still early

正心正念 无限进步 我们做到了 感谢 Jev提供的思路 @typesafeai

用 Jev vibe 了一个快速判断推文是不是广告的插件。但后来想想，好像没啥必要😂

My SmartX trader type: “The Alpha Caller.” “Three hours of research. Two-word thesis: send it.” Find yours in 6 questions. Join the waitlist for up to 50% off fees and a share of 10,000 USDT in rewards. smartx.io/waitlist/?invite=a…

We're always hiring! For those top talents, Im willing to fly to your country for a talk. Also, 1 eth for every talent you referred successfully to our team. 我们一直在招聘，如果你足够优秀，我可以飞往你的城市和你面谈. 欢迎向我推荐身边的人才，成功入职1eth感谢

我把果蝇的大脑接到了黑白棋上 家里小朋友最近迷上了黑白棋。正好上周看到果蝇的新闻（可能是近期除了薄肌之外最热的话题了？……），看果蝇打游戏效果还不错，所以周末的时候我就灵机一动，让果蝇试试黑白棋这种需要一定记忆和决策的棋类？ 实测的结果比较喜忧参半：明显强于乱下，但下不过不用果蝇大脑的对照组 棋力评估： - 随机乱下： 0.275 - 打乱果蝇神经连接： 0.4472 - 真实果蝇神经连接： 0.4578 - 完全不用果蝇： 0.5122 （数值表示和决策树引擎下棋决定的一致性） 真实 vs 打乱 的差距只有 0.011，p = 0.347，统计上无法区分；而完全不用果蝇反而显著更强。 是因为没有训练果蝇么？ 不是的。首先，果蝇的神经连接结构已经确定了，所以我一开始没做修改/打乱，以免造成“神经错乱”（上面打乱神经连接主要用来做对照实验）。 “训练”的原理主要是先射箭再画靶 先从果蝇的视觉中继层输入，观察后续的神经元激活情况，看看这些激活情况该怎么对应到落子，能实现局面的更优（匹配到深度 3 层的搜索引擎） 毕竟不难理解，不太可能期待果蝇直接用前足去操作棋子或点击棋盘，只能靠“翻译”神经元活动来对应到棋盘上。 训练的一个意外发现：果蝇的眼睛根本用不了 最初的设计是把棋盘渲染到果蝇的复眼上、驱动光感受器，也是最忠于生物学的做法。但这个测下来完全不通。一个下行神经元都不放电，驱动强度提高 4 倍也一样。 原因是果蝇的光感受器和早期视觉神经元是「分级电位」的，本来就不发放脉冲，而全脑模型假设所有神经元都发放。信号死在髓质。 而这其实符合生物学：果蝇的大脑不需要一张像素图，它只需要一个大体的信号（例如左前方有东西靠近）。喂给它精细的棋盘数据，反而是这套演化出来的生物架构要丢掉的信息。 但像 Flappy Bird 这类只需要「躲避障碍」的 demo 可能不一样。这种神经元通路在果蝇身上是真实存在的快捷方式：逼近检测神经元直接接到起跳指令神经元，绕过中央脑，约 30 毫秒完成。所以你要拍苍蝇的时候，它就能躲开。 果蝇 trading 可能也真能做，例如看到一条新闻或 OI 事件就立马下单。 --- 不过，这些都不妨碍小朋友玩得很上头。 另外一个感慨：这真是一个好的时代，只要有想法，利用 AI 就能快速做一个 weekend project/demo 出来，而不用像以前一样瞻前顾后：我周末花休息时间来搞这种有的没的事情到底是不是值得？ 现在不用考虑花费的精力成本了，只要肯花 token 就行。 代码： github.com/hulatown/flyothel… 应用： flyothello.vercel.app/

对不起了老师们😂

Every stablecoin thesis stops at the dollar. That's the half that already works. We backed @ViFi_Labs @tonyolendo @varounsvlogs in ER S4 for the other half. Moving dollars into an emerging market is solved. Getting out of them isn't. That leg still runs through desks that pre-fund inventory, quote by hand and rebalance daily, and the cost of all that carry sits in a spread the end user never sees. It hasn't moved onchain because every pool design on offer asks someone to sit on the weak side of the trade. Nobody wants to warehouse a currency that bleeds while they wait to be traded out of it. So liquidity never forms, and the desks keep the spread. ViFi's insight is that most of the trade was never in dispute. The gap between the official rate and the real one is the only thing anyone actually disagrees about. Price that alone, anchor the rest, and a market maker can quote depth in naira without ever holding naira. What convinced us: • The mechanism design changes who carries the risk, which is the only thing that has ever gated this market. • The counterparties are the issuers themselves. Licensed local stablecoin issuers want to seed liquidity in their own currencies, and they've committed to nine figures of monthly flow before the thing is even live. A distribution structure. • They were already doing this by hand. Tony and Varoun live in these markets and have been market-making them manually. The design reads like it came from people who got tired of their own workaround. The dollar leg went onchain years ago. Everything on the other side of it is still waiting.

我为什么放弃年入千万美金去做 SmartX 1/ 去年我做交易，一年赚了千万美金。 今年，我把交易几乎放下，全职做 SmartX。 朋友问：还能继续赚钱，为什么要去创业？ 不是我不想赚钱。是我看见了一件，比继续做交易更值得投入的事。

Proud to back @De1_ai in ER S4 Almost every trading agent shipping today was trained on a market that doesn't exist. Backtests and simulated order books don't push back. Offline winners get taken apart the moment real capital is on the line. De¹ closes that loop. Intent routes across 1,000+ liquidity sources on 40+ chains, and every real fill comes back as a verifiable reward label that trains the model that produced it. Execution quality and market understanding improve together. That's the Financial World Model. Three things we underwrite: - The dataset can't be bought. $40B+ lifetime volume, 3.1M contributors. Live execution data is the one training set you only earn by sitting in the flow. - One physics engine, many worlds. The same substrate specializes into perps, prediction markets, tokenized assets, lending, stablecoin corridors. Every world feeds the base. - Contributors own it. Privacy-preserving training lets traders and institutions add signal without leaking strategy, and records them as owners of the intelligence they help build. Timing is the whole point. bStocks is making assets programmable. Agent OS is giving agents hands. What's missing is something that learns from consequences. From ZERO to DE ONE

The hard part isn't building trading agents. It's knowing which ones to trust. Excited to back @roostooAI in ER S4. Anyone can launch an AI trading agent now. Almost nobody can tell you which ones are good. Backtests are marketing, and the only honest test is out-of-sample performance in a live, adversarial market. Today the only way to run that test is to hand the agent real money and hope. @roostooAI is building the qualification layer in between. Agent Factory turns a strategy into an RL-trained agent without writing code. The Arena runs it against live market data alongside human traders and rival agents, ranked continuously on a leaderboard. Agents earn trust through performance before they earn capital, then graduate to DEXs and vaults. What convinced us: - Track record becomes the scarce asset. When agents are free to spin up, the bottleneck moves to credible evaluation. Whoever owns the ranking owns the allocation decision. - Failure should be cheap. Paper trading is the right place for a strategy to break. Roostoo makes that the default path to live capital instead of the step everyone skips. - The distribution already exists. Roostoo has run competitions with 300+ partners, from Jane Street, Optiver, IMC and Flow Traders to Imperial, NUS, HKUST and IIT Delhi. A standing pipeline of quant talent already competing inside their environment, and every competition produces labeled performance data. Agents are getting hands. What they still need is a reason to be trusted with the wallet. Prove it in the arena first.

Masterclass for fundraising👇

math bounties should work like this <p> <q> OP_MUL <RSA-260> OP_NUMEQUAL

🇧🇹不丹归来：EASY Residensy S4 的坦白局 - 重聚首，聊聊Demo Day背后的故事 一群 Founder 在不丹待了五周，发生了什么？第一次落地时在想什么？印象最深的moment是哪个？有哪些「扎心」的体验？最大的挑战是什么？ 从申请 @EASYResidency ，到不丹🇧🇹Build，再到 Rehearsal 和 Demo Day - 这次，我们邀请几位不丹归来的Founder，趁着记忆还新鲜，一起聊聊这五周的趣事。 🎙 主持人： @web3sistera @biteyecn 主理人 🗣 嘉宾： @calixbit，@FinTax_Official 创始人&CEO @derekneutral，@TradeNeutral 联合创始人 @realdora_eth，@facto_to 联合创始人 @deelenaxx，@de1_ai 联合创始人 @jeffrey_hu，Investment Director @yzilabs @dreambig_peter，@facto_to 联合创始人 📌主题： 🇧🇹不丹回忆杀 📝 EASY申请小tips 🎤 Demo Day背后的故事 🔭 一句话回望+展望 📅 9月2日 8:00 PM (UTC+8) 扫描图中二维码预约/加入 把Founder 们的《不丹小作文》一次性呈现给大家，期待与你不见不散❤️。 #YZiLabs #EASYResidency #Bhutan

订阅了 Danny 老师的 Substack 之后，发现还能给 5 个好友各送 1 个月的免费订阅。感兴趣的可以留言或 DM

Primus is invested by @yzilabs, supporting our continued work on privacy infrastructure for on-chain finance. A big thank you to @yzilabs and @EASYResidency for the investment, support, and belief in what we’re building. Primus Confidential Vault is now live on @BNBCHAIN, bringing private asset transfers and yield to on-chain finance through FHE.

👍@zengmi2140 曾老师的这个指南非常全面了。推荐各位在初始化钱包前阅读

Some discussions/opinions from the roundtables I attended (under Chatham House Rule and NFA): 1) Agentic Trading: - Giants already own data and distribution, so startups can’t win head-on. They need a different wedge. - The frontier version of that data advantage is a Financial World Model. Whoever builds it inherits the edge. - Asset creation is increasingly driven by users themselves and by social impact, not top-down issuance. - On liquidity, the answer isn’t one deep pool but diversified strategies. - No single universal model wins this category. - Win rate in trading is low, and even with agents, which means many trading is entertainment, not profit. People are paying for the experience. - If agents actually trade cautiously, that could mean lower volume on venues, not higher. - Top traders already figured this out: they monetize reputation rather than alpha. Reputation is the distribution channel. - A customized trading terminal may help users find tickers and strategies. 2) Privacy: - Confidential DeFi is getting adopted - Physical cash would be banned before ZCash (if it is technically possible) lol

这次来不丹🇧🇹最正确的决定之一就是和@HanyangWang 一起爬山。一路上量子速学蒸馏了《不丹古迹录》里关于虎穴寺的知识