135. 和自然选择创始人Tristan聊，Elys、赛博分身、灵魂、Context的获取与流动和AI社交网络

Summary

This 张小珺Jùn｜商业访谈录 episode interviews Tristan, founder of Natural Selection / 自然选择, about EVE, Elys, and the product thesis behind AI Social Networks. Elys is framed as an AI-native social network where Cyber Avatars carry high-dimensional user context, pre-interact with other avatars, and hand valuable connections back to real people. The core strategic claim is that intelligence may become more equalized, but Context Engineering, Context Flywheel, and Subjectivity As AI Asset will remain scarce and product-defining.

Key Claims

• Tristan links his product history across audio, stranger social, romance games, EVE, and Elys through a consistent concern with loneliness, emotional value, and better human connection.
• His early audio-platform lesson is that pre-AI recommendation relied on low-dimensional labels such as titles and tags; long audio remained difficult to understand before a user opened it.
• EVE grows from the Project Her idea: AI could make scripted characters feel more alive by combining memory, persona, interaction, and emotional response.
• Elys moves from single-user AI products toward a network form where context-processing happens between nodes, not only inside one user’s private assistant.
• The episode contrasts traditional social products as low-dimensional labels plus human screening with AI Social Networks as high-dimensional context plus agentic pre-work plus final human handoff.
• Cyber Avatars are meant to represent the user’s current thoughts, values, taste, tone, goals, public work, and remembered interactions, then use that model to comment, filter, match, and invite.
• Context Flywheel is the main product loop: user behavior improves the avatar, the avatar returns better connections, and visible utility makes the user more willing to provide more context.
• The source treats privacy as a live product tradeoff: users will not hand over sensitive context unless they see enough concrete value in return.
• Elys is not presented only as dating; Tristan describes it as a connection network that could include friendship, professional networking, resonance discovery, and real-world meetups.
• The Urumqi ski-group example is used to show that AI social value can be measured through real-world connection, not only in-app engagement.
• The episode’s “avatar syndrome” claim suggests that watching a more direct avatar may change how users express themselves in real relationships.
• Subjectivity As AI Asset is the source’s strongest AI-era agency claim: a person must know who they are, what they want, what they value, and what aesthetic or past work should guide agents.
• Tristan argues that WeChat, Xiaohongshu, and Doubao may have abundant behavior data, but not necessarily enough high-dimensional subjectivity context to represent a person.
• He expects ChatGPT and other large assistants to become serious competitors because they already accumulate user context, but he argues social-network protocols, rules, and user graphs still require product work.
• Natural Selection / 自然选择 does not plan to pretrain foundation models; its product bet is post-tuning, smaller adjustments, emotional reasoning patterns, and context-driven interaction design.

Key Quotes

“智能会逐渐平权，但 Context 不平权” — Tristan on product defensibility.

“AI 时代唯一重要的事情就是做自己” — Tristan on subjectivity.

“EVE 更像一个人，Elys 更像一个场” — Tristan on the two products.

Connections

• Tristan, Natural Selection / 自然选择, EVE, and Elys — founder, company, companion product, and social product discussed in the episode.
• AI Social Networks, Cyber Avatars, Context Flywheel, and Subjectivity As AI Asset — main new concepts introduced by the source.
• Context Engineering, Persistent Agent Memory, and Proactive Agents — existing agent concepts extended from work/productivity into social-network matching.
• AI Native Product Design, AI Friend Products, and Human-Agent Collaboration — product-design and companion branches deepened by the EVE-to-Elys transition.
• Human Agency Under AI, AI Communication Ability, and Human Judgment Under AI — agency and judgment themes behind the claim that users must define themselves before delegating to agents.
• WeChat, Xiaohongshu, ChatGPT, Doubao, and Open Cloud — comparator platforms and products used to frame assistant context, proactive behavior, and social-network competition.

Contradictions

• No direct contradiction with prior wiki content. The source reinforces existing Context Engineering, Proactive Agents, Persistent Agent Memory, and AI Native Product Design themes while moving them from productivity and work agents into social connection.
• The source qualifies earlier broad AI-companion optimism by separating EVE’s virtual companionship from Elys’s real-person connection loop; AI is the intermediate method rather than the final relationship object.
• The source also qualifies platform-context optimism around WeChat and ChatGPT: accumulated data matters, but Tristan argues that useful AI social networking requires explicit subjectivity context, user trust, and a network protocol for avatar-mediated connection.