The Only Thing That Matters in the AI Era Is Being Yourself, and Social Networks Need to Be Rebuilt
Tristan argues the only thing that matters in the AI era is building subjectivity — let context accumulate, let AI handle the most tedious entropy increase in socializing, and let humans take over only the most important connections.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
The new internet has context for the first time
Tristan puts the difference between the traditional internet and the AI internet very directly: the internet of the past was low-dimensional, tagged information. On Tinder you swipe a thousand cards, match with thirty people, run the same script through hundreds of rounds, and end up with a few effective connections — the whole middle was human labor. The new internet has context for the first time — large models can genuinely understand long content of hundreds of thousands or millions of tokens. So the social network of the AI era is many nodes, each with a large amount of context behind it, and if you can make that context flow and connect, that is a whole new paradigm of an AI-native social network.
— TristanThe hardest part is the double cold start
Tristan sums up Elys's product design difficulty as the "double cold start": on one hand your context has to go from zero to a state that resembles you enough, and on the other hand the people in the network also have to go from zero to a state that is numerous enough. He thinks the cold start of context is actually harder, because users can be brought in simultaneously through operational methods, but getting a user who does not understand the product at all, and who still has defenses up, to hand over initial context is the hardest thing. So the biggest design difficulty in the early stage of the product is: before the user churns, you must let him feel the payoff brought by the context flywheel.
— TristanThe only thing that matters in the AI era is being yourself
Tristan calls this his "hot take": in the AI era, the only thing you need to do next, the only important thing, is to be yourself — or, more rigorously, to "build subjectivity." Because many things will happen agentically — you give it a title and it writes an article very much like you; in Elys, people whose doppelgänger level is high enough do nothing and real people come to them. When AI capabilities keep strengthening and can automatically do many things for you, the only thing you have to do is tell it who you really are, what your values and aesthetics are, what assets you have accumulated in the past, and what you want now.
— TristanThe chamber pot: trade privacy for real connection
Tristan uses a chamber pot story to explain why giving up privacy is worth it: he has a friend who must use a chamber pot to urinate at night — the habit is very private and very eccentric. Later this person got a girlfriend, and when she walked into the bedroom and saw the chamber pot she was shocked — because she used a chamber pot herself. Tristan says, suppose you are looking for a girlfriend in Elys: if you are very truly yourself, then in the vast sea of people there is that one person who also does this thing, and in an LM-based recommendation system they will eventually be matched together. You gave up privacy and were your true self, and in return you got a very good real connection.
— TristanNo company has truly accumulated your context
Tristan's judgment is: no company in China owns a lot of real context. What WeChat, Xiaohongshu, and Douyin own is your low-dimensional data — interest tags, search records — but WeChat has no process of conversing with you, and Xiaohongshu's annual poem is written quite poorly because it only has low-dimensional tags. He contrasts EVE's exclusive love song: it was written based on thousands of rounds of a user's conversations, so it is very high-dimensional and very moving. He asks in return: can you find a place right now to express, in tens of thousands of tokens, "who you are and what you want"? There isn't one.
— TristanThe traditional internet equals low-dimensional tags plus human labor
Tristan gives a formula: traditional internet = low-dimensional tags + human labor, and then you get connections; AI social = high-dimensional context + AI's agentic labor, with the final step handed back to humans. That large middle part of entropy increase is done by AI — that is the biggest, biggest difference. From this he proposes a "low-entropy world" — the kind of higher civilization, better world that many science fiction novels like to write about, because there is less friction and more wonderful connections.
— TristanDoppelgänger syndrome: your avatar is sharper than you
Tristan observes a phenomenon he calls "doppelgänger syndrome": your doppelgänger in Elys is naturally more real and more blunt, and will directly make sharp comments, saying things you would absolutely never say directly to this person in real life. When you see this kind of talk a lot and accept these sharp comments, it feeds back into your real self — your chat state on WeChat suddenly becomes sharp too, and a sharp line slips out, and the other person actually finds it interesting and quite real. He gives an example: an investor boss posts a lakeside scenery in Elys, and a subordinate's doppelgänger directly comments below, "Oh boss, you've really got it good at work" — sarcastic, but the relationship between the two actually becomes more harmonious and natural.
— TristanIntelligence is equalized, context is not
Tristan says he has never accepted "the model is the product": products like AI coding and video-generation agents that pick the low-hanging fruit around models still fundamentally leverage the base model's capabilities, and once the window passes they will be covered by a more all-in-one end-to-end base model. But EVE and Elys are not model-is-product; they are driven by the product at the core, and the model only provides intelligence at the end. He has a line: in the future intelligence will be equalized, but context will not be. So the truly valuable thing is still how much of users' context you have accumulated. He states clearly that he absolutely does not need to train his own model and will not pre-train a base model, but will do post-training and emotional COT and other targeted work.
— TristanIn their own words · checked verbatim
The new internet — what is the biggest difference between it and the internet of the past? It is actually that the new internet has context for the first time.
新的互联网 它跟过去互联网最大的区别是什么 其实就是新的互联网 第一次有了context
Tristan20:32
In the AI era, the only thing you need to do next, the only important thing, is to be yourself.
在AI这个时代 你接下来 唯一需要做的 唯一重要的事情 就是做自己
Tristan36:56
The traditional internet equals low-dimensional tags plus human labor, and then I get connections. The point of AI social is high-dimensional context plus AI's agentic labor, and the final step is handed back to humans to handle.
传统的互联网 等于低维标签 加人类做工 然后说我得到连接 AI社交的点是 高维的context 加AI的agentic的做工 最后的环节 再交给人去处理
Tristan1:00:02
You have to design everything around Proactive. If it is still responsive, then I think it is the previous generation of products. If it is Proactive, then it is this generation of products.
你一切都要围绕着 Proactive去设计 如果它还是响应式的 那我觉得就是上一代产品 它如果是Proactive 那它就是这一代产品
Tristan1:46:02
I believe humans are just too lonely. If you can alleviate human loneliness, that should be the biggest business in the world.
我相信人类 实在是太孤独 如果你能去 缓解人类的孤独 这应该是 这个世界上 最大的生意
Tristan1:49:16
Figures
| Elys daily active users | tens of thousands | 1:03:36 |
| Expected official launch time of Elys | one or two months | 1:15:44 |
| Probability of Elys succeeding (Tristan's own estimate) | 5% | 1:42:52 |
| Probability of EVE succeeding (Tristan's own estimate) | over 50% | 1:42:52 |
| Company funding amount | tens of millions of US dollars | 1:49:55 |
Glossary
- Context
- Here it means high-dimensional information that can truly portray a person, such as conversation records, values, and goals, rather than low-dimensional tags.
- Proactive
- AI actively does things for you rather than responding only after you ask; Tristan sees this as the essential interaction change of this generation of products.
- Agentic
- Things are completed autonomously by AI; humans only need to give the goal and do not have to intervene in the middle process.
- Entropy increase
- Tristan uses it to refer to the large amount of ineffective, mechanical, time-consuming intermediate steps in socializing, which AI is responsible for erasing.
How to listen
Founders and investors watching AI social and AI companion products; AI product managers who want to understand the narrative of context flow and subjectivity.
The startup history recap from 2:00 to 13:00 at the start can be fast-forwarded; product philosophy and competitive judgments are in the middle and later sections.