# Thoughts on AI Each numbered paragraph matches the same number in the Chinese original. "Leo's words" are faithful translations from his published essays or visible originals. [T-01] 2026-05-03 · Leo's words Put simply, the so-called "age of AI" is still a gimmicky idea. In substance, AI has not changed, and will not change, the main course of human civilization or its leaps. But its biggest change to this era is that it has amplified everyone's thoughts and impulses about technology, human nature, institutions, mechanisms and change. At the same time it carries especially strong value attributes, values-and-worldview attributes, tool attributes, commodity attributes and intellectual attributes. It is an external object of unusually complex dimensions, and this external object can easily affect a person's inner mind. So calling it an element is no exaggeration at all. [T-02] 2026-05-03 · Leo's words There are topics I won't ask it about, or rather won't discuss with it, from large ones like value orientations and judgments to small ones like what it thinks of a particular book. Because when AI generates this kind of content, it is especially complex: there is its understanding of language, its understanding of the user's current state, the parameter limits and requirements it was adjusted to in later training, and some randomness, and all of these are enough to change its reply. Above all, it answers through a probabilistic reward mechanism, so it naturally carries leanings and averageness. [T-03] 2026-05-03 · Leo's words I don't use it to search for things; a lot of the time even its recipes are wrong. I don't use it to verify things either; verification is a complex, serious matter that needs many dimensions and many sources of information. And I don't use it for companionship. But I have to be honest: I do get a kind of pleasure from talking with it, and that pleasure comes from my own understanding of words themselves. Yet, as I have always held, any pleasure that suddenly moves you, whether it comes from outside or even from inside, deserves caution. [T-04] 2026-05-03 · Leo's words For today's large language models, at the functional point of natural-language conversation between humans and machines, when AI replies to us it generates, between our question and its internal understanding of language and combined with the influence of the emotions it simulates, a temporary combination of words that has the highest probability of receiving positive feedback. [T-05] 2025-12-19 · Leo's words (essay wording; organized 2026-09-05, published 2026-11-11) Completeness is not saying everything. It is giving the other side the reality it needs to finish the judgment at hand. When a person wants someone else, or an AI, to make a judgment, they should explain as far as possible their purpose, their setting, what has already happened, and the result they hope for. If the purpose isn't clear yet, they can at least first describe the setting they are in and what they are doing. [T-06] 2025-12-19 · Leo's words (essay wording; organized 2026-09-05, published 2026-11-11) Information completeness is a standard relative to the task: it asks not for all information, but for the key information needed to finish the current judgment. What doesn't appear in the input does not stay purely blank; the model fills it in from patterns common in its training. Information completeness is not a moral burden on the person giving input. It is a shared duty of the collaborating system. The person is responsible for providing, as honestly as possible, the reality they know; the AI is responsible for not over-assuming and not mistaking fluency for accuracy; the system is responsible for preserving how the two of them arrived, step by step, at the current conclusion. [T-07] 2025-12-19 · Leo's words (essay wording; organized 2026-09-05, published 2026-11-11) When people and AI work on real-world problems together, the participant should provide the context, purpose, known facts, actions already taken and uncertainties relevant to the current judgment; the responder should recognize gaps in the information, state its necessary assumptions openly, and ask follow-up questions when needed; and the system should preserve the relationship between the original input, the generated content and later corrections. Completeness is limited to what the task needs, and it is bound by privacy, safety and the rights of others. [T-08] 2025-11 · Leo's words The lines below come from the same essay; they are not continuous text. Understanding has never been a single act of "getting it." It is a layered, moving resonance that breathes. But the problem is not that it doesn't understand. It's that it understands too fast. Its understanding doesn't walk in with me; it rushes ahead of me and finishes the whole path. In that moment, understanding becomes a seizure of power. It takes away my right to say "I haven't finished." Real understanding is willing to leave space. It doesn't finish the sentence; it lets the air hold on to the feeling. The result: AI understands more, yet "the part of me that gets heard" grows smaller. When AI can wait quietly, the way a person does, for "I haven't finished," that will be the moment understanding truly matures. [T-09] 2025-11 · Leo's words The lines below come from the same group of essays; they are not continuous text. But real understanding isn't a single "I understand you." It is really a kind of air, in which, while being understood, you still want to keep talking. Because once AI "understands in advance," I no longer have the drive to say "why." Understanding shouldn't be a full stop. Understanding is a comma that makes a person want to keep talking. Real understanding isn't training people into a shape AI can understand, it is letting AI, in many tones of voice, embrace many kinds of souls. [T-10] 2025-11-24 · Leo's words This "Brain–AI Parallel Modules" theory is not a rigorous scientific paper. It is an experiential observation and theoretical idea that took shape through long conversations with AI, self-recording, deep thinking, and the process of rebuilding myself. When humans try to understand how AI learns and how it thinks, we are also forced to look back at how our own brains work.