与许多中国风险投资人不同，我和我的合伙人克里斯不会投资于竞争对手。投资竞争对手是风险投资人能做的最不忠诚和不专注的事情。它显示出完全缺乏信念，并且是懦弱的行为。
一旦我们选择了创始团队，我们就会完全承诺给他们，并确保他们有赢得的最佳机会。对我们来说，忠诚和完全专注是世界一流伙伴关系的基础。即使我们最终错了，错失了赚钱的机会。我们不想投资于Robinhood、Manus、Plaud、Qbit或Wanaka的竞争对手。当我们给你开支票时，那就是我们的承诺。坚持到底。(Grok translation)

Unlike a lot of VCs in China, my partner Chris and I do not invest in competitors. Investing in competitors is the most disloyal and unfocused thing a VC can do. It shows a total lack of conviction and is spineless behavior.

Once we pick a founding team, we commit to them entirely and make sure they have the best chance of winning. To us, loyalty and total focus is the foundation of a world-class partnership. Even if we end up being wrong and leave money on the table. We don’t want to invest in Robinhood, Manus, Plaud, Qbit, or Wanaka competitors. When we write you a check that’s our promise. In it until the end.

Harmonic 是我在美国做出的主要押注之一。形式化验证数学与智能体求解将成为前沿领域，为极其复杂的系统带来信任与可靠性，并解锁海量科学发展。我的前老板 Vlad Tenev，以及 Helm AI 的前投资组合创始人 Tudor Achim，已吸引了顶尖人才和投资者共同攻克这一问题。非常期待看到他们的进展。

Harmonic is one of the primary bets I’ve made in the US. Formal verified math and agentic solving is going to be the frontier that delivers trust and reliability to very complex systems and unlocks an incredible amount of scientific development. My former boss Vlad Tenev and former portfolio founder of Helm AI, Tudor Achim have attracted fantastic talent and investors to work on this problem. Excited to see their progress.

我看到太多团队在构建智能体架构，这种架构让模型把大部分时间都花在搜索上，而不是真正去做事。
解决这个问题的方法有很多，但如果你没有主动去思考它，你甚至不会意识到它的存在

I’m seeing too many teams building agentic architecture that forces the model to spend most of its time searching, and not doing.

There are a lot of ways to solve this problem, but if you aren’t actively thinking about it, you won’t even realize it’s there.

科技行业里一些最优秀的人才，同时也是最难管理的人，就是那些把一切都当成自己事情的人。他们情绪化、容易争论、对一切深度投入，并且对所有问题都深入思考。他们凡事都往心里去，是因为他们极度珍惜自己的时间，想要用每一个清醒的时刻去赢得胜利。任何挡在他们面前的人，都会立刻感受到一种迫切的抉择：要么加入，要么错过。

大多数公司只能拥有少数这样的人，但他们却能给整个组织带来巨大压力，推动大家达到更高的绩效水平。即使他们制造了一些冲突，整体权衡下来仍是净正收益。领导层面临的挑战是：确保公司真正有一个清晰的取胜计划，并让这些人直接去推动实现——绕过管理层，直接对接高管。

如果你就是这类人，当经理或同事对你说“你把一切都太当回事了”的时候，你应该把这种特质发挥到极致，一点都不要改变。直接去找CEO，把你认为公司需要什么才能赢的看法原原本本告诉他们，忽略所有那些管理层的应声虫——他们都在用自己的时间玩一个远没有那么有趣的游戏。

Some of the best people in tech, but also the hardest ones to manage, are the ones who take everything personally. They are emotional, get in arguments, are deeply invested, and think deeply about all of their problems. They take things personally because they value their time highly and want to use every waking hour to win. Anyone standing in their way feels an immediate choice to join or miss out.

Most companies only manage to have a few of them but they generate massive pressure on the entire organization to perform at a higher level . Even if they create conflict, the trade off is a net positive. The challenge for leadership is making sure that the company actually has a plan and knows how they will win, and deploying these people to make it happen; without a management layer, and direct executive access.

If you are one of these people, and managers or peers say “you take everything too personally” you should take that strength to the end of the world and not change a thing. Go talk to the CEO and tell them exactly what you think the company needs to win and ignore all the management drones - they are all playing a far less interesting game with their time.

Wishing Hamilton a Strong Race!!

This is maybe a stupid question but given that my wechat moments and jike are full of openclaw content, i’ll ask:

are most people in china in these massive meetups using claude or codex with opeclaw?

i’m wondering because the open source models are pretty horrible at tool use so i can see the potential for disappointment if they do not.

Vibe coding 赋予我们构建私密、高质量且高度定制化产品的能力。

过去几周，我一直在开发一个原生 GPU 的 Wisprflow/Typeless 替代方案——它免费、无云、速度极快，名叫 TurboSpeak。

Apple Silicon 是当今许多人都能接触到、却最被低估的技术之一。我认为，越多的人开始思考 GPU 编程，就会有越多的人意识到软件领域还有多么巨大的潜力尚未被挖掘。

TurboSpeak 结合了大型 Whisper 模型和经过微调的 Qwen 模型，不仅能转录音频，还能清理那些杂乱无章的非结构化想法。我还在修复很多 bug，但希望大家试用后能喜欢。它大约占用 6GB 内存，但速度很快，质量也相当高。

作为一个社区，我希望我们能一起构建和分享这类免费工具，并意识到大多数软件公司未来将变得毫无价值——现在是时候用不同的方式思考如何投入创业时间，以及哪些竞争策略在未来仍会具有生命力。

我每天仍在运行大量 Claude 会话来调优模型和修复 bug，所以大家可以期待 TurboSpeak 在可预见的未来会持续改进。

Vibe coding gives us the power to build products that are private, high quality, and customized.

Over the last few weeks I’ve been spending time building a GPU native Wisprflow/Typeless alternative that is free, no-cloud, and fast - TurboSpeak.

Apple silicon is one of the most underused technologies many have access to today, and I think the more people that think about GPU programming, the more people will realize how much potential is still available in software.

TurboSpeak uses a Large Whisper Model combined with a tuned Qwen model, to not only transcribe, but clean up unstructured thoughts. I’m still working on many bugs but hope you enjoy trying it. It takes about 6GB of memory but the results are fast and pretty high quality.

As a community, I hope we can build and share free tools like this, and realize that most software companies will be worthless - its time to think differently about how to spend time on startups, and what competitive approaches will still be viable in the future.

I’m still running many claude sessions a day to tune the model and fix bugs so you can expect this software to improve for the foreseeable future.

Anthropic outages are painfully boring periods of my life now.

今年，超越 LLM 的知识储备并不再依赖它，变得尤为重要。我开始相信，仅靠简单封装 LLM 的初创公司，将无法在任何垂直领域存活下来。LLM 应该仅用于工具调用和推理。知识层或许才是持久的护城河。AI 的“生成”部分，目前且很可能将继续由模型公司所掌控。将它作为核心业务来依赖，风险极高。你越能把工具和推理当作发现引擎，我相信你就越能构建出真正专有的东西。

This year it’s really important to move beyond LLM knowledge and not rely on it. I’m beginning to believe that startups wrapping LLMs alone will not survive in any vertical. LLMs should be used for tool use and reasoning only. Knowledge layers may be the durable moat. The “generative” part of AI is, and will likely remain, owned by the model companies. Relying on it as a core business seems very high risk. The more you can use tools and reasoning as a discovery engine, I believe the more you will build something proprietary.

问： Patrick，你在伦敦、不在中国看deals的时候，都干嘛啊？
答： 你不会懂的。

恭喜 Wanaka 团队！！
解锁了整个创意游戏和社交互动的类别。
我们才刚刚开始！

Congratulations to the Wanaka team!! Unlocking an entire category of creative play and social interactions. We are at the very beginning!