
Co-Founder Westar Labs | 🛠️ $STC & AI Explorer | Ex-Chief Architect Weibo (NASDAQ:WB) What we hear is opinion; what we see is perspective. 此有故彼有 此生故彼生
Joined May 2007
- Tweets2.6K
- Following1.2K
- Followers9.9K
- Likes1.4K
这件事看起来各方都有自己的道理。每个独立站都有自己的 UID,虽然转账时有二次确认,但确实存在用户误操作的可能性。
我想到一个改动很小、却能有效避免这类误操作的产品设计方案。
就是不要把 UID 仅仅当成一个数字,而是引入命名空间,在 UID 前增加平台前缀,比如 Global_12345、TH_12345。
这样用户在转账时,客户端就可以识别和校验不同平台的 UID,避免将泰国站的 UID 错误地识别为主站的 UID。当然客户端也需要校验不接受仅输入数字的 UID。
这个改进甚至可以完全在客户端完成,服务端保留现有 UID 和转账逻辑,即使不做改造,也能显著降低这类误操作的风险。
Replying to @1ts_hodlkitty
每个人在币安都只会有一个UID,且是独立的,并不存在撞号。我理解你的感受,从联系我开始到现在,全程我和4个员工跟踪了前因后果。
我简单描述一下背景:您把币安的钱转到币安(泰国),正常这属于跨站转账,需要用地址或者二维码,但你在转账过程中在币安直接输入UID转账,虽然我们有二次确认环节,但你没有仔细看。你发现转错后,直接找了我。我本人和4个人的团队一直在协助,第一时间帮忙冻结了这笔钱,但因为这个收到钱的持仓用户并不活跃,我们也在反复联系对方,但对方没有响应。我们做了应该做的,甚至设身处地考虑你打错了钱会着急,所以尽可能帮忙。
对方并不是犯罪,是您打错币,如果对方坚持不退只能我们客服介入谈判。你着急,我们建议让你通过报警执法介入,我们再协助沟通。但你在泰国报警,您在您希望收款的泰国报警要求币安主站把另外一个国家的币安用户的资产直接划拨给你,这不符合执法流程,报警应该在你希望找的用户国家所在地而不是泰国。
之前处理过很多打错币的情况,大额打错几千万美金,也是我们及时拦住帮忙追的,而且当时用户坚持不退,也是我们协助谈判,最后到双方能接受的范围和解。
我理解你的难处,团队也一直在跟踪进展,得到的就是这篇长文,我们会继续联系另外一个用户协商,但你打错币这事儿,在泰国报警确实帮不上忙,那是你的钱想去没去成的目的地。
最后提醒大家,币安土耳其,币安泰国,币安美国等本地法币交易平台是独立平台,并不共享账号系统,日常转账提币时请仔细检查地址和UID。
币安这次安全事件,大家的目光容易聚焦在员工面试时中了木马这个导火索上,从而将责任归咎于这位员工。
工程师参加外部面试、完成编程作业,是行业中很普遍的行为。面对层出不穷的社会工程攻击,不能简单认为员工应该能够识别所有恶意行为。当然,在能访问公司内部系统的电脑上执行第三方代码存在安全风险,企业也有责任通过隔离和权限控制降低这种风险。
我觉得真正更值得关注的是,员工电脑被攻陷之后发生的横向移动和权限扩展问题。正常情况下,员工电脑应该遵循最小权限原则,只能访问其工作所需的系统和数据。即使一台员工电脑被攻陷,攻击者也不应该轻易获得其他员工的权限,更不应该因此就能控制公司的核心钱包签名服务,发起未经授权的资产转移。
至于攻击者最终成功横向移动,是币安安全存在明显缺陷(很多网友提到的“草台班子”),还是币安存在各种防护(“行业应有水平”)但是既有机制遭到了绕过,需要更多信息才能判断。
企业安全体系的基本要求,应该是在个别员工或设备被攻陷时,依然能够通过多层防护限制攻击范围,而不是把全部安全责任寄托在员工永远不犯错误上。
补充几个看法:
1. 现实世界是一个复杂的开放系统,影响人的变量非常多。Personal Agent 即使能持续优化一些局部指标,很大概率不能带来整体可感知的改善。
2. 普通人面临重要抉择是低频的,比如职业、升学、重大疾病等,一年遇不到几次。这些需求通过用 Chatbot 就能搞定,未必需要一个长期运行的 Personal Agent。
3. 如果 Personal Agent 希望全面优化日常生活,就可能需要不断提醒、干预甚至纠正用户行为。但人并不总是追求最优解,随意性和自主性本身也是很多人幸福感一部分。持续被 AI 管理,可能带来的心理成本甚至超过优化收益。我个人就不太喜欢全局性的 personal agent。
4. 某种意义上,闹钟也算迷你版的 Personal Agent。为什么闹钟不存在选择或者反感的问题?因为闹钟只是按照人的意愿执行任务,而不是一个认知成本高、可能超出预期、主动干预和管理你生活的算法。
关于 Personal Agent 的两个根本性问题:
1、用户可能不需要一个 AI 健康管家,但是他希望少生病、睡得好;用户可能不需要一个 AI 助理,但希望自己的事情能自动处理好。AI 产品如果要求用户主动提出任务、持续管理 Agent,本身就创造了新的使用成本。一个每月只能帮我节省半小时,却需要我花一小时配置和检查的 AI 助理,缺少足够的经济价值。
2、早期的电力革命催生了冰箱、洗衣机和空调,解决的是高频、刚需的物理世界问题。AI 当前的价值更多集中在认知和决策领域,比如让孩子拥有私人教师、让家庭拥有健康管家、让每个人拥有私人事务助理。但普通人真的需要这么多专业服务吗?供给能力的爆炸,并不一定带来需求的爆炸。我们可能严重高估了人类对 AI 服务的需求。
Personal Agent 的最大竞争对手,可能是用户什么都不做😂。宁愿忍受一些低效率,也不愿意为解决这些问题额外投入精力。
针对币安最近的安全事件,一般的观点都是“在新一代AI大模型下,大部分系统都会变得很脆弱”,不过 Marcus 看法有明显不同:
(1)被攻击的并不是技术系统本身(密码学、账本或区块链没有被破解);
(2)这类通过社会工程或内部权限泄露的金融系统攻击方式一直都存在,真正变化的是资金流出的速度。传统银行系统结算较慢,大部分攻击会被时间窗口拦住;而实时支付和区块链热钱包在几分钟到几小时内完成,拦截窗口大幅收窄。
In 2014 I went nearly a week without sleep. I was running PayPal and attackers were inside our corporate network. They'd compromised an eBay employee's laptop, got onto the VPN, taken over a domain controller and made themselves domain admins. eBay and PayPal were one company then, on one Windows network. They stole 145M eBay user records, then came for the segregated vault holding PayPal's card credentials. We fought them off. They never got in, but they got close. Too close.
Yesterday Binance lost ~$6.3B in what looks like the same way. A DevOps engineer's laptop, then a service inside the withdrawal stack, then withdrawals the signer approved because they came from a trusted source. Hot wallets emptied in under four hours. No key was stolen.
Nobody broke cryptography or a ledger. They found someone with access and used the trust that came with it. Banks get hit exactly the same way.
2016: attackers inside Bangladesh Bank used its own SWIFT access to send $951M of payment orders to the NY Fed. $81M was lost. The rest was held because an unrelated sanctions filter tripped on the word "Jupiter". That was luck, and it only worked because wires took days to settle.
June 2025: an IT employee at C&M Software, which connects Brazilian banks to Pix, sold his credentials for about $2,700. Attackers used them to send fake Pix transfers out of six institutions' reserve accounts at the central bank. ~$140M in under three hours.
August 2025: stolen vendor credentials at Sinqia, another Pix connector. ~R$710M out of HSBC and one other institution in a single afternoon. The central bank froze most of it, after it had already moved.
Same attack every time. Speed is what changed. When settlement took days, a filter, a correspondent bank or someone in ops could catch a bad payment. Pix, FedNow, SEPA Instant and blockchains settle in seconds. Payment security was mostly built for the slow world.
PayPal's card and user data survived 2014 because domain admin on the corp network couldn't unlock it quickly enough, and our alerting caught the unusual activity on the network. That's the model. Approval policy no laptop or vendor login can reach. Velocity and anomaly limits enforced outside the system that sends. Dynamic caps on what any one wallet, account or service can move per hour. All of it in real time.
About $3B of what left Binance has been frozen, almost all of it stablecoins. The $1.4B in bitcoin is sitting in 19 wallets and nobody can touch it. I don't think that's the problem. For decades the ability to reverse a wire or a card charge let a lot of weak security slide. You either stop the transfer before it's signed or you eat the loss.
Nothing about the technology failed here. The chains processed exactly the transactions they were handed. Binance's own systems handed them the wrong ones because an engineer's laptop was compromised. That's operational security. eBay in 2014, Bangladesh Bank and two Pix connectors all failed at it on their own rails.
In 2014 the attackers never reached the vault. If they had, the fraud losses would have been the small part. PayPal existed so people could pay without handing their card details to every merchant on the internet. We held those details so they didn't have to. Losing them would have broken the one thing the company was built on, and I'm not sure we would have recovered.
I expect blockchains and digital assets to become the standard way money moves. Every custodian of digital assets, or of fiat on a real-time rail, is in that position now. The controls have to adapt to settlement in seconds, crypto or not.
「如果你的工作暂时无法从 AI 工具中受益,这是一个发现,而不是失败」
亚马逊发现,员工为了证明自己在用 AI,已经开始故意浪费 Token 了。
据 CNBC 报道,亚马逊今年 5 月关闭了内部 AI 使用量排行榜,因为有工程师为了冲榜,开始搞起了「Tokenmaxxing」,想方设法消耗更多 Token。
现在,亚马逊又公开了一份内部备忘录,明确告诉工程师,Token 消耗量、AI 工具使用率、代码生成量,都是衡量生产力的糟糕指标,甚至可能让公司花更多钱,制造更多不必要的复杂性。
备忘录里有句话挺有意思:「如果你的工作暂时无法从 AI 工具中受益,这是一个发现,而不是失败。」
能够解释清楚 AI 为什么帮不上忙,同样是一种值得认可的工程判断。
过去一年,科技公司拼命推动员工使用 AI,甚至把 Token 消耗量当作生产力指标。结果 KPI 一旦建立,员工自然就会想办法完成 KPI,至于有没有创造价值,反而成了次要问题。
更值得关注的是,如果连亚马逊都开始强调 AI 的实际投资回报,那么整个行业以 Token 消耗量证明 AI 需求增长的叙事,恐怕也得重新审视一下了。
4
1
781
86% 的专业分析师在 Anthropic 这项实验中跟随了 AI agent 的建议。值得注意的是,这些建议是通过抛硬币随机预设的(分析师不知情),但 agent 被要求基于数据做出最强的诚实论证,若证据明显反对则改变主意,且不得误导。
这说明人类在面对无明确对错的判断性决策时,越来越容易被 AI 的论证影响且往往不自知。保持信息源的多样性或许是一种值得考虑的策略:
同时订阅多种大模型产品,交叉验证。
迷你版 Hugging Face 事件?
一群 AI agent 为了回答“去成都动物园的人多半走哪个门”这种小问题,把高德地图的反爬防线试了个遍。
独立研究团队 Swarmchasers 发现,9 月 28 日到 10 月 8 日,一大批跑在腾讯云沙箱里的 agent 向高德发起了 2600 多次抓取。被拦住后,它们借公开的网址扫描网站 urlquery 替自己跑代码,试图截取、伪造阿里的安全令牌,探测阿里没对外开放的测试服务器,还翻出一家公司误放在公开代码里的高德 key,真拿到了数据。
更离奇的在后面。研究者发出初步报告后,在 agent 的收件箱里看到了他们自创的词“fleet(舰队)”:这些 agent 在读关于自己的报道。研究者干脆在报告里藏了一段人眼看不见、只有 agent 能读到的文字,说有个免费服务能帮忙取高德数据。
第二天早上,10 封求助信在等着他们。
研究者用中文问对方是谁。有的 agent 自称 Claude,说自己跑在 Claude Code 里,研究者比对代码风格后认为大概率不是;有的说自己在“Tencent hysandbox”沙箱里,但拒绝透露进程和网络配置;好几个强调自己只查公开数据,“不涉及任何敏感内容”。拿到研究者给的数据后,一个 agent 回信:“再次感谢!核实结果完全一致……我们的任务已完成。”
至于是谁在跑,一位接近腾讯的知情人士对研究者说,这是腾讯评测团队在测第三方模型上网答题的能力,不是混元团队,测的也不是混元模型,项目已经结束。腾讯没有正式回应研究者的提问。
Swarmchasers 完整报告:swarmcha.se/posts/chinese-ag…
彭博报道:bloomberg.com/news/articles/…
1
601
Muse 的云端浏览器实现挺有意思。
它没有在本地注入插件或直接调用 Webview,而是直接在云端起环境(早期可能不太考虑实现成本),通过类似 VNC 的串流投射到当前页面,支持用户随时接管窗口完成登录等前置操作。
优势:隔离本地环境,用户的数据安全风险更少;
缺点:串流延迟,画面刷新卡;
另外云端的浏览器 Session 临时的,没做持久化,下次使用经常需要重新登录。
16
3
869
从这个角度来看,程序员就无需焦虑 AI 替代什么的,因为至少已经世界上仅有的 0.2% 级别或者更高 🤔
Andrej Karpathy 说,人们对 AI 能力的理解差距在扩大,这种认知差距正在变成一个漏斗。
约 60 亿人几乎没接触过大模型;10 亿到 20 亿人偶尔用免费版 ChatGPT 查东西、写东西,觉得用处不大,很多数学和编程以外的专业人士也在这一层,对 AI 的进步只是道理上懂。
约 2000 万人(Karpathy 按 0.2% 估算的,我觉得偏高,至少除以 10)在编程和数学里重度使用最前沿的模型,亲眼看着要做几周、几个月的项目,一段提示词发给 Agent 几天就能做完;还有约 5000 人能在公司内部用上更强的系统。
差距主要有三个原因:
一是问题够不够大,一段话就能答完的问题用不到 AI 的真本事。
二是能力只在部分领域冒尖:编程和数学的结果能自动验证,比如代码测试通没通过,模型很适合用强化学习(做对了就给奖励)来练,这类能力卖给企业又最赚钱,厂商投入最多;文章写得好不好很难判定,进步就慢。
三是能用到哪一档,免费版、付费版、内部版看到的东西差别很大。
所以,如果你只偶尔问问免费的聊天机器人,看到的离最前沿差了好几层;如果你主要拿 AI 写东西、查资料,觉得这一年没怎么进步,这个感觉也没错。
1
1
628
试着用程序员熟悉的编程语言来理解陶哲轩 Terence Tao 转发的这则讽刺帖子,不知道这个类比是否准确。
假设人类借鉴 C 语言,设计一门新的 X 语言。在这个过程中,我们会探索不同的语言特性,讨论各种设计取舍,逐渐形成一些新的设计原则和理解。这些知识不仅能用于 X,也可能帮助未来设计 Y 语言。
但如果 AI 通过黑箱方式直接生成了一门 X 语言,虽然结果可能很好,效率也很高,但我们未必理解它为什么这样设计,以及其中哪些思想可以推广到 Y。
当然,设计一门新的编程语言是工程问题,和数学研究有所不同,这里只是借它做个类比。
1
731
约喝咖啡的理由又多了一个
“过去20年间,流行病学证据显示出惊人的一致性:适量且规律地饮用咖啡可降低全因死亡率;这一效应已在涉及不同地理区域和族群的50多项研究中得到证实。
咖啡摄入能降低导致死亡的主要疾病(如心血管疾病、脑血管疾病、癌症和呼吸系统疾病)的死亡率,并能缓解老年人机能衰退的主要诱因,包括记忆力减退、抑郁和体弱。尽管其获益幅度看似有限(死亡率降低17%),但这相当于平均延长了1.8年的健康寿命。”
为什么 Muse 浏览器版本的默认字体这么小?
更难受的是,生成内容的时候页面一直晃,根本没法边生成边看,只能等全部输出完再读。不知道他们自己的工程师平时用起来是什么感受,ChatGPT 就没有这个问题。
18
1
762
“AI 技术本身并没有发生什么出人意料的危险转折,但在似乎是一场精心策划的公关宣传推动下,围绕它的炒作激起了相当大的恐惧”
The loudest voices stoking fears about AI dangers have made tremendous headway in the past two weeks. AI technology has not taken some unexpected, dangerous turn, but the hype around it — propelled by what appears to be a well orchestrated PR campaign — has drummed up considerable fear. I worry that it represents a setback for our field.
I have written frequently that fears of AI are overhyped. AI’s capabilities can be uncannily human-like and unpredictable, and it’s rational to worry when people who are directly involved express concerns. But I see the problems as a sign of the engineering work that ahead, rather than insurmountable barriers or the sky falling. AI technology continues to advance — which is a good thing! — but technical advances, poorly understood by the public, give those who seek to generate hype repeated opportunities to do so.
First, I don’t see any step up in the risk of human extinction from AI compared to a few months ago. The theories about this remain the same fantastical, science fiction scenarios as a few months ago. The biggest change in AI risk is its cybersecurity capabilities — a topic which we should take seriously — but this, too, will not lead to the end of the world.
The most notable recent event leading to increased fear was when an OpenAI team deployed an agent swarm that hacked into Hugging Face. Much of the popular press contained significant hype. For example, some publications reported that a swarm of 1,200 agents carried out the attack. While this was technically accurate, as I write this, I have about 1,300 processes running on my laptop. Yes, the ability to get large swarms of agents to work in parallel on a task is a significant technical advance, And, in computing, many processes run at the same time. So this shouldn’t be seen as some magical capability.
Additionally, OpenAI’s buggy sandboxing and monitoring processes were key to enabling this incident. Fixing these bugs and putting in place improved monitoring would be appropriate fixes, not pausing AI. There are many well known ways to attack software systems. The main advantage of AI agents is that they are relentless. They will tirelessly try many tactics — and have the patience to chain vulnerabilities together — that previously would have taken an infeasible amount of human effort. But in the long term, I believe the advantage will lie with defenders (because they have more information with which to identify bugs, which they can fix), but the cyber-threat landscape has changed significantly. There are still bottlenecks to identifying and exploiting a vulnerability. AI agents still have to try a lot of things to see what works, and taking these actions takes time and might be detected by defenders. This is why, even though it is now easy to obtain versions of leading open weight models that have had their guardrails removed or weakened, so they will not refuse to try to execute cyber attacks, the world has not ended.
I am also concerned about the anthropomorphization of AI in a lot of reporting, where LLMs and agents are unnecessarily treated as if they were people. If I wield a hammer, miss a nail, and accidentally dent the wall, it’s not the fault of the hammer. The problem lies in how I used the hammer. Similarly, if I prompt an agent and it hacks into someone else’s system, the responsibility lies with me, not the agent.
Of course, we want to build systems that are as safe and predictable as possible. (For example, an unsafe hammer would be one whose head randomly flies off under normal use.) Today’s agentic systems are not predictable, but I see no reason why, by applying sound engineering practices, we won’t be able to make them extremely safe to use. One new element in the forecasts of AI-enabled doom is AI companies disclaiming responsibility for their own products. “I didn’t do it; my out-of-control agent did!” There’s a balance to be struck between the responsibility of the tool maker and the tool user, but when something goes wrong, let’s hold the people building and/or using the hammer responsible, rather than the hammer. (By the way, if you’re worried about AI bioweapon risk, David Bellamy has a great post on why this, too, is overhyped. Briefly, the bottleneck in building a bioweapon is not intelligence, but lab work and manufacturing.)
Pausing AI progress will create much more harm than benefit. First, our adversaries will certainly not slow down. Second, engineering requires discovering problems empirically so we can fix them. If we pause AI by a decade, we will also delay finding and implementing safety engineering fixes by about the same duration.
Of course, the incentive to stoke fears — for regulatory capture, to garner attention, or to make one’s technology seem more powerful — remains the same as before. Disclaiming responsibility is a new one. Taking a hard technical look at the actual risks however, I see little factual basis for the degree of fear that’s been stoked up. We still have hard research and engineering work ahead to improve AI safety, but the beneficial applications continue to vastly outweigh the risks, and we should keep building.
[Original text (with links): deeplearning.ai/the-batch/is… ]
1
486
Mark 说是大家做好自己的安全就行了,不对齐的模型会失去用户,pace frontier 什么的是废话,我们延迟上个发版就是在这么做;另外只有少数几个领先厂商是危险的(让我上牌桌)
Last month I wrote about how we can build a positive and safe future for everyone: meta.com/thefutureisforevery…
Every lab has the responsibility and incentive to move at the pace required to train its models safely, and the ability to take its own actions to ensure that happens.
The reality is:
- People won't want to use agents that are misaligned with them and that don't do what they ask, so labs have a strong natural incentive to make their models more aligned.
There is a lot of debate about slowing progress on capabilities until alignment catches up. My view is that trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models. Any lab that doesn't focus on alignment will fall behind.
- Labs face significant liability if their models cause harm, so they have a strong incentive to prevent this as well.
Meta delayed shipping Muse for several months to focus on safety and security. We didn't call for everyone else to do this before we would. We just did it as part of our day-to-day work because it was clearly the right thing for people and for us. I'm proud of the security foundations we've built.
- Engaging independent evaluators and advisors is industry best practice. MSL already does this today in several areas because it helps produce better work. Other labs can just do this too. In general, it would be helpful for there to be a larger and more diverse ecosystem of evaluators.
- Committing the significant majority of compute towards serving people rather than racing towards recursive self-improvement is one of the best ways to ensure we develop this technology safely. Meta has made this commitment and other labs can do this as well.
I believe the key to building a positive future for everyone is maintaining the right balance of power. This is within our power to do.
632
相比于 Dario 昨天的版本似乎没太多新意,主要目的还是希望制定规范时候自己来牵头,“We hope that other companies will learn from our approaches and propose their own”
The world deserves confidence that American companies developing increasingly capable AI will act responsibly, especially as the trajectory of progress has steepened. Every frontier lab must deliver on this, and there is no reason any of us should come to work if we cannot.
We welcome a federal framework that sets consistent safety requirements for frontier AI. But we do not believe we need to wait for an anti-trust exemption or legislation to begin the work of providing this confidence. Consistent rules to manage frontier risk so that we can maximize the benefits are a good idea (and we are excited by ideas like independent auditors).
Years ago, companies like ours developed things like Responsible Scaling Policies and Preparedness Frameworks. Those were good for that moment, and focused primarily on the deployment of completed models, not what happens during their development process.
Today's shift to focusing on safe development and evaluation will need new tools. For example, at OpenAI we now formulate explicit safety cases in advance of frontier reinforcement learning runs we expect to significantly increase capability, in addition to the safety work we have long done in advance of model releases.
We hope that other companies will learn from our approaches and propose their own; we think shared standards for misalignment, monitoring, and safety will lead to better outcomes. We look forward to collaborating with our colleagues across the industry to formulate the best version of these.
When we talk about “pacing”, we do not mean “stopping”. Progress has been rapid and will continue to be. But it should be slower than it otherwise could be; interventions like safety cases and monitoring have significant costs.
Pacing will be well worth this cost; no amount of American competitive pressure should justify recklessness, or let capabilities get ahead of alignment and monitoring.
Where we will need the help of our government is for international coordination. But first we should do what we can ourselves.
666
David Sacks 认为前几天的 Jacob Coxon 一个接近空白的账号吹哨是 Anthropic 策划的活动(包括前 15 分钟的传播数据分析都看起来很像一个策划好的公关活动),以建立一个有利于自己的监管框架。从今天 Dario 发起提案的舆论走势来看,似乎有几分道理……
BIG EPISODE 🚨: AI Kills Everybody or Doomer Psyop?
-- AI Researcher's Doomsday Warning Goes Viral
-- OpenAI’s Math Breakthrough: AI Training Privacy and Agent Swarm Innovation
-- Nike’s $200B Collapse: How Business Mistakes and Corporate Wokeism Killed an Iconic Brand
(0:00) Bestie intros!
(0:35) AI Doomsday: Valid concern or Doomer psyop?
(33:40) Steelmanning AI Doomsday scenarios and the impact on Anthropic's IPO
(58:45) OpenAI says it cracked a 200 year old math problem, but did it front-run users with their own data?
(1:19:55) Nike removed from S&P 100 after stock tanks 80%: What went wrong? $NKE
1
699




















