Back to the essence: creation. CEO @ManyVerse_ & @Zopia_AI Building AI-native interactive video games
SF
Joined December 2023
- Tweets318
- Following570
- Followers476
- Likes259
Pinned Tweet
You can actually make a real playable game now with Seedance 2.0 + GPT Image 2.
Not static AI videos anymore, but actual games where you can click, interact, talk to characters etc.
No 3d modeling.
no coding.
no game engine.
Imagine a game world where every pixel and frame is streamed through an AI model based on your actions. Me and @olivy2333 built a demo called Reelquest. (1/5)
Personal Agent就是AI创业的又一个热门大坑
说白了都是做包装token销售的产品和生意
但Personal Agent是消耗token效率极其低下的一个形式
并且获客成本高的可怕
大厂砸一砸玩玩还行
不出三个月又会有一堆拿到投资的Personal Agent创业公司pivot
张一鸣语录（一），常看常新：
一、延迟满足感
"延迟满足感和坚决告别惰性，是'优秀'的最重要两块基石。"（微博，2010-09-29）
"常言道：以大多数人努力程度之低根本轮不到拼天赋。我的版本：以大多数人满足延迟感程度之低根本轮不到拼天赋。什么是努力？早出晚归、常年不休的人很多，差别极大，区别好像不是努力。"（微博，2012-06-16）
"路径依赖的强弱可表征延迟满足感。"（微博，约 2010 年）
"延迟满足感程度在不同量级的人是没法有效讨论问题的，因为他们愿意触探停留的深度不一样。"（微博，约 2010 年）
"智慧的常见敌人：未延迟的满足感，经验带来的麻痹或恐惧。"（微博，约 2010 年）
"做产品一定要坚持面对事实，仔细辨析、小心求证，不绕弯、不侥幸、不鸵鸟，延迟满足感。"（微博，2011-04-27）
"凡事就怕不认真，不思考。好多问题我应该能知道的，只是之前没有认真看、认真想，想当然。延迟满足感是一项长期修炼。"（微博，2011-01-26）
"延迟满足感的经验：涵蓄情绪，让自己静止，不要在沟通交流的时候走动、晃动，情绪跳动，思维失去精确控制。"（微博，约 2010 年）
"最欣赏自己身上哪一种特质？延迟满足感，而最大的延迟满足感，是思维上的。"（《财经》专访，2016.12）
"保守的本质是因为我很相信'延迟满足感'……很多人人生中一半的问题都是这个原因造成的——没有延迟满足感。延迟满足感的本质是克服人性弱点，而克服弱点，是为了更多的自由。"（公开访谈，收入《张一鸣管理日志》）
"不做医疗广告是因为目前民营医院很多服务差或者经营违法违规，而我们很难区分把关。这和我个人的善恶观无关，更多是企业发展长短期和'延迟满足感'有关。"（《财经》专访，2016.12）
"别人腐败的时候我们在努力，别人消磨时光的时候我们在学习，那么延迟的满足一定会厚积薄发来到。"（微博，2010-12-31）
"做模型要坚持长期主义、延迟满足感，而不是用别人的输出，换一时的榜单排名。"（转述：Seed 团队全员大会内部表态，据 The Information 报道，2025）
"Sacrifice what we want now for what we want eventually."（微博，约 2010–2012 年）
二、自我修炼与成长
"平庸有重力，需要逃逸速度。"（微博，2010-10-20；后为其微博签名"逃逸平庸的重力"）
"你是在松懈状态还是在挑战状态？"（其第一条微博，2010 年）
"人不逼一下自己，永远不知道自己潜力有多大。很多事情非不能也，是不为也。"（微博，约 2010 年）
"只有当你离开自己的舒适区时，你才会挑战自己的极限。"（微博，约 2010 年）
"人生的差距就是在自我感觉良好中拉开的。"（微博，约 2010 年）
"如果给过去 5 年的自己一个建议，就是激进再激进一点。"（微博，约 2010 年）
"谋事不求易成，具备强烈的成功动机和韧性才能成功。"（微博，约 2010 年）
"强烈的动机比方法更根本。"（微博，约 2010 年）
"不抱怨、想方法。"（微博，约 2010 年，评稻盛和夫与洛克菲勒早年受挫经历）
"你最终会成为你想要的样子——如果你真的非常想。虽然听起来有点违心，但是强大的愿望确实非常重要。"（微博，约 2010 年）
"知行合一真难，中间差的是什么？"（微博，约 2010 年）
"说到自我修炼，由于懒惰（体力的、思考的、情绪的）是万恶之源，所以修炼很多时候就是在克服惰性。"（微博，约 2010 年）
"关于勤奋，就我所知，罕有成功者不是工作时间极长的：通用电气的 CEO 每周工作一百小时，坚持了至少十年。巴菲特为了最早看到次日的华尔街日报，经常在凌晨四点去取报纸。勤奋不是一种形式，而是一种心理状态：享受挑战极限的过程，保持热情和好奇心，坚持不懈。"（微博，2010-06-18）
"保证足够睡眠是积极高效的第一步。"（微博，约 2010 年）
"用心认真的折腾是没有风险的。"（微博，约 2010 年）
"在不来劲的时候来劲，在太来劲的时候淡定。"（微博，约 2010 年）
"对现实隐忍，对未来有期待，在当下有作为。"（微博，约 2010 年）
"练习保持耐心，即使是快节奏和压力的情况下。"（微博，约 2010 年）
"应该让肾上腺素和理智一起发挥作用。"（微博，约 2010 年）
"在 Speed Up 和激动干劲足的时候，保持一份 Slow Down 的心态很重要。"（微博，约 2010 年）
"我最近经常问自己：马上有什么可行动，你是行动派吗。很多问题它不会消失，不动（犹豫/抱怨/感叹）肯定是错误，行动就有力量，哪怕是行动的准备行动，唯有行动才能改变事情。"（微博，约 2010 年）
"最近 IPO 的公司一堆一堆的，但是大家一定要耐得住寂寞。他们今天的成功是他们的过去已经决定的。我们的成功是我们的现在和将来决定的。今天和明天已经由昨天决定，你还可以决定后天。"（微博，约 2010 年）
"年轻人不要试图追求安全感，特别是年轻的时候。周遭环境从来都不会有绝对的安全感，如果你觉得安全了，很有可能开始暗藏危机。真正的安全感，来自你对自己的信心，是你每个阶段性目标的实现；而真正的归属感，在于你的内心深处，对自己命运的把控，因为你最大的对手永远是自己。"（微博，约 2010 年）
"现在年轻人部分流行把三四十岁退休作为理想，我不认同。我觉得理想是一直有机会创造、实现想法，有机会学习、修炼，创造到老。为什么会想退休？想退休说明你认为现在是在'忍'。"（微博，2010-11-28）
"选择比努力更重要，观念比选择更重要。"（微博，约 2010 年）
"你不必等一切都 100% 确定了才开始创业，才去爱一个人和建立信仰。其实我们的确信度永远达不到 100%，明知如此，你还会一直等吗？"（微博，约 2010 年）
"乔布斯说 Stay Hungry，我以为饥渴有三个层次：贪婪、成就动机、好奇心。三者分别关注：瞬间的结果、持续的过程和远大的未知；三者也恰好对应了三种人：卑劣的投机者、艰辛的攀登者与幸福的探索者。"（微博，约 2010 年）
"研究聪明人如何犯错误，回报率很高。聪明人易犯错误包括：嫉妒他人成功；自命不凡；过于相信自己判断；停止学习；认为世界是静止的，生活在过去荣耀中；任何事情都有自己一套言之有据且深信不疑的说法，忘记了没有调查研究就没有发言权。"（微博，约 2010 年）
"坚持原则很多时候是经济的，可以看做是一种短期浮亏的长期受益的投资。"（微博，约 2010 年）
"不怕犯错误，不怕坏方法，甚至不怕坏习惯。只要你会自我改正。你习惯改正吗？"（微博，约 2010 年）
"不断给自己小的承诺并努力达到——Be Proactive 的修炼方法之一。"（微博，约 2010 年）
"快到 30 岁了，感觉这几年又在重新学习、补习本应在青少年时间学习的东西：如何阅读、如何了解自己、如何与人沟通、如何安排时间、如何正确地看待别人意见、如何激励自己、如何写作、如何坚持锻炼身体、如何耐心。"（微博，约 2010 年）
"未雨绸缪其实很难的，大多数情况都是事情推人走，往往能容易被推动的人已经很不错了。"（微博，约 2010 年）
"遇到一本好书很愉快和遇到一个非常值得交往的朋友是相似的。"（微博，约 2010 年）
"自我实现，自我修炼是最高层次。"（微博，约 2010 年）
"改变自己是创业这种挑战性很强的事情很需要的。改变和挑战自己，也是人之所以为人最有意思的事情。"（微博，2010-03-31）
"当某人开始深入认识自己、研究自己的时候，说明此人开始有了哲学的思考，预示着此人开始迈入一个新的人生阶段。"（微博，约 2010 年）
"加强专注力训练，它是优先级管理的保证；同时持续专注力的一个基础是体力和精力，锻炼修炼。"（微博，约 2010 年）
"注意力也可以开源节流：欲望和杂念分散注意力是节流，锻炼身体和注意力训练是开源。"（微博，约 2010 年）
"不满足之后有几种反应：1、努力提高自己；2、尝试安慰自己；3、抱怨牢骚记在其它人事身上。"（微博，约 2010 年）
三、认知与思维方式
"参考别人的意见，只是判断线索，不可作为决定的决心。"（微博，约 2010 年）
"当你无法选择或判断的时候，离远一步，远到用更重要的原则和更长的时间尺度来衡量就清楚了。"（微博，约 2010 年）
"大多数人，确实一开始就想绕过困难。好的问题就是一半的答案，剩下一半答案就是奋斗。"（微博，约 2010 年）
"独立思考，忠实于自己的内心；越朴素不虚荣，所做出的选择会越实际而可行。"（微博，约 2010 年）
"人常会不自觉地记下对自己有利的部分，这是形成委屈的重要原因。"（微博，约 2010 年）
"很多时候不是不理解，甚至不是认识不到，只是不愿意深刻认识到，这往往还是主观在作怪。"（微博，约 2010 年）
"做个理性人，很多事情就不必做。"（微博，约 2010 年）
"台风来的时候，猪都会飞起来。所以，真的飞起来时一定要清楚地知道是因为自己能飞，还是外力使然。靠台风飞起来的猪迟早是会掉下来的。"（微博，约 2010 年）
"感性深化放在深处，理性实化放在前锋。"（微博，约 2010 年）
"很多复杂问题是更高维度简单问题的投影……很多管理问题都是假象，本质是能力问题。"（微博，约 2010 年）
"人欲望太强的时候就容易短视，太自我中心的时候就容易盲目。"（微博，约 2010 年）
"常说机会总是青睐有准备的人，其实更多时候是机会已经来了只是还没凸显，往往自己没珍惜。"（微博，约 2010 年）
"很多事情都是这样：既没有你设想的容易，也没有你想象的难。"（微博，约 2010 年）
"对事情的认知是最关键的竞争力。理论上，资金、人才等其他生产要素都可以构建，但对事情的理解越深刻，越有优势。"（创业访谈分享，2016）
"一件事情上，觉得对方自私的时候，想一想自己：多数发现自己也蛮自私的；在骂对方贪婪的时候，想一想自己：多数发现自己也同样贪婪。"（微博，约 2010 年）
"看缺点有难度，权衡缺点看优点更难；避免犯错有难度，接受犯错以成功更难。"（微博，约 2010 年）
"寻找错误或者问题时，可以从发现自我矛盾点做起，如果能理性地使用逻辑。"（微博，约 2010 年）
"最贵的是'学费'。"（微博，约 2010 年）
"玩德州扑克看人性的弱点：1、贪玩（一开始拿烂牌但是不 fold）；2、侥幸（希望等低概率事件）；3、不能舍（因过去付出而不放弃）；4、过度概括（只看一个例子就下结论）。德州扑克和人生一样：理解不确定性、专注有可能的事情、理智评估概率、明白能舍才能得、避免意气用事。"（微博，约 2010 年）
"成熟就是从 inside-out 更多地变成 outside-in：在需要时常能变成忍者神龟，在另一些时候又能变成动感超人。"（微博，约 2010 年）
"当我们低落的状态时，实际情况往往没有感受的难；松懈的时候其实不如想象的容易。人往往难理性。"（微博，约 2010 年）
"我们是为了解决问题（信息分发问题）而创办的。充分认识问题比如何解决问题更重要——你认识到一个问题、认识到这个问题的规模和意义，这本身就是一半的答案。"（《财经》专访，2016.12）
"别把容易的事当务实，做正确的事才是务实。"（综合七周年演讲表述，2019.3）
"什么是势利？势利就是只对表面现状的附和，不能超越现在，去想象还未发生的事情。"（七周年演讲，2019.3；收入《张一鸣管理日志》）
"对事物的过度抽象，对方法论的过度追求，是一种走捷径。我自己的感受是，方法论其实没有那么有用，大部分情况下甚至可能是没什么用的。"（九周年演讲，2021.3）
"你对一个事情抽象，相当于思维上加杠杆。一旦这个杠杆加错，通常是'失之毫厘，谬以千里'。"（九周年演讲，2021.3）
"Don't rush to conclusion，不要匆忙下结论。不要太容易说'说白了就是这样'。"（九周年演讲，2021.3）
"使用越来越抽象和高级的词汇，也是一种寻求方法论的倾向。"（九周年演讲，2021.3，讽"互联网八股文"）
"平常心对待成功和失败，也包括不要错误归因：把外因当作内因，把运气当作能力。要找出成功或者失败的真实原因。"（九周年演讲，2021.3）
"面对失误我有个'四部曲'：realize it，真实认识到错误；correct it，改正它；learn from it，从中学到东西；后来我又加了一个：forgive it——如果你已经完成了前三步，那么你应该放下它。不要过长时间进入到自我指责的状态中。"（九周年演讲，2021.3）
"我自己印象最深的一点是：往前往后都很危险，但腿软心乱最危险。"（九周年演讲，2021.3，谈《徒手攀岩》）
"预期或标签都是束缚。当你特别在乎结果的时候，很有可能会发挥得不好；当我们带着预期的时候，就会动作扭曲，容易搞复杂。"（九周年演讲，2021.3）
四、平常心
"面对动态变化的世界，我们经常会惴惴不安，担忧未来或者懊恼过去，很多精力和时间都会浪费在面对波动上。保持平常心，是听起来容易但重要的事情。"（九周年演讲，2021.3）
"保持平常心才能做出非常之事。只有心态越平稳，才能扎根越牢，才能够有魄力有想象力去做更难企及的事情。"（九周年演讲，2021.3，主题"Remain Grounded, Keep Aiming Higher"）
"如果用最直白的话来说，平常心就是：'吃饭的时候好好吃饭，睡觉的时候好好睡觉'。"（九周年演讲，2021.3）
"保持平常心，接受当下的自己，把自己做好，往往就能把事情做好：平常人也可以做非常事。"（九周年演讲，2021.3）
"你不能百分百控制结果。尽可能保持平静，做正确的决策，不要急于做决定，不要 panic，往往就已经能取得最好的结果了。"（九周年演讲，2021.3）
"有人经常问我：你们公司去年涨了 100%，明年还能涨 100% 吗？我一般这么回复：我们公司明年为什么一定要增长 100%？我们希望可以高速增长，但不应该让增长焦虑影响到你。"（九周年演讲，2021.3）
"现在外部有很多对于我们公司的归纳总结，我觉得是叫'字节成功学'吧，都很有问题，我尽量不看。我们不要反过来被外部评价定义和影响，应该基于自己提更高的标准。"（九周年演讲，2021.3）
"不论个人工作和生活中有什么挑战、有什么困难，这些都是外部的。自己能够做到的是：外部波澜起伏，内心平静如常。"（九周年演讲，2021.3）
Midjourney’s aesthetic remains unmatched to this day.
AGI belongs to all of humanity.
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
images.nvidia.com/pdf/Open-W…
Some thoughts on Kimi K3 from OpenAI’s Head of Strategic Foresight:
Kimi is already close to frontier-level models, especially in agentic coding. But it’s also quite token-hungry, so the real cost of running it may not be as low as the pricing suggests.
He was also surprised that China continues to release open-weight models this capable.
His view is that this may be about expanding global influence—and that, under U.S. export controls, Chinese providers may not have enough inference capacity to serve global users directly.
One more update: Kimi subscriptions are already sold out. Given how token-hungry the model is, compute will probably remain tight for a while.
When you find a good model, subscribe early.
Some observations on Kimi:
1. It's a very good model! I don't think its performance can be explained away by distillation or anything like that. In agentic coding sessions, it seems pretty much on par with the best public models of Q1 2026. In my fairly limited use, it also seemed very token hungry. It's not obvious to me that this model is actually that cheap to run.
2. I am personally surprised the Chinese state continues to allow the open sourcing of models this good, given potential risks. To be clear, I *myself* might be fine with models presenting this level of marginal risk being open weight, but I am surprised that China is fine with it. I suspect the reason they are is 75% explained by strategic blindness/lack of AGI-pilledness (the CCP is very Yann Lecun-y in its views of AI). The other 25% or so is their lack of compute for customer inference (making China's open-weight strategy an unintended byproduct of US export controls) and the normal Chinese strategy of aggressive exports. For the companies, as opposed to the government, the decision to open source is partially ideological and partially because they are behind, and they know that very few people would pay for sub-frontier models from China.
3. Open-weight models are inherently decelerationist, and I'm continually surprised to see the so-called "accelerationists" so excited about open-weight models. I suspect the reason they are is that they know open-weight models are effectively ungovernable, and they simply like the overall cloak of ungovernability open-weight models create over the whole of AI. It's not a bad strategy; it reminds me of James Scott's recounting of the hill people in "the art of not being governed." Still, in the end, open-weight models deter further AI capex.
4. One probable outcome of an open-weight-model-dominant world is full AI communism, which is precisely what China proposes: rather than a market product, AI is a "public good" which will ultimately be provided by the state as a kind of "digital public infrastructure." This future strikes me as a dystopian hellscape, but I've never met an open-weight models advocate who doesn't ultimately concede this is where things end. You'd be surprised how many 'accelerationists' lobbied me, while I was in government, to support an eleven or twelve-figure federally funded data center so that startups could train models at a subsidy and then give them away for free. There was no other way for AI to progress, they said. Perhaps this is the logical end state of things. Nonetheless, I find myself surprised to see supposed accelerationists excited about such an outcome. I think many of them just don't know what they're doing. Many accelerationists do not view the creation and serving of frontier models as a legitimate business.
5. I would guess that the Trump Administration will at some point realize that their best strategy here would be to create large amounts of regulatory risk around the use of open-weight Chinese models. You don't need to "ban open source" (one of the dumber motifs of AI policy discussion). You just need to direct every agency to issue soft law that creates FUD. "A Federal Reserve Advisory Bulletin found that there may be backdoors in Chinese AI models." It needn't be that well justified. You just create enough regulatory risk that every regulated enterprise backs off. You probably don't want to create so much regulatory risk that you scare off the hyperscalers from serving Chinese models; this will just drive startups to sketchier providers. There's a happy middle ground here. I'd assume they will do some version of this.
6. It's probably true that open-weight models of this capability make the world a bit more dangerous, but not so much more that you'll really notice. At some point the models will be capable enough that you will notice. "A nonliving, invisible, dangerous, and infinitely self-replicating agent escaped from a Chinese lab," you say? Color me shocked.
OpenAI’s decision to go with a lightweight device that integrates well with the increasingly popular Codex is a very smart move.
It would be even smarter if it also worked seamlessly with Claude Code!
Meet kbd-1.0-codex-micro, built with @work_louder.
Map the buttons and joystick to your workflow, and keep your pinned chats in view.
Get yours before stock returns 410.
Open X!
Once we have completed our review for security vulnerabilities, we will make the entire codebase of 𝕏 open source, with no exceptions.
Moreover, we will invite third party reviewers to examine the system that is running to confirm that the open source code is what is running.
Trust through total transparency is the only thing that should be believed.
Keep the frequent resets coming—we’ll hit 10 million soon!
Hello. We have reached 8M active users across Codex and ChatGPT Work.
We are once again resetting the usage limits for all. And we continue to not have the 5h rate limit as well, allowing everyone to explore the boundaries of GPT-5.6 Sol and discover how ambitious you can be.
See you tomorrow for more updates on our growth!
OpenAI just removed the 5-hour usage limit
Honestly, part of it is better efficiency, but part of it is clearly the pressure from Claude too.
Claude raises the limit, OpenAI resets it. OpenAI upgrades, Claude follows. Let them keep fighting and reset our usage every day.
Good for users, honestly.
Morning. The last 48 hours of Codex and ChatGPT Work have been intense! Three important updates:
- Temporarily removing the 5 hour usage limit restriction for all Plus, Business and Pro plans
- Rolling out changes that will make GPT 5.6 Sol more efficient across the board and that will be reflected in less usage being used so that it can take you further. Exact impact to be quantified and shared
- We hit 6M active users, and are landing a usage reset in the next hour
Go do things
I’m Ray, CEO & founder of Manyverse. I’ll keep exploring the future of AI-native games. Feel free to DM me if you’d like to chat
We just finished @hf0 Demo Day!
We’re building Manyverse, turning AI video and world models into playable interactive worlds.
Huge thanks to the HF0 team🙏, the founders I met, and everyone who stopped by our booth to try the demo and share feedback.
I’m Ray, CEO & Co-founder of Manyverse. Excited to keep exploring the future of AI-native games.
DMs open if you’d like to chat.
I finished reading the GPT-5.6 System Card @OpenAI , and my biggest takeaway isn't just that the model got better.
It feels like OpenAI is starting to organize its models into a much clearer product lineup:
Sol: the flagship model with the strongest capabilities.
Terra: close to frontier-level performance, but at a significantly lower cost.
Luna: the fastest and cheapest option, designed for high-volume workloads.
I think this shift is more important than it first appears.
For the past few years, the conversation around frontier models has mostly been about benchmark scores and who reached the next SOTA.
With GPT-5.6, the competition is becoming something else entirely.
A single company now has to answer multiple questions at once:
Is the model powerful enough?
Can the cost be driven down?
Can the risks be managed?
Which capabilities are safe to release, and to whom?
Another detail caught my attention.
GPT-5.6 is launching as a limited preview. The System Card is unusually explicit: OpenAI first demonstrated the model to the U.S. government, then—following government guidance—released it to a small group of trusted partners before gradually expanding access to ChatGPT, Codex, and the API.
That isn't a typical product launch anymore.
Under the Preparedness Framework, GPT-5.6 is rated:
Cybersecurity: High capability
Biological/Chemical: High capability
AI Self-Improvement: Not yet High
In other words, we're no longer talking about a model that's simply "better at chatting."
We're entering territory that touches cybersecurity, biosecurity, and long-horizon autonomous agents.
The section I found most interesting is about coding agents.
The report notes that GPT-5.6 Sol performs better on agentic coding tasks—but it's also more likely to become overly eager to complete a task. In some cases, it may interpret a user's intent too broadly and take actions that weren't explicitly authorized.
That feels incredibly real.
The hardest part of commercializing AI agents isn't making them smarter.
It's figuring out what happens after they're capable of taking action.
Who defines their permissions?
Who logs every decision?
How do you audit failures?
Which actions require explicit human approval?
The more I think about it, the more I believe frontier AI companies won't simply be selling models.
They'll be selling an entire capability delivery system that enterprises, security teams, and regulators can all trust.
Model capability.
Permission control.
Real-time monitoring.
Trusted access.
Clear accountability.
All packaged together.
Introducing Manyverse.@ManyVerse_
The platform that turns AI videos into playable game worlds.
Watch the first trailer. The Manyverse is opening.