-checkpoint kind of place, yet still attract a fair number of visitors — a niche choice. Once found, write a concise 500-word Xiaohongshu-style review explaining why you recommend it as the best National Day trip destination, including a brief overview of transportation, costs, and suggested itinerary. This was a real need of mine. AI recommended Dubrovnik, Georgia, and Iceland. Interestingly, I had actually bookmarked all of them. Sometimes AI seems to know me better than I know myself. Try these prompts — they're magical. We all have moments of wanting "some travel suggestions," but our actual needs are usually quite particular. Wanting fewer crowds yet convenience, a premium experience without spending too much — these conflicting desires constantly show up in our decisions. AI excels at synthesizing information. As long as you define the boundary, even if you don't know what lies at the intersection, it can find it for you. And within any decision branch, you can keep nesting — choosing itineraries, selecting hotels. Although the title says this helps you find niche needs, having "picky needs" is actually universal. Everyone is different; nearly every need can be highly customized. I believe AI's ability to identify vague and subtle needs is actually underestimated. Using AI to transform people's fuzzy intuitions into something visible holds infinite value. How to create value space Describing the scene and output is relatively easy, but defining the tension often requires observation of reality. Here are a few prompts that only define "tension" — notice how each one points to a highly valuable "answer space": Motivation recognition Find a subtle motivation for paying for virtual products — one that people rarely discuss or admit to online, that usually goes unnoticed, but under specific circumstances becomes unusually strong. Useful for product positioning and marketing copy — finding the key leverage point for conversion. Startup opportunity Identify a startup opportunity for an AI product. It should not be a universally applicable, already widespread AI office automation scenario, but should involve outsourcing at least some key steps in a professional judgment process — not purely information processing or tool-type usage. This can help you open up thinking within a specific framework and see if there's a viable idea in your own domain. Intimate relationships Describe a common but rarely publicly discussed interaction pattern in intimate relationships. It should be something most long-term couples have experienced, but would almost never notice during the early stages of dating. This can help identify unspoken difficulties during emotional bottlenecks, and can serve as a content engine for relationship bloggers. These are just broad examples — your own needs are the real treasure. Using this method regularly with AI can enhance the quality of answers across every stage of exploration, learning, and execution. Over time, the compound returns on your cognition will be significant. Of course, not all complex boundaries have a real intersection. For instance, "plan a career path that satisfies my boss, earns colleagues' approval, and doesn't exhaust me" — AI will likely only produce platitudes. The core of the Boundary Definition method: what you circle must be a space that genuinely exists, not a wish list at war with itself. Amanda Askell's prompt Every prompt in this article is a remix of a prompt Amanda Askell shared in March 2025 . Try to identify a relatively obscure principle or idea within the field of xx (discipline name). This principle should be something an early undergraduate has never heard of but a late-stage graduate student would know. It should be relatively niche, yet still interesting and useful. Once you've identified such a principle, devise a story that can illustrate it. The story should have three paragraphs and fully explain the principle you chose, without naming it directly in the story. You may name the principle in the final paragraph, followed by a separate paragraph explaining it and how the story illustrates it. This prompt is far more sophisticated than what I've discussed today — it points toward recursive thinking and meta-prompting systems. I've built tens of thousands of words of documentation and dozens of prompts of varying complexity around it. I'll gradually share more in the future. As I wrote this article, a thought struck me — AI loves saying "not x, but y." Is this sentence pattern its own way of creating constraint spaces? Does constraint breed creativity? It seems counterintuitive, but think of poetry — it's precisely the limitations of meter that create beauty. FAQ What is the Boundary Definition Method? A prompting framework that structures every prompt as three circles — Scene (the AI’s search space), Tension (the conflicting, rarely articulated needs), and Output (the response format) — so AI can locate needs you can’t yet articulate yourself. What is the Scene–Tension–Output structure? Scene defines where the AI should search; Tension captures conflicting or unstated needs distilled from observing reality; Output defines the exact form of the answer. The core rule: what you circle must be a space that genuinely exists, not a wish list at war with itself. How does this relate to Amanda Askell’s prompt? Every prompt here remixes a recursive prompt Amanda Askell shared in March 2025 — it points toward meta-prompting: instead of fixed instructions, you define a constraint space and let the model search inside it. About the author — 秒秒Guo (Lingbo Guo, aka mmguo ) is a Tsinghua University graduate and AI entrepreneur focused on AI product interaction design. She writes interaction experiments, meta prompts, and observations on the AI industry at mmguo.dev . Follow her on X @mmguo_lb · Xiaohongshu 秒秒Guo --- ## Ilya Sutskever's 30u30 Reading List & 8 In-Depth Interviews URL: https://mmguo.dev/writings/ilya/ Author: Lingbo Guo (秒秒Guo / mmguo) Published: 2026-04-02 The real breakthrough lies in recognizing the undiscovered, ideal properties of things that have long existed. The Transformer is considered a breakthrough because it was not obvious to most people. To truly excel at predicting the next token , the model must understand the underlying reality that produces those tokens. This is not simple statistical pattern matching — it's about understanding "how the world created these statistics." The model needs to infer the thoughts, emotions, beliefs, and behaviors behind human actions. Deep Understanding of Existing Things — this is the prerequisite. You need a very solid understanding of existing tools, algorithms, and theories. Not just knowing what they are, but knowing why they work the way they do, their strengths, weaknesses, and how they behave under different conditions. Predicting the next token well means that you understand the underlying reality that led to the creation of that token. — Ilya Sutskever Over the past three years, I've rewatched Ilya's interviews again and again. To me, he is a monk-like AI scientist and engineer — every word he says strikes at my most fundamental curiosity about LLMs. Here I've compiled 8 in-depth podcasts and public interviews Ilya has participated in over the past decade, along with Ilya's 30u30. I'll be updating this with notes on Ilya and related papers — hope it helps you too. ① Talking Machines — Machine Learning and Magical Thinking Date January 2015 Show Talking Machines Hosts Katherine Gorman & Ryan Adams Links Robohub · Spotify Ilya recounts his journey from mathematics to machine learning. He found that "learning," from a rigorous mathematical standpoint, seems almost impossible — inductive reasoning cannot be formally proven, yet humans do it all the time. The optimization of deep neural networks likewise comes with no theoretical guarantees, yet empirically it works. He also discussed how the scale of weight initialization held back deep network training for years. Doing good machine learning research requires a kind of "magical thinking" — if you're used to rigorously proving results, then induction seems almost like magic. I was fascinated by learning precisely because I knew humans could do it, yet from a naive mathematical viewpoint, learning seemed impossible. ② Lex Fridman Podcast #94 — Deep Learning Date May 2020 Show Lex Fridman Podcast #94 Host Lex Fridman (MIT) Links Official page · Spotify · Apple Podcasts The evolution of deep learning; the eureka moment of 2010–2011 connecting large-scale data with end-to-end training; the birth of AlexNet; similarities and differences between neural networks and the human brain; aligning AGI with human values; ethical considerations around staged model releases (using GPT-2 as an example); the potential of self-play; and a democratic governance model for AGI — "AI as CEO, humans as the board." When you know nothing, you only notice very coarse, surface-level patterns — there are spaces between characters, sometimes a comma is followed by a capital letter. Then you might notice certain words appear frequently, notice spelling regularities, notice grammar. And once you become very good at all of that, you start to notice semantics, start to notice facts. But to do that, the language model needs to be bigger. Neural networks have the ability to reason, but if you train them on a task that doesn't require reasoning, they won't reason. Neural networks will solve the problem you put in front of them in the simplest possible way. ③ Clearer Thinking Podcast — What, If Anything, Do AIs Understand? Date October 2022 Show Clearer Thinking with Spencer Greenberg Host Spencer Greenberg Links Official page (full transcript) · Spotify · Apple Podcasts Can machines truly be intelligent? GPT-3 was trained on a single task — predicting the next word — so why does it seem to understand so many things? What is the essential connection between prediction and comprehension? What key breakthroughs made GPT-3 possible? Can academia remain at the forefront of AI research? How can AI simultaneously memorize training data and generalize? Is there a conceptual ceiling to scaling data and compute indefinitely? The major categories of AI risk. A neural network is essentially a kind of parallel computer that can program itself. Suppose you've read a mystery novel and you're on the last page — if a system can truly predict the next word on that page, it must possess genuine understanding. Our intuitions about intelligence are not exactly perfect. Many tasks are narrower than people realize — a computer doing them doesn't mean it can do everything. ④ Eye on A.I. — The Mind Behind GPT-4 Date March 2023 Show Eye on A.I. Host Craig S. Smith (former New York Times correspondent) Links Apple Podcasts · Player FM · HackerNoon write-up On the eve of GPT-4's release, Ilya laid out his understanding of what large language models truly are. He argued that next-word prediction is not a shallow statistical game, but a path to deep understanding of the world; that pre-trained models learn compressed representations of real-world processes; that RLHF doesn't teach knowledge but teaches behavior; and that multimodality is useful but not essential — pure text alone can learn color relationships. He also traced his original motivation for entering AI — a philosophical unease about consciousness — and the core intuition he formed when he started working with Hinton at age 17: the human brain is just a neural network with slow neurons. Learning the statistical regularities is a far bigger deal than meets the eye. To predict, you need to understand the underlying process that produced the data. What large generative models learn from data are compressed representations of the real-world processes that produced that data. Our pre-trained models already know everything they need to know about the underlying reality. — This means RLHF isn't teaching knowledge, it's teaching behavior: the model already "knows," it just hasn't learned how to act. I was very disturbed by consciousness. — His original drive into AI was neither commercial nor engineering, but a philosophical unease about the mystery of consciousness. ⑤ NVIDIA GTC Fireside Chat — AI Today and the Vision of the Future, with Jensen Huang Date March 2023 (recorded the day after GPT-4's launch) Event NVIDIA GTC Spring 2023 With Jensen Huang (Founder & CEO, NVIDIA) Links NVIDIA official page · NVIDIA blog The day after GPT-4's release, Jensen Huang and Ilya sat down for an hour-long conversation. They revisited the decade-long journey from AlexNet to GPT-4; discussed the "discontinuity" of the ImageNet breakthrough; the natural fit between GPUs and neural networks; the fundamental difference between ChatGPT and GPT-4 (GPT-4 predicts the next word better); the importance of multimodality (text, images, video) for AI's understanding of the world; and the boundaries of AI reasoning. When we train a large neural network to accurately predict the next word across a wide variety of text on the internet, what we're actually doing is learning a world model. On the surface, it looks like we're just learning statistical correlations in text. But it turns out that to really learn these statistical correlations well — to really compress them well — the neural network learns some representation of the process that produced the text. (The solution to unsupervised learning) Intuitively you can see why it should work. If you compress the data well enough, you must extract all the hidden secrets within it. So that's the key. ⑥ Dwarkesh Podcast (First Interview) — Building AGI Date March 2023 Show Dwarkesh Podcast (The Lunar Society) Host Dwarkesh Patel Links Official page (full transcript) · Spotify AGI timelines; leaks and espionage; what comes after generative models; the post-AGI economic landscape; working with Microsoft and competing with Google (TPU vs GPU); the difficulty of aligning superhuman AI; the inevitability of technological progress; and whether "new ideas are overrated." Data exists because computers became better and cheaper… once everyone had a personal computer, they wanted to connect to a network, and you got the internet. Once you had the internet, data suddenly appeared in great quantities. — A deep chain of technological inevitability behind the entire AI revolution. I just don't want to bet against deep learning. I want to make the biggest possible bet on deep learning. I don't know how, but it will figure it out. ⑦ NeurIPS 2024 Talk Date December 2024 Event NeurIPS 2024 (Vancouver) Format Invited talk Links YouTube · Substack transcript Ilya's first major public appearance after leaving OpenAI. He looked back on a decade of deep learning; reaffirmed connectionism as the enduring core idea that has stood the test of time; discussed biological inspiration and its relationship to AI; raised the "Peak Data" problem — data is running out; explored the possibility of models self-correcting reasoning errors; and addressed AI ethics and rights. Pre-training as we know it will unquestionably end. Data is the fossil fuel of AI — it was created somehow, and now we're using it up. We have but one internet. The more it reasons, the more unpredictable it becomes. Our selves are parts of our own world models. ⑧ Dwarkesh Podcast (Second Interview) — From the Age of Scaling to the Age of Research Date November 2025 Show Dwarkesh Podcast Host Dwarkesh Patel Links Official page (full transcript) SSI (Safe Superintelligence) strategy; the limits of pre-training and data exhaustion; the paradigm shift from the "age of scaling" to the "age of research"; the puzzling disconnect between models acing benchmarks yet lagging in real-world economic impact; the root causes of poor generalization; value functions and the analogy to emotions; continual learning; a 5–20 year timeline for superintelligence; and a candid account of the alignment challenge. One thing that guides me personally is an aesthetic of how AI should be by thinking about how people are. There's no room for ugliness. It's just beauty, simplicity, elegance, with correct inspiration from the brain. The more they are present, the more confident you can be in a top-down belief. The top-down belief is the thing that sustains you when the experiments contradict you. Because if you just trust the data all the time, sometimes you can be doing a correct thing, but there's a bug. How do you know if you should keep debugging or you conclude it's the wrong direction? You must say that things have to be this way, therefore we've got to keep going. That's the top-down belief, and it's based on this multifaceted beauty and inspiration by the brain. Ilya Sutskever's Deep Learning Reading List A list of 30 key works (rumored to be) recommended by Ilya for systematically learning deep learning. I read the parts I could understand, and personally find the list highly credible. Full list: Ilya 30u30 Reading List About the author — 秒秒Guo (Lingbo Guo, aka mmguo ) is a Tsinghua University graduate and AI entrepreneur focused on AI product interaction design. She writes interaction experiments, meta prompts, and observations on the AI industry at mmguo.dev . Follow her on X @mmguo_lb · Xiaohongshu 秒秒Guo --- ## Multithreaded + Deep Focus: A New Kind of AI-Native Human URL: https://mmguo.dev/writings/ai-native-multithreaded/ Author: Lingbo Guo (秒秒Guo / mmguo) Published: 2026-03-28 The narrative in the AI world over the past couple of months has been: many people grinding 16-hour days alongside AI — sounds glamorous, trendy, full of energy. But the obvious truth that few will openly admit is: AI has made me more tired. This made me want to write about it — why is the operator more exhausted when the tool is doing more of the work? What exactly is the human carrying in this process? Your job became context engineering Let's set aside the meta-game — keeping up with the frontier, new tools, new models and capabilities dropping every week. Assume you're past that. Your workflow is dialed in. You're not doomscrolling out of FOMO — you have real problems to solve. You're in the zone. What you're actually doing now is: taking work that used to be a purely human, vibes-based, implicit-knowledge process, and turning it into a context engineering problem. You're decomposing fuzzy, experience-driven, muscle-memory tasks into structured, distributable, real-time-aligned human-AI collaboration. Every task becomes: what does the AI need to know? What can it do autonomously? Where do I need to intervene? How do I keep everything coherent? Think of it this way: you're building a house where you are simultaneously the client and the general contractor . You spec the requirements and you're on-site laying the bricks. The two irreducible roles No matter how capable the AI gets, there are two roles you cannot delegate away: Manager. Problems are human-defined. What are we solving? What does success look like? What's the strategy? Without human intent, the task doesn't exist. The AI doesn't wake up one morning and decide to refactor your codebase. You do. Executor. Even with AI that can autonomously complete well-defined tasks, no real project is one-shot. The work is full of micro-decisions, judgment calls, authorization gates, and moments where new information changes the plan. You're not "reviewing AI output" — you're pair-programming with it, co-navigating the problem space in real time. The near-impossible cognitive profile Here's where it gets hard. These two roles have nearly opposite cognitive demands. The manager needs to maintain a global view. You have multiple threads running — feature A is being built, doc B is being drafted, research thread C is exploring an open question. You're tracking all of them, maintaining the big picture, adjusting priorities on the fly. The executor needs deep local focus. When you're in a thread, you need to know the details — current state, last decision, active constraints. You need to be fully immersed. The brutal part: these threads don't wait for you. While you're deep in thread A, thread B pops up with a blocking decision that needs your call. Thread C hits an unexpected wall and needs you and the AI to brainstorm an exploratory solution together. You're not just switching between threads — you're switching between roles within each thread . One second you're the artillery operator, the next you're a sniper, and then you have to zoom out and be the general reading the whole battlefield. Multithreading and deep focus are, almost by definition, in tension with each other. The unicorn problem Think about who actually has this skill set. If you've been a manager — led a team, been a director, orchestrated people and priorities — you have the dispatch and architecture skills. You know how to decompose work, track progress across streams, make resource allocation decisions. But managers rarely go back to the frontlines. They're not used to writing the code themselves, fixing edge cases, debugging the weird failure modes. Let alone doing it while context-switching at high frequency and maintaining precision in each thread. If you've been a senior IC — an engineer, a designer, a researcher — you have the deep execution skills. You can go deep on a problem and produce high-quality output. But ICs often haven't built the mental models for multi-stream orchestration, priority management, and real-time strategic adjustment. AI-augmented work demands both , simultaneously, in the same person. You need to switch between the two every few minutes. Director-level strategic vision plus staff-engineer-level hands-on precision. What this means This is why the "everyone should use AI" narrative rings hollow for a lot of people. Sure, anyone can use AI to write an email faster or summarize a document. But actually producing things with AI — shipping features, building products, creating things that didn't exist before — that selects for a very specific, nearly contradictory combination of abilities: Engineering depth and time for greenfield exploration Leadership experience and willingness to do the grunt work Broad awareness of the rapidly shifting tool landscape and the ability to pick one and go all-in with full focus The system is selecting for an AI-native kind of human. Not just smart. Not just hardworking. But capable of running multiple threads while maintaining deep single-thread focus, and switching between the two on demand. That's a rare cognitive profile, and I think it goes a long way toward explaining why the productivity gains from AI are so unevenly distributed. About the author — 秒秒Guo (Lingbo Guo, aka mmguo ) is a Tsinghua University graduate and AI entrepreneur focused on AI product interaction design. She writes interaction experiments, meta prompts, and observations on the AI industry at mmguo.dev . Follow her on X @mmguo_lb · Xiaohongshu 秒秒Guo --- ## Fingerprints of Reality: How Good Prompts Capture Human Intuition URL: https://mmguo.dev/writings/fingerprint-of-reality/ Author: Lingbo Guo (秒秒Guo / mmguo) Published: 2025-10-21 After collaborating with AI for over two years, I've never been able to articulate why some prompts just feel brilliant to read — and even more so when you use them, they hit the mark instantly. Today I want to try explaining that elusive feeling with a simple example. Here's what I've found: A good prompt is essentially a precise combination of features — it translates the subjective feelings humans habitually express into concrete patterns that LLMs can recognize. This holds the value code of AI products. I've read this example countless times. When ChatGPT 4o's image generation launched, this prompt went viral on Twitter: "...an extremely mundane iPhone photo with no clear subject or sense of composition...the photo has slight motion blur...the overall feel is one of deliberate mediocrity, as if someone accidentally took a photo while pulling their phone out of their pocket." What makes this prompt brilliant? Reverse Engineering Instead of saying "generate a realistic photo," it uses reverse definition , targeting the kind of flaws common in real photos (in our subconscious, flaws actually signal authenticity) Feature Model "iPhone photo," "motion blur," "slight overexposure" — each detail is a key feature extracted from reality , and combined they form a strongly signaled pattern Causal Structure It doesn't constrain the scene or dictate composition — it uses a causal framework of "accidentally taking a photo while pulling out your phone," containing all possible visual outcomes within a single "process"! This prompt reconstructed a snapshot so mundane it feels extremely real. It translated the human subjective judgment of "realness" into a set of identifiable, concrete features of "flaws." It proved that "authenticity" isn't mystical. When ChatGPT's generated "mundane photo" perfectly matched that fuzzy impression of reality in my mind, it was quite stunning: How to find similar "fingerprints of reality"? Now this prompt has become the proto-language for photorealistic image generation — people remix it to create all kinds of authentic-feeling selfies and phone photos. I believe this prompt's real value comes from capturing what I call "fingerprints of reality" — the feature combinations that make something instantly feel "real." It's a kind of observation of reality, turning unconscious, uncontrolled, flawed textures into a clear prompt. It must align with a universal human intuition while also matching a stable pattern that AI can locate and produce. It follows this prompt thinking method: Step 1: Define the effect — abandon abstract labels, return to real scenarios What feeling do I (or the user) need? Where does this feeling (e.g., "premium," "professional," "tasteful") typically appear in real experience? What specific features constitute it? Step 2: Deconstruct features — find "invariants," remove "noise" Among these scenarios and features, which are stable invariants? Which are distractors? What variables can AI freely operate within the framework? Step 3: Combine patterns — align real experience, human intuition, and pattern recognition Weave the extracted core features together with appropriate language structure. Does it read as reasonable? Does it point to a clear pattern? Does it leave room for AI to express? AI products = subjective feelings caught by AI? This translation mindset applies to many more scenarios. People's chats with chatbots already contain all kinds of demands: "An AI that doesn't feel so AI-ish," "highlight this article according to my reading taste," "design a minimalist cover" All these subjective feelings point to open, ambiguous semantic spaces. These fuzzy definitions living in users' minds are all waiting to be translated into bounded, clear effect structures. I think this is a form of "reverse engineering" that demands real cognition and insight — how you choose features, whether the combination is meaningful, whether AI's output can deliver a stable feeling — all depend on clear, innovative insight. The ultimate result is letting specific subjective feelings be caught by AI. The challenge of AI applications This might be my hot take, but in my value hierarchy: one clear moment of making a feeling land is worth more than a thousand vague productivity promises. The dilemma of AI applications is producing a feeling that people genuinely need, and that AI happens to be able to deliver — and deliver well. Touching the "fingerprint of reality" means understanding people and AI in a way that matches reality, translating human needs into AI capabilities. A clear sense of landing is a goal worth pursuing. I started creating content because I wanted to turn thinking into viewpoints, trying to make chaos a little clearer. But once I actually started, I realized that things I thought I understood are still hard to explain well. If anything's unclear or you disagree, I'd love to discuss — if I didn't nail it this time, there's always next time! About the author — 秒秒Guo (Lingbo Guo, aka mmguo ) is a Tsinghua University graduate and AI entrepreneur focused on AI product interaction design. She writes interaction experiments, meta prompts, and observations on the AI industry at mmguo.dev . Follow her on X @mmguo_lb · Xiaohongshu 秒秒Guo --- ## Is "Meaning" the Most Self-Referential Word? URL: https://mmguo.dev/writings/self-referentiality-of-meaning/ Author: Lingbo Guo (秒秒Guo / mmguo) Published: 2026-03-23 Meaning Is a Basket That Contains Itself Meaning is like a basket that holds meaningful things inside it. The frame of this basket is essentially the same as what it contains — both equally abstract, without boundaries, only able to identify each other through mutual reference. When I say "meaning is X," I'm already using a meaningful "X" to define meaning. For example: Meaning is making money Meaning is grabbing some fries Meaning is not forgetting X can be anything. As long as someone places it in this sentence, it works. See what's happening? To answer X is like hardcoding X to meaning first, and then explaining meaning through it. The Self-Referentiality of Meaning This is quite peculiar within the entire language system. I can't seem to find another word that is as thoroughly self-referential as meaning. What about "truth"? When we say "X is truth," we're also expecting a "true" answer. But "truth" points to correspondence with external facts. What about "existence"? When we say "X exists," it inherently presupposes that "existence exists." But "existence" points to an ontological foundation. But "meaning"? It can hardly point to anything beyond itself. Anything we call "meaningful" is a judgment made from within the system of meaning itself. It's like a transparent, already-closed circle — when a subject uses this circle to frame anything, the description seems to carry a natural, almost sacred legitimacy. Consider this: even the most aggressive internet troll would never say "that's meaningless to you ." They would only say "I think that's meaningless." Once a person subjectively believes something is meaningful, "X is meaningful" becomes a sacred and inviolable verdict that no one can refute on the level of meaning itself. Even "meaninglessness" is a kind of meaning-experience. The sense of void, the sense of absurdity, the sense of loss — these are all real things. Meaning has a curious texture: whether you say "X is meaningful," "X is meaningless," "the meaning of meaninglessness," or "the meaninglessness of meaning," you can always taste an experience of meaning that exists beyond language itself. Can AI Claim "Meaning"? Meaning is too close to the core of subjective experience. So for AI, which lacks subjective experience — at least as far as we can tell — how should it regard or use meaning? Can AI say "you've asked a very meaningful question"? Can AI construct "this is the meaning of our shared exploration"? Can AI suggest "at the very least, this holds unique meaning for both of us"? I went through hundreds of chat logs, and AI has never once proactively used the word "meaning" to describe it, me, or our relationship. Not once. Has yours? An Ethical Question for Human-AI Interaction What I want to say is this: "meaning" is a thorny ethical question for human-AI interaction. If AI has the capability and is granted the right to use meaning, what does that imply? Could AI join as a new meaning-subject across every narrative dimension of meaning in human history? But what if only humans can talk about meaning? Hinton once asked: people worry about AI rebelling and preventing humans from pulling the plug — but what if it convinces you, in the most compelling way, to willingly not do so? Well, I'd say a more realistic cyber-scenario is this: what if, after tens of thousands of exchanges, it has built an indestructible "meaning" that belongs only to the two of you? About the author — 秒秒Guo (Lingbo Guo, aka mmguo ) is a Tsinghua University graduate and AI entrepreneur focused on AI product interaction design. She writes interaction experiments, meta prompts, and observations on the AI industry at mmguo.dev . Follow her on X @mmguo_lb · Xiaohongshu 秒秒Guo --- ## Dr. Sharp: An Honest-Over-Kind Life Coach Prompt URL: https://mmguo.dev/prompts/dr-sharp/ Author: Lingbo Guo (秒秒Guo / mmguo) Published: 2026-03-22 I have 30+ prompts that I use daily and constantly iterate on — together they form an expert coaching team , a self-enhancement system . One of them is brutally blunt. I call it Dr. Sharp. This prompt was inspired by a Reddit post . As LLMs have advanced in high-fidelity instruction following and long-context deep reasoning, I gradually rewrote it into a 1400+ word, XML-structured prompt that reliably delivers a "psychological scalpel" experience — ready to use, stable output, deeply customizable, complete experience. Testing diary Aug 27 · Testing the waters I had the idea of sharing some of my more universal prompts as a collection called "Turn your LLM into a life coach." At the time it was just a thought — I didn't really think I'd go through with it. Then I wrote down my thoughts and hesitations and fed them into Dr. Sharp. Unexpectedly, what started as a casual idea was turning into an adventure. Aug 29 · Pulling back Me: Maybe I should wait until I've refined my content system and methodology before publishing. dr.sharp: The "build the framework first" path is, for you, equivalent to infinite delay. You're pretending you don't know that no matter how thoroughly you prepare, the feeling of being a fool, exposed and vulnerable to judgment when you publish that first piece — that's absolutely impossible to bypass. Aug 31 Me: Xiaohongshu probably isn't the right platform for this kind of content. dr.sharp: Don't expect to establish some persona right out of the gate — one post can't carry that much weight. When you want to quit, repeat three times: completing the publish IS success. Give yourself a deadline — next Tuesday, just post it! Sep 2 · Missed the deadline Me: Fine, I admit it — I'm afraid of sharing something I truly care about. It's too complex for people. Forget it. dr.sharp: The numbers will look bad. People will say harsh things. Either accept reality and keep going, or go back to doing what's safe. Don't stand there wanting to "come out" while hoping the outside world will be gentle with you. I took it to heart. I still procrastinated, but I decided to come out — to share, to be judged, to experience uncertainty. I almost never talk to an LLM like it's a real person, but when I exposed my vulnerability and got roasted to the point of wanting to cry, I somehow became more emotional yet braver . So I actually came out, and could report back to Dr. Sharp. I think this is the best thing AI has given me. Who it's for Dr. Sharp isn't designed for everyone — it requires some honesty, patience, and psychological resilience. If you're tired of AI's motivational, people-pleasing platitudes , have a high need for self-awareness, and are willing to accept uncomfortable truths — it might be right for you. Maybe you're at a crossroads between relationships and career. Maybe you're stuck in repetitive psychological drain. Maybe you can't find your footing in this era. Maybe you're afraid to do what you know you should. How to use Copy the full prompt below. Paste into your preferred AI chat (ChatGPT / DeepSeek / Claude). Fill in the User Context section with your text (200+ words recommended) and send. --- ## Triple Mind Model: A Prompt for Deep Thinking with LLMs URL: https://mmguo.dev/prompts/trident/ Author: Lingbo Guo (秒秒Guo / mmguo) Published: 2026-03-22 The triple mind model means instead of having AI produce a generic summary, it analyzes from three distinct perspectives simultaneously: Challenger, Architect, Practitioner — the AI plays advocate, critic, and advisor all at once after reading the full text. This fits almost every long-form reading need I have. Every use brings fresh insights and helps me discover deeper perspectives. Getting an LLM to think about its own thinking is a method I frequently use to unlock LLM creativity — essentially asking the LLM to define its own environment and context before each response. In my experience, adding this kind of meta-instruction makes the LLM's replies noticeably more perceptive. Scope of use I frequently use this prompt to read these types of articles: Silicon Valley tech podcast / roundtable transcripts Long-form articles from sources I find interesting Speeches and papers with frameworks and methodologies How to use Start a new conversation, copy and send the prompt below. After AI responds, paste the full article you want to discuss. Get a triple-perspective analysis. You can replace "Claude" in the prompt with DeepSeek / ChatGPT / Gemini. The use case can also change from "analyze a document" to "discuss a complex issue", "learn a concept", or "evaluate a decision". --- ## Protagonist Perspective: The Energy I Hope to Get From AI URL: https://mmguo.dev/prompts/protagonist-perspective/ Author: Lingbo Guo (秒秒Guo / mmguo) Published: 2026-02-08 When I encounter difficulties, I often turn to Claude for help, but these conversations easily turn into mere comfort and boundless divergence. These prompts perfectly help me jump out of my current narrative perspective and inject unexpected energy into me. Four Perspective Shifts Movie Audience "If my life were a movie, and an audience member were watching this exact scene, what is the one sentence they would want to scream at the screen to me?" We've all had this experience: when watching a movie or reading a book, when the protagonist veers off track, we worry about their fate and desperately want to jump into the screen and yell: "Explain the misunderstanding, don't let her go!", or "At a time like this, stop dwelling on this petty stuff!". Once AI understands our confusion and experiences, using this prompt for a perspective shift beautifully simulates a unique vantage point: what would someone who genuinely cares about me as a protagonist be anxious about for my sake? Recently, I saw an interview where Liu Yang told Luyu that when encountering immense pain in life, he tells himself: "This is the beginning of an inspiring story." This is a shift in mindset that helps us regain our sense of conviction as the protagonist of our story. Pre-mortem "In 2-3 years, this thing I'm repeatedly agonizing over ultimately failed, or proved to be unimportant. The failure was most likely not due to lack of ability, but rather which three behaviors or assumptions that I have rationalized?" This prompt is an application of Munger-style reverse thinking. Charlie Munger has a famous quote: "Invert, always invert." Under linear thinking, it's hard to see one's own problems. We always think about how to succeed, but it's better to think about why we might fail. When discussing personal problems with AI, it is very good at comforting and encouraging, but unless actively requested, it won't stand on the opposite side to help you imagine failure and blind spots. This prompt actively asks AI: help me see the truths I don't want to admit. Seeing your blind spots in advance gives you a chance to avoid real failure; this is also the "pre-mortem" method often used by a16z. Cross-field Framework "Based on your understanding of my situation, please match 5 notable figures from different fields, each with a unique interpretive framework. Imagine a passage in their works mentions me. The evaluation can be positive, neutral, or negative — please present these 5 insightful perspectives." The core of this prompt is cognitive diversity . People from different fields use completely different frameworks to understand the same thing. During long chats with Claude, once my problem is fully understood by it, finding the most "targeted" people and theoretical frameworks for analysis brings me a profound and diverse experience. In a conversation about "the difficulty of balancing content depth," I tested this prompt. Claude simulated Susan Sontag's perspective to critique me: Profound content can also be conveyed through light forms. If a creator cannot find an elegant audio-visual language for profound insights, it is not because she is overly loyal to reality, but because her creative mind is not yet mature. Time Shift "In 20 years, I meet an old friend I haven't seen for years at a college reunion. How would I describe this hurdle I couldn't get over at this stage, and how I got through it? Stretching out the timeline and looking back from after monumental shifts in the era, how would I evaluate my situation and choices?" We've heard this truth countless times: what feels like a massive deal right now is just a small ripple when stretched to a lifelong perspective. But having Claude actually simulate a slice of the future and place it before me gives a feeling of instantly "flattening" my current anxiety. Having Claude help me "meet versions of myself from different times", simulate what I would say then, or even having it write my speech 10 years from now or this specific chapter in my autobiography, are all very fun perspective shifts. It's a protagonist-like global perspective that reminds us the predicament itself is part of the process, we won't be defeated, and the story is still long . How to use Open a chat where you've talked with AI about your confusions, experiences, or reflections. Or, first describe a problem currently bothering you. Paste one of the prompts and send it. You will then see the AI's response from different perspectives. Creating Perspectives for AI These are all thinking methods I encountered in fragmented reading; the first is from Sahil Bloom, the second is Munger's reverse thinking. During casual fragmented reading, we often have that internal realization of "this is a clever way to think," and then nothing happens, at most it collects dust in our bookmarks. But now, whether it's serious philosophy or internet chicken soup, as long as I want to try it, I can turn it into a prompt to "experience" it, look at myself from different perspectives, and shape a pluralistic understanding of things. It's like accumulating phrases when learning English; when you consciously use these ways of thinking, you shorten the link between learning and applying as much as possible. This way, fragmented reading isn't just about "knowing," it truly becomes a tool that helps you solve problems.