Prediction Markets Analyst: https://xitter.cf/t.co/aYTBwX1Qz4 | MACRO CAL: https://xitter.cf/t.co/UvpAfaJQZp 播客：https://xitter.cf/t.co/o03KpdaxKa 公众号：预测市场那些事儿
Joined September 2024
- Tweets972
- Following641
- Followers5.2K
- Likes1.5K
polymaster retweeted
我们爱猫，爱咖啡（从靠咖啡续命开始），是热爱一切静谧与美好事物的创作者。
我们也曾在高强度的工作压力卷过扛过，深知在这个充满不确定的时代，年轻的灵魂藏着怎样的紧绷与不安。我们懂得那份能让人慢下来、沉进去、进入心流的宁静有多宝贵。
于是我们就想要拿起画笔，用代码在数字的世界里构建一隅让思绪落定的静谧角落：
也许是一间微风轻拂、醇香漫溢的咖啡小屋，
也许是一座饱浸融融暖阳、盛满花香与硕果的花房；
又或是一湾浮光跃金、翻涌着鲜活渔获的潺潺清溪。
又或是一路静影沉璧、掩映着幽芳奇木和群山远岫的幽幽曲径。
待到溯流而上的远方， 更有巍峨雪山横绝天地，迢递高路直入青云……
你准备好来坐坐了吗？
#ArtistQ #IndieDev #CozyGames #GameDev #专注
polymaster retweeted
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
polymaster retweeted
我们是 ArtistQ。
用插画和游戏，留一处温柔。
敬请期待。
私たちは ArtistQ です。
イラストとゲームで、やさしい居場所を。
近日公開。
We are ArtistQ.
We create illustrated games for gentler moments.
Coming soon.
#ArtistQ #IndieGame #ComingSoon #FocusTimer
polymaster retweeted
The US may be decoupling from direct trade with China, but it is not decoupling from “Made with China.”
Over the past several years, Washington has repeatedly raised tariffs on Chinese goods, aiming to reduce its reliance on Chinese manufacturing and shift supply chains back to the US or toward other countries.
Today, I came across an excellent analysis that looks at the issue from a different angle. Instead of counting only direct US imports from China, it also measures the Chinese components, equipment and services embedded in goods imported from other countries.
Here are the key takeaways:
1/ US tariffs have reduced direct imports from China, but they have done far less to reduce America’s actual dependence on Chinese production.
The supply chain has changed its route, not necessarily its source.
2/ From 2017 to 2024, China’s share of direct US imports fell from about 18% to 11%.
But when Chinese components and services embedded in third-country exports are included, China’s share fell only from 17.7% to 15.4%.
3/ By 2024, the Chinese value added contained in total US imports was 34% larger than direct US imports from China.
About 25% of this Chinese content reached the US through third countries, up from 16% before the trade war.
4/ Mexico and Vietnam are the two most important intermediary hubs.
Mexico carries the largest volume of indirect Chinese value added, while Vietnam has seen the fastest growth, especially in electronics, machinery, textiles and transport equipment.
5/ This is not all simple transshipment.
In many cases, final assembly has moved out of China, while Chinese components, machinery, capital and technical capabilities remain embedded upstream. China is moving deeper into global production networks.
6/ The report concludes that blanket tariffs are a poor tool for strategic decoupling.
They lengthen supply chains, raise costs and make Chinese dependencies less visible, without necessarily creating viable alternative suppliers.
Tariffs have changed the label on many US imports. They have not necessarily changed what is inside them.
polymaster retweeted
Hi @intel, I noticed that quite a few students were having difficulty running PyTorch models for the course I teach. After the session, I wrote a longer message to students discussing computer hardware in general.
While looking into suitable options, I found that one of the most affordable devices currently available in Australia with the Intel Core Ultra X7 358H is an ASUS barebone NUC, which is obviously not very student-friendly.
After giving it some more thought, could @intel consider something along the following lines?
1. Create an Intel online experience store featuring selected small-form-factor PCs, laptops, handhelds, and other Intel-based devices from different brands.
2. Let customers configure storage based on their needs and budget—for example, lower-cost SSD options or higher-end Phison solutions designed to support very large AI models.
3. Negotiate volume Windows licensing with Microsoft so customers can purchase Windows 11 at a lower cost, while also offering Ubuntu as a no-cost alternative.
4. Provide step-by-step video tutorials covering Windows/Ubuntu installation and practical use cases such as PyTorch, local AI applications like OpenClaw, and gaming with XeSS.
5. Offer robotics and AI development platforms based on Panther Lake through the same store.
6. Build an Intel community that engages students from secondary school through university.
The broader idea is to make Intel hardware easier to discover, configure, learn about, and build with—especially for students who may not know which hardware best suits their needs.
These are just some ideas. Below is what I wrote to students after class under the title “Computer Hardware in General”:
Hi, I noticed that some of you are having difficulties running the code in Studio 3. This could be due to issues with your computer or hardware setup. Later in the course, we will introduce Google Colab, which should make things easier. It is also probably the most economical option.
An alternative is to use a laptop or desktop with a capable iGPU or dGPU.
Here is the laptop I used in class:
notebookcheck.net/Refreshed-…
The most important component is the processor, the Intel Core Ultra X7 358H. The “X” here indicates that it comes with the Intel Arc B390 iGPU, which is a very capable integrated GPU. Its performance should be higher than the NVIDIA T4 GPU typically available in the free tier of Google Colab for many of the workloads we use in this course. The iGPU is the same as the one demonstrated here:
youtu.be/UU359ZLD5FU?si=GV3S…
Here is a list of products using the X7 358H chip:
scorptec.com.au/search/go?w=…
The cheapest option currently listed is the ASUS NUC 16 Pro Barebone Mini PC with Intel Core Ultra X7 and 32 GB RAM, advertised at around $1,755. However, you would need to purchase an NVMe SSD separately and install the operating system yourself. A 500 GB NVMe SSD may cost around $150. You would also need a Windows 11 licence if you want to use Windows. Alternatively, you could install Linux and avoid that cost. The total cost can add up, but it is still likely to be lower than that of a laptop with the same processor. That said, a laptop may be more versatile and convenient for everyday use.
You might then ask: since this is not a laptop, how could I use it while I am in class? One option is to leave the machine at home and remotely connect to it via SSH using VS Code.
Another alternative is to build a desktop computer. However, given current memory prices, I am not sure whether this is the most economical approach at the moment. In any case, here is a video I made previously if you are interested:
youtu.be/xknquSjqMmA?si=RjBb…
There are also suitable solutions from NVIDIA, Apple, and AMD. I have experience with NVIDIA products, although they can be relatively expensive. I have less experience with Apple and AMD solutions, so I cannot provide as much practical advice on those platforms.
For most students, I would recommend trying Google Colab first before purchasing new hardware.
polymaster retweeted
其实他真的很传奇，我还特地去了解了这位ceo，2025 年 3 月 18 日，英特尔请来了陈立武出任 CEO。我认为如今的intel你如果理解了这位新任CEO，那你就理解了英特尔反转故事的一半因为后面发生的所有事，几乎都和他这个人的能力与人脉直接相关。
他的履历有三点值得特别描述的，每而其中的每一点都正好补在英特尔的命门上：
首先他是顶级的「翻盘型」经营者。2009 到 2021 年他担任 Cadence（楷登电子，EDA 设计软件龙头）的 CEO，把这家公司彻底改造了一遍：任内营收翻倍以上、利润率大幅扩张，股价上涨超过 3,200%（公开资料）。这不是守成，这是实打实把一家公司带上新台阶的记录。
其次他还是半导体圈的「创投教父」。1987 年他创办了 Walden International三十多年来投遍了半导体、AI、先进计算领域的突破型公司，是这个行业最有人脉、最懂技术趋势的投资人之一；他还拿过 2022 年的 Robert N. Noyce 奖（半导体行业协会的最高荣誉）。
再加上他有别人没有的「关系网」。这一点是隐性的、却最关键。一个在半导体生态里深耕了三十多年、几乎和所有大佬都熟的人，意味着他能调动的资源、能搭上的合作是普通职业经理人想都不敢想的。
我一直相信Lip-Bu Tan 的翻盘记录和人脉网是目前 $INTC 能下活的关键
polymaster retweeted
最近美股📉很多人都开始说历史少有，如果熟悉美股过去二十年调整的规律就不会这么说了。这次调整幅度无论大盘还是个股其实远低于去年关税战，当时连英伟达都腰斩了，更不用说22年、20年3月了。去年2月底有梳理过过去二十年美股调整的规律，再加上过去一年内四次调整，25年11月、26年2月到4月初、26年6月前两周、以及最近7月那就全齐了。可以去看看这篇做个参考。
过去二十年美股历史上的调整幅度：昨晚上美股进一步大跌纳指此次最高点调整幅度-9%多，一月前有说今年美股阵风局应该会看到-15%左右的调整（不用那么精确），可以回顾下过去二十年美股调整的幅度来为当下做个参考：
1、在个人判断体系里美股调整幅度（以纳指为例、标普大部分时间幅度会比纳指小一点）：
-10%上下属于小级别的调整；
-15%～-20%上下属于中级别调整；
-25～-30%往上属于大级别调整。
2 、过去二十年里，大级别的调整有四次
1）22年初到11月，纳指-28%、标普-32%，主因通胀创四十年最高纪录，联储胡近十年最快加息节奏；
2）20年3月，一个月之内纳指和标普均是-30%多的大跌，主因疫情全球蔓延引发的流动性大冲击；
3）18年10月到18年12月底，纳指-25%、标普-21%，主因是美联储上上一轮加息最快最后的阶段和中美贸易战；
4）07年10月到09年3月，纳指-56%、标普-58%，主因还是金融危机。
3、中级别的调整：
1）24年7-8月，纳指-17%，标普-11%，衰退担忧、套息交易退潮冲击；
2）23年8-10月，纳指-15%、标普-11%，通胀反弹、国债回购危机；
3）15年7月-16年2月，纳指-20%、标普-15%，首次加息、科技股估值消化；
4）11年5673月-11月，纳指-20%、标普-23%，主因是欧债危机、美国经济增长放缓；
4、小级别（-10%）的调整，最近两次就是24年4月（纳指-10%），23年2月-3月，如果再往前看每年都会有一到两次。
5、其实当下美股的调整只是达到了小级别调整的范畴，看到有人开始喊金融危机了，如果对历史上每一次调整的幅度和成因很熟悉的话其实大可不必。现在还远没金融危机的地步，此次调整核心原因：
一是川普政策的“阵痛期”引发市场预期转向（从之前的“通胀走低经济走强“转向“通胀反弹下不去经济放缓停滞”的滞涨预期）；
二是 市场估值高位市场变得敏感，需要估值回归完成估值消化。
细看市场并没有系统性的风险：
当下金融压力指数还在低位没有飙升；
银行间隔夜担保利率sofr也是正常状态；
十年美债收益率还处于下行趋势；
前两天聊到市场进入极度恐慌状态可以买点（开第一枪），但是市场预期转向惯性还很大所以还会有进一步逢低买入的机会（分批买），现在接近-10%的调整幅度其实开始进入之前预期-15%的调整范围之内，分批买还是合适的。
polymaster retweeted
$INTC
"We intend to accelerate various of our manufacturing expansion projects, though the scale and pace ... will ultimately be dictated by the amount of committed demand for Intel 14A that we are able to obtain ... and design wins with potential SIGNIFICANT external customers."
Money statement, quite literally, from Intel 10-Q.
There is so much captured in this statement people told me I'd be wrong about....
$INTC
polymaster retweeted
Dear US government,
Since you've just blocked Fable and Mythos on critical national security grounds, here are some other tools that pose a similar threat to the American people:
- Microsoft Teams
- SAP
- Salesforce
- Jira
- Outlook
Please do what you must to save America 🇺🇸
‼️🚨 BREAKING: Amazon researchers snitched to the US government about jailbreaking Fable 5 and Mythos 5, forcing Anthropic to immediately shut down worldwide access.
A security export control directive from Commerce Secretary Howard Lutnick enforced the action.
Anthropic is fighting the directive and calls it a misunderstanding.
This isn't the first clash. The Trump administration had already tried to get Anthropic to pause the release of its latest models before this directive landed.
polymaster retweeted
关于这个领域（概率与结果市场）的 "原子化" 和打包交易，42 space 的 @Leozayaat 给过我一些启发。
他们的体育赛事盘把每一个具体比分的盘当作一个采购原子，UI 网格即所有的比分可能性矩阵，然后你会发现传统足球的几乎所有流行玩法（标准胜负平、总分盘、让分盘、BTTS、零封盘等）都可以通过这个事件网格的组合 "描述" 出来。也是异常精巧的 UI 设计。
其实在保险领域是类似的，也是把各种可能性具体揭示（或自己精算）了以后再进行组合打包售卖。很多情况下现实世界的 “风险” 是一种恶性事件组合的叠加态，真正的大买家也并不在意具体的其中一个事件，因为单一事件没有痛感，而是希望打包购买一整批对于组合结果与风险的描述。
Polymarket 最近会推出 Combo (组合概率) 市场。我对 Combo 市场的理解是，相当于 Polymarket 开始有可以自由打包和定价无限风险因素叠加的 “保险类资产” 的能力。
如果说以前我们经常聊 “预测市场接 sportbook” 动的是传统博彩业的蛋糕，那么现在预测市场创建 Combo 市场显然动的就是传统保险业的蛋糕。
有些生意就是要满足多个风险条件同时发生才没法做，比如对于一家演艺经济公司 or 足球俱乐部，可能要对冲各种天气/舆情/地区政局/演员球员身体状况/演员球员主动违约（跳槽 or 公信力受损）等多种因素组合，才能让自己的 PnL 高度平滑，以及财报好看。
之前风险概率头寸的定价权由保险业垄断，你只能找他们询价（即发起 RFQ 请求）——但如果卖饼的和定价饼的同一帮人，你必然要隐性承担巨额经济 fee。同时，他们愿意做你生意的门槛不低，因为市场中这些对冲需求高度定制化（每个人或企业的获利模式完全不同），每个资产包的构建都有传统执照、传统人力模式下的固定成本，所以几乎不会为小额需求者的服务——那么我们一般认为这种市场是低效的，具备较低定价意义的。
而预测市场的创新能力在于，预测市场开放出了一个“任何人可以持有一个具体的事件概率头寸”的关键能力。那么自然任何人也都具备将这些事件的风险因素进行原子组合并打包交易的能力。这也就是原先保险公司在做的事情。
对于这种高度定制化的市场需求，交易模式一般是用RFQ场外叫价。RFQ 是 Request for Quote 的缩写，中文通常翻译为“询价单”或“报价请求”。在商业和采购领域，这是一种标准的业务流程：企业（买方）为了购买特定的商品或服务，向一家或多家供应商（卖方）发出一份详细的清单，邀请他们提供具体的价格、交付条款和付款方式——保单采购也是类似的。
Polymarket Combo “保单” 也用的是 RFQ 场外叫价，订单簿是无法为这种定制化需求提供集中流动性的（我的读者应该了解，定价方式、定价能力不同的系统，自然也会存在套利空间），但订单簿胜在可以把单一原子的风险定价精确。
统一的定价力量是巨大的，链上的流动性虹吸意味着定价权的转移。博彩公司和保险公司两者的历史角色都可能从地区意义上高度分散的 “流动性构建者” 以及 “公允定价者”，变成转发预测市场全球统一流动性的 broker（即专注负责揽户）。
polymaster retweeted
We’ve identified a security incident that involved unauthorized access to certain internal Vercel systems, impacting a limited subset of customers. Please see our security bulletin:
vercel.com/kb/bulletin/verce…
polymaster retweeted
流动性本身就是 “即时交易的匹配效率“，传统期货做市商定价的是连续波动、可高度对冲的资产价格风险；而 Polymarket 做市商定价的是离散结果、对冲效率较低的单一事件风险。
期货做市商的应对能力会高很多，具有可分散、可对冲、可量化的特征，做市策略可以依赖统计模型与微观结构持续优化。
Polymarket 做市商所承受的逆向风险具有更强的不可对冲性与信息不对称性。其损失往往是跳跃式、不可连续调整的，而非像期货那样可以通过价格波动逐步修正。
polymaster retweeted
new improved polymarket docs are live!
docs.polymarket.com
mcp + agent skills included. drop the link into cursor, claude, whatever - your agent learns and builds on polymarket directly
next up: builder fees, sdks, and more
polymaster retweeted
Polymarket twitter is full of ppl constantly talking about building bots for 15 min crypto markets which imo are the most oversaturated and hardest markets to enter. I think people should be looking for inefficiencies they can quantify in other markets and build bots to exploit them because you will make a lot more in a market where u are the only one who has a good model to build a bot off of while everyone else tries to trade that market manually