M1kes1

[GREG ISENBERG @gregisenberg]

I just got back from SF and I FEEL INSPIRED.

I spent 5 days with frontier AI model teams, AI startup founders, and 3 billionaires.

My takeaways:

1. I had lunch with 3 billionaires. All of them are buying SaaS companies and rebuilding them agent-first. They were deeply inspired by Bending Spoons and Ryan Cohen's eBay deal. Buy the company, cut the headcount, rebuild the tech, add agents, add features, make more valuable experience, raise prices.

2. The frontier model companies are hungry for usage data from the field. They can see API calls and token counts. They can't see the actual workflows. If you're deep in a niche using these models in ways the model companies haven't seen, that understanding is incredibly valuable. Usage intelligence is the new alpha.

3. Consumer AI is massively underbuilt. Every billboard in SF is either B2B inference infrastructure or vertical agent companies. The entire city is optimized for enterprise. Meanwhile you have companies like Cal AI doing $50M ARR in 18 months as a consumer app. I met with a cool few teams doing consumer AI (@paulscherer / @ekuyda)

4. MCP came up in literally every conversation. The companies exposing their product as MCP endpoints are getting pulled into deals they never pitched for. The ones that aren't are becoming invisible to agents. This is the new SEO. If agents can't find you, you don't exist. Building products for agents is the new zeitgeist in general.

5. Not uncommon for hot seed rounds to be $25-50 million valuations. I saw a Series A at $450 million

6. If I had a dollar every time someone mentioned "forward-deployed engineer" this trip I could have funded a seed round. It's the hottest role in SF right now. The person who sits between the agent and the customer, making sure everything actually works.

7. The mood around open source shifted. A year ago it felt like open source was chasing the frontier models. Now founders are telling me Gemma and DeepSeek are good enough for 80% of what they need at a fraction of the cost. The "which model do you use" conversation is being replaced by "which model for which task." Model loyalty kinda feels dead.

8. Voice agents came up more than I expected. Multiple founders told me voice is the interface for the next billion users. The billion people who will never type a prompt will absolutely talk to one.

9. The Obsidian community in SF is weirdly intense. Multiple founders showed me their vaults unprompted. Like showing someone your home gym. It's a flex now. The quality of your knowledge base (second brain?) is becoming a status symbol among builders.

10. Maybe it was just the people I met but the age of the founders is shifting. I met more founders over 40 this trip than any trip before and more founders under age 21 than ever before. Founders getting older and younger at the same time.

11. I spoke to a lot of fast-growing startups, VCs and frontier models who are hiring content creators right now.

12. The restaurant scene in SF is actually better than it's been in years. Founders are going out more. Alcohol is out, not surprisingly.

13. SF doesn't feel like the only place anymore. We all have access to the same frontier models. We all read the same X feed. A founder in NYC or Lagos is calling the same APIs as a founder in SoMa. So in the past it felt like SF was always lightyears ahead, doesn't feel that way anymore. It's okay not to live in SF and have BIG DREAMS.

14. The coworking spaces in SF are half empty but the coffee shops are packed. People want to be around people. I had a few startup ideas here....

15. Walking around the Mission I noticed something: the street-level businesses, the taquerias, the barbershops, the laundromats, none of them use any AI at all.

16. I heard the phrase "agent debt" for the first time. Like technical debt but for agents. When you hack together an agent workflow fast and never clean it up, the system prompts conflict, the memory gets polluted, the tools overlap. 6 months later the agent is doing weird things and nobody knows why lol.

17. Met a few people who carry two phones now. One for personal. One that's basically an agent terminal running Telegram or iMessage connections to their agent fleet.

It's always amazing to get that dose of inspiration in SF. I FEEL INSPIRED.

But I'm so happy to be back home, locked in and building.

We're 12-18 months into a shift that will take 15 years to play out. The urgency in every conversation was real.

What an incredible time to be building.

[M1kes1 @Mixessi]

@SuperFPF @leifuchen @chunxiangai 老哥 想求一份Poly心经 感谢🌹

[M1kes1 @Mixessi]

@yaosiscom 老哥 求个存档🌹

[Andrej Karpathy @karpathy]

"Finding the Best Sleep Tracker"
Results of an experiment where I wore 4 sleep trackers every night for 2 months. TLDR Whoop >= Oura > 8Sleep >> Apple Watch + AutoSleep. Link simply right here instead of in a reply because ¯\(ツ)/¯
https://t.co/3kGg4mKAaL

[M1kes1 @Mixessi]

@MEJ50749 +1

[M1kes1 @Mixessi]

@BTCBruce1 抽

[李自然 Nate Lee @nateleex]

📢 发点干货

Claude 3.5 进阶秘籍 | 5大绝招助你成为AI达人👩🎓

1. Claude官方宝库：针对商业和个人任务优化的提示词，助你事半功倍。💼✨

https://t.co/RDnpUJuLxE

[indigo @indigox]

Perplexity Page - 这算对互联网少量人类原创内容的进一步污染？还是 AI 全面接管内容生产的前夜😐

[宝玉 @dotey]

推荐阅读：《我们在使用大语言模型 (LLMs) 构建产品一年中的经验总结 (第一部分) [译]》

文章主要分享了大语言模型核心组件的最佳实践，包括提示词设计、对 LLM 输出结果的评估，什么时候该用 RAG 还是微调等等。

一、提示词设计

1. n-shot 多样例
在提示词中加入 5 个以上不同类型的示例，覆盖各种可能情况

2. CoT 思维链
让 LLM 在输出结果之前解释其思维过程。简单的加入“让我们一步步思考”还不够好，最好是明确步骤

3. 输入输出的结构化
在使用结构化输入时，请注意每种大语言模型都有自己的偏好。Claude 偏好 xml，而 GPT 则偏好 Markdown 和 JSON。

4. 编写小而精的提示词，专注做好一件事
不要试图在一段复杂的中解决所有问题，尝试拆分成多个小而专注的提示词

5. 精心设计上下文信息
上下文信息并非越多越好，而是要提供关键的有效信息，而不是冗余、自相矛盾和结构混乱的信息。

二、信息检索/检索增强生成（RAG）

1. RAG 输出的质量取决于所检索文档的质量
所检索的信息相关性越高、信息密度越高、文档细节越丰富，检索效果越好。

2. 关键字检索同样很重要
现在大部分 RAG 教程演示都是基于嵌入（Embedding），但很多应用场景关键字检索效果可能更好，最好是将嵌入和关键字结合使用

3. 优先使用 RAG 而不是微调
大部分时候是不需要微调的，RAG 便宜、容易实施，并且 RAG 提供了更精细的文档检索控制。

4. 即使是模型的上下文越来越长，RAG也不会过时
需要 RAG 来筛选输入模型的信息来提升上下文信息质量得到更好的结果，另外长上下文成本比较高。

三、从提示工程到工作流

1. 逐步多轮的工作流能显著提升效果
将一大段复杂提示词分解为若干段短小提示词可以取得更好的效果，而多个短小提示词需要有工作流来支撑

2. 优先采用确定性工作流
工作流越确定越容易管理和控制

3. 合理使用温度（temperature）参数
如果希望 LLM 生成结果更加多样化，那么调高温度参数，如果希望更加确定性，则调低。

4. 不要忘记缓存
对于 LLM 生成的结果，可以缓存起来，下次有相同的请求可以重用节约成本。

5. 微调
有一些任务，即使是最巧妙设计的提示也无法胜任。例如，即使经过大量提示工程，我们的系统可能仍然无法返回可靠的高质量输出。如果是这样，那么可能有必要为特定任务微调模型。

如果决定微调，为了减少收集人工标注数据的成本，可以 生成并在合成数据上进行微调，或 在开源数据上引导。

四、评估与监控

1. 基于实际输入输出样本编写单元测试
收集实际应用的输入输出样本，用这些数据编写测试。由于 LLM 输出存在一定的开放性，可以用一些特殊的验证方式，比如字数、句子数在范围内，比如格式是否符合要求等。

2. 用 LLM 评估 LLM 生成结果
可行，但不是万能的。

3. “实习生测试”
把 LLM 当成大学实习生，如果这个任务实习生能完成，那么模型是不是也可以？如果实习生都不能完成，那么是不是应该简化拆分？

4. 过分强调某些评估指标可能损害整体性能
“当一个衡量标准变成目标时，它就不再是一个好的衡量标准。”

其他还有几个评估与监控方法，不再总结，有兴趣请阅读原文。

原文：https://t.co/PTKPZzigdt
译文：https://t.co/847rAw26CI

[The Technology Brother @thetechbrother]

Someone should make this app for vision pro

[M1kes1 @Mixessi]

lol

[TechCrunch @TechCrunch]

In case you missed today's #GoogleIO keynote presentation, we summed it up for you

[Lyson Ober @lyson_ober]

🔥 Midjourney cref 海马体写真教程来啦！10 分钟不到出写真～视频在下面 （记得mark 看文字笔记）👇 超低学习成本，你只要会用任意手机修图软件...就能解决 Midjourney Cref 角色一致性直出之后的「形似而神不似」的问题。

📝 简而言之，文字笔记：

1️⃣ 得益于 --cref [image url]和 --cw [0~100]，midjourney 能够迅速达成角色一致性的目的而无需 Lora。

其中，cref 代表 character reference（角色参考）；cw 则代表 character weight（参考权重），默认为 100，会参考角色的人脸和服装特点。如果你只想关注人脸，请调低到 0。

2️⃣ 你不需要对角色原图使用 /describe 来贴近目标生成角色的形象，因为 --cref 会帮你完成这件事情。

👉 于是，风格复制和服装的替换，则应当用提示词（prompt）和 --sref [image url] 来解决。

顾名思义，--sref 代表 style reference，权重参数为 --sw [0~1000]，默认值 100，越高越接近参考图像的风格

举个例子，现在我们要海马体这张图像，我们先用 /describe 指令反推海马体的提示词，再配合 --sref [海马体图像的 url]，就可以轻松复制图像风格。

在回车之前，检查你的提示词，不要出现和 sref 风格差异过大的 prompt；也可以增加一些你之前学过的有用的 prompt。

3️⃣ Upscale 最满意的那一张。接下来解决大家最关心的「形似而神不似」的问题。其实非常简单，这个问题其实是：MJ 直出和原人物在眼睛、眉毛、鼻子等其他地方的大小、间距有一些细微的差距。

如果你不控制变量一个个调整，就会觉得「怎么眼睛嘴巴合在一起就不像我了？」

此时，用醒图（英文版叫做 hypic）或者美图秀秀等 App 调整脸部，拖拖修图进度条就好了～完成度就从50%->80%甚至更高！

如果你用的是醒图，我建议关注一下【人像 -> 面部重塑 -> 面部 -> 比例】，调整一下上庭、中庭、下庭。

---

📝 其他备注 1:

🎨 在转绘这部分有一个重点案例是照片 -> Minecraft 这种风格幅度变化很大的。

有小伙伴的问题是发现即便 --sw 风格权重拉满了还是复制不了像素小人风格。

经过实验发现：原因是 --cw 太高。猜想原理：cref 提取的人物本身包含了一定程度的风格特性。MJ 把这部分特质也当成了「作为你 character 的一部分」。

调低 cw 发现成功复制，猜想正确。（见视频）

---

📝 其他备注 2:

🤔 cref 和 sref 还可以同时参考多张 url，那是不是说我只要上传同一个人不同角度的照片，就更容易直出呢？

🧪 实验发现不是..用语言描述的感觉是：MJ 它会把多张照片理解为多张人脸，直接把形状揉在一起。结果非常容易畸形。

[@levelsio @levelsio]

👨‍💼 People always ask me how I invest my money, so I open sourced my investment portfolio at https://t.co/sQ0aiU7v02

There's no big strategy here and no certainty that this is a good portfolio either, right now it returns about $120K/mo

I try buy mostly ETFs and they have the highest return for me overall

Sometimes stocks rise so much though it becomes disproportionate % of portfolio (like NVIDIA)

I don't invest in startups now, but maybe one day I will. Why not now? From my studies of VC funds: you need to invest 100 to 1000 times @ $100K = $10M to $100M to have one home run exit that makes your fund profitable statistically

Unless you for example put in $1M yourself and then raise another $9M to $99M from rich people and pension funds (via LPs) that means you statistically you make no chance of ever making a profit on your investments. So not sure I should do it at all now.

ETFs though are nice, they let me invests in tens of thousands of companies all around the world while paying for ex 0.07% in fees (called TER, Total Expense Ratio)! That means investing $100K in 10,000 companies costs me only $7 per year! The cheapest investment products on the planet.

Since I'm not American I buy the European versions (but still in USD) of ETFs (called UCITS), for ex VOO in US is called VUSD in EU, there's some tax savings if you buy it in Europe (if you're not American) etc.

Anyway I don't know much about it, just trying to do the right thing and not chasing short term profits but long term 20-30 years

Some of my inspirations for investing are John Bogle (inventor of ETFs) and Ray Dalio (tl;dr Asia is the future) and those depict my investment strategy

[Cos(余弦)😶‍🌫️ @evilcos]

Web3/Crypto 新手入门资料现在推荐什么？我学习学习下。不要技术资料。

[徐冲浪 @cyrilxuq]

作为前内容从业者，给几点国人外网交流的建议:

0.最安全的外网访问方式是一线城市，只看不发言，不评论任何内容，当工具使用

1.外网可以讨论经济，中国经济只有市场派和特色市场派，大家讨论都是改革开放的力度，为了赚钱，国家也是为了赚钱，没有问题

2.人在国内的不要键政，国外的随意，成年人了，别最后国外留不下又回不了国，自己承担后果

3.不要提具体的人物、地方，只有企业是你可以随便欺负的(华为除外)，而且欺负民企比较安全

4.国家有部门专门盯着，不要觉得外网没人管

5.如果有生意在国外，或者要收美金换为人民币，请在北京注册一家公司，北京因为不透明性和特权阶级较多，安全系数最高，上海、杭州、深圳已经出现多例奇怪事件，北京还没有。如果有条件，直接国外企业收美金不回国。

6. 昵称上网，微信昵称尽量独一无二，不要实名上网，避免误会躺枪

7.美股/加密货币交易尽量少晒单，晒只晒收益率，不要晒本金，问就是10u战神

8.不要线下见推特认识的网友，没必要，对方多有钱多大哥都别见，没用，而且你不一定知道对方是做什么的

[千寻｜AI 分享 🌸 @Crypto_QianXun]

ChatGPT 终于可以安装 Canva 插件了🐮

试了一下，安装 Canva 插件后，即便是设计新手也能在几秒钟内自动创建出各种各样的横幅、海报、短片等内容

ChatGPT + Canva，完美搭配！强强联手，省时又省心！

在插件中搜索「Canva」

[M1kes1 @Mixessi]

@RoieeeTTV 加油百神！

[Greta008 @Greta0086]

我的天，这些二维码简直炫酷到爆炸，一个这样的微信二维码展示给别人扫，是不是超赞！想知道怎么做的吗？

刚找到的炫酷二维码制作教程： https://t.co/7W34hL0aen （第一步改一下，就在微信生成）

完成任务的留下你的二维码，发现特别炫酷的我就扫了啊。

努力尝试中……

#ControlNet #ai

[外汇交易员 @fxtrader]

微软在发布会上展示Edge浏览器整合GPT技术后的新功能——财报分析
打开一份15页的GAP财报，要求总结，随后Bing Chat展示出核心要点。
要求和Lululemon三季度财报对比，Bing从网上抓取数据，很快得出各项指标对比结果。
😲现在还只是演示效果，但是助力投研这方面还是令人期待。

[Coaches' Voice @CoachesVoice]

• Goalkeeper coach: Martín Tocalli
• Assistant coach: Roberto Ayala
• Assistant coach: Walter Samuel
• Assistant coach: Pablo Aimar
• Head coach: Lionel Scaloni

The team behind the team. 👥👏

#ARG • #WorldCupFinal