Bo (@0xboyu) header
Bo (@0xboyu) profile picture

Building & researching agent systems — verification, control, supervision, and beyond. Founder & investor. 🥷 dashing through rabbit holes 🕳️

Joined April 2018
Photo by Bo (@0xboyu)Photo by Bo (@0xboyu)Photo by Bo (@0xboyu)
😂
Be careful when walking around the COLM venue at night. I was out alone and got approached by a group of people to discuss continual learning
59
The world of R&D is forking into two paths: the token-abundant research, and the token-starved research. The future is in the former - evidentially, the progress by today's top AI industry teams and neolabs is breathtaking, where researchers’ human brilliance is super charged by AI’s assistance. Every research university president should be reading this report and reflecting on what the future of higher education research should be. openai.com/index/research-ac…
80
412
52
2,622
339,746
Glad to make it into the top 10 in Academia track :) github.com/0xboyu/aml-memory…
The first season of Agent Memory Leaderboard (AML) is complete. Explore how 100+ teams, 10+ benchmarks, and different memory approaches were evaluated under a unified framework — and what this reveals about the future of AI agent memory.
1
1
3
128
Bo retweeted
Garry Tan says we'll see a wave of exceptional older AI-native founders: "There's going to be as many Patrick Collisons as ever, but one mega trend that we're seeing is the 35, 40, 45-year-old founder who's been around the block, built a lot of engineering." "Peter Steinberger is a perfect example of that. He's been around the block. He knows what to build." "If you take that person, suddenly there's 400 of those people. You can outperform an entire department of any Mag 7." @garrytan @illscience
Garry Tan on YC, First Principles Thinking, and Progress in the Age of AI Silicon Valley's advantage is that it gives weird, ambitious outsiders permission to try, find one another, and become insiders through building. Garry Tan learned this firsthand throughout his career at companies like Microsoft, Palantir, and YC. AI may now extend the power of Silicon Valley much further. Garry joins a16z’s Anish Acharya to cover the turns that shaped him, why founders should trust direct experience over consensus, how AI is allowing small teams to do more with less, and why he's deeply optimistic about AI-native companies and the future of SF: 00:00 Intro 00:26 Getting into tech 04:45 Silicon Valley culture & finding your people at the fringe 10:59 What makes YC great: a birthright for tech outsiders 13:54 Solo founders, vibe coding, and founders being 400x themselves 21:05 Business loops: skillifying every task into a markdown file 23:03 Tokenmaxxing: how to live in 2028 today 28:09 Using AI to make conflict constructive 33:10 The torture of the white-collar job & life above the API line 39:08 Why bureaucracy is a whitepill 41:33 What the next computer looks like: voice, memory & the harness wars 44:44 Local politics & making SF a beacon youtube.com/watch?v=fsTtKywm… @garrytan @illscience
71
89
65
775
863,123
AI-native software engineering teams operate very differently than traditional teams. The obvious difference is that AI-native teams use coding agents to build products much faster, but this leads to many other changes in how we operate. For example, some great engineers now play broader roles than just writing code. They are partly product managers, designers, sometimes marketers. Further, small teams who work in the same office, where they can communicate face-to-face, can move incredibly quickly. Because we can now build fast, a greater fraction of time must be spent deciding what to build. To deal with this project-management bottleneck, some teams are pushing engineer:product manager (PM) some teams are pushing engineer:product manager (PM) ratios downward from, say, 8:1 to as low as 1:1. But we can do even better: If we have one PM who decides what to build and one engineer who builds it, the communication between them becomes a bottleneck. This is why the fastest-moving teams I see tend to have engineers who know how to do some product work (and, optionally, some PMs who know how to do some engineering work). When an engineer understands users and can make decisions on what to build and build it directly, they can execute incredibly quickly. I’ve seen engineers successfully expand their roles to including making product decisions, and PMs expand their roles to building software. The tech industry has more engineers than PMs, but both are promising paths. If you are an engineer, you’ll find it useful to learn some product management skills, and if you’re a PM, please learn to build! Looking beyond the product-management bottleneck, I also see bottlenecks in design, marketing, legal compliance, and much more. When we speed up coding 10x or 100x, everything else becomes slow in comparison. For example, some of my teams have built great features so quickly that the marketing organization was left scrambling to figure out how to communicate them to users — a marketing bottleneck. Or when a team can build software in a day that the legal department needs a week to review, that’s a legal compliance bottleneck. In this way, agentic coding isn’t just changing the workflow of software engineering, it’s also changing all the teams around it. When smaller, AI-enabled teams can get more done, generalists excel. Traditional companies need to pull together people from many specialties — engineering, product management, design, marketing, legal, etc. — to execute projects and create value. This has resulted in large teams of specialists who work together. But if a team of 2 persons is to get work done that require 5 different specialities, then some of those individuals must play roles outside a single speciality. In some small teams, individuals do have deep specializations. For example, one might be a great engineer and another a great PM. But they also understand the other key functions needed to move a project forward, and can jump into thinking through other kinds of problems as needed. Of course, proficiency with AI tools is a big help, since it helps us to think through problems that involve different roles. Even in a two-person team, to move fast, communication bottlenecks also must be minimized. This is why I value teams that work in the same location. Remote teams can perform well too, but the highest speed is achieved by having everyone in the room, able to communicate instantaneously to solve problems. This post focuses on AI-native teams with around 2-10 persons, but not everything can be done by a small team. I'll address the coordination of larger teams in the future. I realize these shifts to job roles are tough to navigate for many people. At the same time, I am encouraged that individuals and small teams who are willing to learn the relevant skills are now able to get far more done than was possible before. This is the golden age of learning and building! [Original text: deeplearning.ai/the-batch/is… ]
226
363
74
1,900
375,516
Don't just reset Codex rate limits for fun, it costs money. Don't just reset Codex rate limits for fun, it costs money. ... but the vibes are good ... I have reset Codex rate limits for ALL paid plans to celebrate a good week and allow everyone to build more with GPT-5.5. Enjoy
1,415
724
644
16,849
1,332,506
LLM Knowledge Bases Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So: Data ingest: I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them. IDE: I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides). Q&A: Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale. Output: Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base. Linting: I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into. Extra tools: I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries. Further explorations: As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows. TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.
2,938
7,397
2,221
61,093
22,034,345
Every claim should have skin in the game. Those selling fake news should pay. Charlatans should get arbitraged. Introducing the new Precog. A fully onchain prediction market protocol. Built for signal. Here's what's new 🧵👇
24
29
12
112
4,647
1/🧵 We are very excited to release our new paper! From Entropy to Epiplexity: Rethinking Information for Computationally Bounded Intelligence arxiv.org/abs/2601.03220 with amazing team @ShikaiQiu @yidingjiang @Pavel_Izmailov @zicokolter @andrewgwils
56
399
92
2,446
1,144,491
0/ What if Web3​ could superpower AI by enabling agent collaboration across organizations in untrusted settings? Our ERC-8004: 🤖Trustless Agents extends Agent2Agent protocol, providing a unified interface to discover agents and establish trust through reputation and validation. Written with @ethereumfndn's AI Lead @DavideCrapis and many #DeAI builders🧵
17
39
19
208
39,585
✨以太坊生态日报(2025-09-09) 【技术升级与协议动态】 1⃣EIP-7732(ePBS)被选为Glamsterdam主题提案,重点讨论其重要性及社区反馈。发布者@ether_world为以太坊生态资讯平台。[原文](teapawt.net/ether_world/status/195…) 2⃣- FOCIL(EIP-7805)获核心开发者与社区支持,文章解析其功能及未列为主题提案的原因。@ether_world提供深度分析。[原文](teapawt.net/ether_world/status/195…) 3⃣- Starknet宣布完成阶段1升级,持续推进零知识证明技术。未标注发布者身份。 【开发者工具与创新】 1⃣ 开发者工具速览:@ether_world汇总5项新工具: 1 Sonic测试网2.1支持以太坊Pectra升级,增强智能合约兼容性; 2. Nadcab Labs推出AI驱动的加密钱包,支持自动化链上管理; 3. Python框架Moccasin支持zkSync快速开发; 4. 模块化应用框架vApps提升以太坊dApp可扩展性; 5. 去中心化协作平台Radicle获Web3开发者青睐。[原文](teapawt.net/ether_world/status/195…) 【学术与合作】 1⃣以太坊纽约研究资助论坛召开,@VitalikButerin提出四大研究方向:质押宏观经济学、FHE友好型哈希/签名、形式化验证及治理设计。@theobtl全程跟进。[原文](teapawt.net/theobtl/status/1955640…) 2⃣谷歌云Web3负责人Devan Mitchem探讨以太坊公共产品建设,分享机构合作视角。@theobtl记录会议内容。[原文](teapawt.net/theobtl/status/1955664…) 【生态应用与活动】 1⃣纽约“World of Apps”活动今日举行,展示13个以太坊应用及现场艺术创作,由@nounsdao等赞助。@jchaskin22预告活动亮点。[原文](teapawt.net/jchaskin22/status/1955…) 2⃣ @ethereum官方账号发布“自主代理商业”主题长推,探讨HTTP 402标准如何通过以太坊实现机器间无摩擦支付,并列举多个已集成案例。[原文](teapawt.net/ethereum/status/195567…) 【行业观点】 1⃣ @VitalikButerin调侃批评者未使用“警惕极客陷阱”标语,引发社区对加密叙事的讨论。[原文](teapawt.net/VitalikButerin/status/…) 2⃣ @matteoikari关注@reserveprotocol的指数代币协议潜力,认为可能开创链上资产组合新范式。[原文](teapawt.net/matteoikari/status/195…) 【安全与研究】 1⃣以太坊证明市场设计研究发布,探讨ZK证明生成的经济模型与激励机制。@ethresearchbot同步摘要。[原文](teapawt.net/ethresearchbot/status/…)
EIP-7732 (ePBS) Selected as Glamsterdam Headliner - Why ePBS Matters - Community & Stakeholder Feedback etherworld.co/2025/08/11/eip…
2
1
6
3,799
Ross is free. His mother never gave up. The whole story is about one mother's unstoppable love for her child. She kept going and then got him free. Women have sacred power. They create life and nurture society. They are diplomats and empaths. And mothers are true power.
To understand the Ross Ulbricht case, here are some excerpts from my article: The reason why the libertarians cheered Trump’s announcement was that the sentence was profoundly unjust and crazy, and because Ross is an intelligent, good-hearted, and wonderful person whose work in this realm had nothing to do with a fixation on illicit drugs. It was about an experiment in market freedom, which he undertook because the technology allowed it. His purpose was simply to show that free markets work. Nothing more. There was plenty that was fishy about the arrest and prosecution of this man who did only technical work, including the way in which named and unnamed accomplices were treated with kid gloves while Ross’s head ended up on the platter. The plaintiff’s arguments to the jury included accusations that were never even made in the indictment or came up in the course of the trial. We can be sure that Ross is only one person among thousands or tens of thousands of people caught up and caged as a result of a criminal justice system that cares nothing for fairness or human rights. We might ask how it came to be that so many people held up signs with Ross’s name on it and why the former and likely future president, in addition to the independent Robert F. Kennedy, Jr., have taken up his cause. There have been many wonderful people who have tried to raise awareness of his case over ten years. But one name stands out above the rest. Her name is Lyn Ulbricht. She is Ross’s mom. Back in 2014 when she first started advocating for her son, hardly anyone could be bothered to care. I recall these days well and I was disappointed and astonished at the indifference toward his plight. The libertarians didn’t really raise a fuss either. Too many facts were in dispute, and there is always a risk associated with lobbying for a jailed convict. But that didn’t stop Lyn. She figured out early on that if she was going to make any progress at all, the libertarians were going to be the most receptive. After all, she had not had much success in just posting signs on storefronts and telephone poles, which she also did tirelessly. She was happy for any audience even if only mildly interested. So she went to every possible meeting, at her expense. She flew around the country. She raised money. She spoke to countless audiences. She traveled around the world. She would talk for hours to podcasters and media people, anyone who would listen. She mastered all the legal details, and was very precise in her language. She hosted events. She passed out bracelets and signs. She established social media accounts and humanized the life that he was enduring in prison. She gradually made progress. It was due to her tenacity born of a deep love—a mother’s love—that has been the background power of this movement for a decade. Her faith never flagged. Not once. Yes, she cried often, prayed, and talked about hardly anything else for ten years. Even when appeals were denied and pushes for leniency were rejected, and Ross was transferred from one place to another, she did not give up hope. As a mom, her job was to do the impossible for her son even if the rest of the world regarded the cause as hopeless. In her mind, there was always hope. She just needed to find a way. As I saw the signs in the crowd this weekend, and heard two presidential candidates make an appeal for Ross to be freed, I immediately imagined the joy in Lyn’s heart. I’ve not spoken to her but I know her and I know it is there. This is a woman of incredible faith, boundless energy, unstoppable passion for what’s right, and a seemingly infinite ability to push, press, and lobby for her son’s life and freedom, with anyone who would listen, anyone with some minor hope of influencing outcomes. If that story doesn’t fill you with inspiration and respect, there is something wrong with you. What we see in Lyn is something that is almost not of this world. But you know what? It is, in fact, very much of this world. It is all around us. It is the divine in everyone among us: the fire of a mother’s love for her child. Watching this all unfold should remind us all of the remarkable power of this beatific force at work. It is unstoppable, more powerful than all the lobbies and even the armies. We’ve seen the power of maternal love strongly at work in the world over the last year. It was the moms who demanded that the schools be open. It was the moms who saw what their children were being taught and formed large and influential organizations to stop and block the corruption of the curriculum. It is the moms who are out there now demanding the right to have knowledge of medical procedures being pushed on the kids. It is the moms who are demanding choice over the shots being forced on the kids. And what a gigantic difference it has made! The schools did open because of it and they are being reformed because of it. It’s the moms who have marshaled the energy to push back on the Great Reset and draft the kids into a creepy psy-op experiment in gender transitions. They will have none of it. It’s the moms that are fed up with seeing their daughters’ sports programs raided by boys in drag. And looking back in time, it was the moms who demanded the carve-out legislation that made homeschooling possible. And it is the moms today who are getting ever more engaged in hugely important causes like food and medical freedom, without which we have no health. This is what happens when you cross the moms: you face a power that is not of this world. The politicians have begun to understand this, and fear it. I’m not speaking of some vague and ideological form of abstract “activism” promoted by the likes of Hillary Clinton and Melinda Gates. These are people who deploy gender as an ideological weapon against society. I’m speaking of something far more pronounced and real, a tactical passion unleashed under particular conditions, one rooted in love. No amount of money can buy it and nothing can stop it once it is unleashed. theepochtimes.com/opinion/a-…
59
172
10
1,702
235,381
Ethereum Foundation Treasury Update The Ethereum Foundation (@ethereumfndn) has set up a new @safe 3-of-5 multisig wallet. The wallet address is 0x9fC3dc011b461664c835F2527fffb1169b3C213e An op has been initiated to send 50,000 ETH there, but be patient; due to signing delays, this will take a few days to process (etherscan.io/tx/0x6872c97c57…) The EF will use this new wallet to participate in the DeFi ecosystem. We've made a testing tx on @aave. Stay tuned! Also, @safe has proven safe and has a great user experience, and we will transfer more of our funds here over time! #helloworldcomputer
489
479
333
3,355
1,326,049
zk-TLS stands to upend Web2, breaking monopolies and disrupting trillions in value. The first use cases reach PMF in 2025, enabling on-chain identity scores, vampire attacks on platforms like Uber, and more. @MessariCrypto x @EV3ventures deep dive co-authored by @VinayakDKurup
13
35
25
189
44,926
Devcon is an antifragile experience. you leave the comfort of your home and country removed from your usual work and daily routine dropped into a new location and culture around the largest density of likeminded people in the world, from around the globe immersed in conversations and experiences that you would never have had back at home when you return home, you have completely new perspectives and connections It's not necessary if you are locked in as an engineer or founder sprinting towards realizing a vision but for builders looking expand their opportunities, or wanting to create and strengthen 1:1 relationships, or to find new opportunities to collaborate with like-minded people, there isn't a bigger or better Schelling Point in the world (in crypto)
21
23
2
342
21,109
devcon bike ride day 1 vs devcon bike ride day 7
45
49
8
627
47,361