每日信息简报｜2026 年 08 月 08 日

DAILY BRIEFING · 2026.08.08

把 AI 的能力叙事拉回证据、成本与边界

今天的线索横跨模型安全、智能体浏览器、生物设计、训练基准与劳动力市场：真正值得追踪的不是单个标题，而是能力、约束与验证如何相互咬合。

今日要点

OpenAI says it slowed Astra model development over security concerns

OpenAI 表示因安全顾虑放缓 Astra 的开发；报道聚焦其被认为具备更高网络能力风险的阶段。能力推进与安全评估在这里不是两个独立流程：部署节奏本身就是风险治理的一部分。

Responding to the next frontier of critical cyber capabilities

OpenAI 说明其对更前沿关键网络能力的应对方式。值得关注的不是单一门槛，而是评测、访问控制、监测与外部复核能否在能力变化时同步迭代。

Cloudflare launches Kitesurf, a browser built for AI agents

TechCrunch 报道 Cloudflare 推出面向 AI 智能体的浏览器 Kitesurf。若智能体要持续操作网页，身份、权限边界、可审计性与失败恢复会比“能否点击”更决定其实际可用性。

AI 与科技

This A.I. Just Created Viruses Not Found in Nature

纽约时报报道一套 AI 系统生成了自然界未发现的病毒序列。这类研究同时扩大生物设计的实验空间，也要求把筛选、实验验证与生物安全审查连成闭环，不能只看模型生成结果。

Beyond Next-Token Prediction: A Performance Characterization of Diffusion versus Autoregressive Language Models

Apple 研究比较扩散式与自回归语言模型的性能特征。它提醒我们：模型路线的差异应落到可测的延迟、吞吐、质量和资源成本，而不是停留在架构标签。

NVIDIA Blackwell Tops MLPerf Training 6.0 with Industry-Leading Scale and Performance

NVIDIA 发布其对 MLPerf Training 6.0 成绩的解读。基准可用于比较特定提交配置，但不能直接外推为所有模型、数据集或集群下的训练成本与速度。

值得深读

Employers unexpectedly cut 23,000 jobs in a sign of a wilting labor market

NPR 报道雇主意外减少 2.3 万个岗位，并将其置于劳动力市场走弱的背景下。单月数据噪声很大，但就业、工资与招聘意愿仍是判断宏观需求韧性的关键联动指标。

A new vaccine could vanquish a major cause of deadly diarrheal disease

NPR 介绍一种针对严重腹泻病因的新疫苗研究。公共卫生价值取决于临床证据、生产能力、价格和高负担地区的实际可及性，而不仅是候选产品本身。

科技爱好者周刊（第 407 期）：国家为什么需要开源软件？

阮一峰从国家与开源软件的关系切入。开源的长期价值不只在“免费使用”，还在于关键基础设施的可审计、可维护和人才生态能否避免单一供应商锁定。

Colossal Biosciences Said It Cloned Red Wolves. Is It for Real?

Longreads 追问一家生物科技公司的红狼克隆说法。物种保护中，“做出样本”与建立可持续、具遗传多样性的野外种群是两件不同难度的事。

快速浏览

Aeon · 2026-08-07Sasquatch science：保留原文，适合结合上下文继续阅读。↗ Planet Money · 2026-08-07Older workers aren’t retiring. Should they be forced to?：保留原文，适合结合上下文继续阅读。↗ 纽约时报 AI · 2026-08-07China’s Unitree Prices IPO in Bet Investors Are Ready for Humanoid Robots：保留原文，适合结合上下文继续阅读。↗ Machine Learning Mastery · 2026-08-07Identifying Token Costs Hiding in Your Agentic Loop：保留原文，适合结合上下文继续阅读。↗ The Verge AI · 2026-08-06Suno shares plans to combat spammy AI music：保留原文，适合结合上下文继续阅读。↗ TechCrunch AI · 2026-08-07Airbnb says AI is helping it ship features faster as it tests a new search function：保留原文，适合结合上下文继续阅读。↗ TechCrunch AI · 2026-08-07After Rippling blew millions on AI in months, it built an employee ROI tool：保留原文，适合结合上下文继续阅读。↗ Hacker News 精选 · 2026-08-07Assembly Hall of Shame：保留原文，适合结合上下文继续阅读。↗

@jukan05 · 请求账号 @jukan05 · 2026-08-08 00:18:30 UTC · 原创原帖：AWS Leadership: Cut CPU Resource Waste to Address Capacity Constraints Apparently, engineers now even have to wait days to secure CPU server capacity.↗ @jukan05 · 请求账号 @jukan05 · 2026-08-07 23:18:53 UTC · 原创原帖：I recommend checking out @zephyr_z9's comments on memory. Our Citrini Semi team posts commentary like this in the Citrini Substack chat.↗ @jukan05 · 请求账号 @jukan05 · 2026-08-07 23:09:17 UTC · 原创原帖：Citi cuts Micron price target on peaking memory prices Citi downgraded Micron, arguing that both DRAM and NAND prices will see QoQ gains decelerate over the next four quarters, with pricing peaking in the May (F2Q) quarter of next year. Specifically, Citi revised its QoQ DRAM price assumptions for the Aug/Nov/Feb/May quarters from 23%/10%/2%/0% to 23%/9%/2%/0%, and now models 2H27 down 3% HoH. NAND price assumptions were changed from 29%/9%/1%/0% to 29%/7%/0%/-1%, with 2H27 now cut to down 5% HoH. Citi cites capacity expansion at CXMT and YMTC as the driver of the DRAM and NAND price declines. US-listed memory names sold off sharply.↗ @sama · 请求账号 @sama · 2026-08-07 22:54:58 UTC · 原创原帖：astra is a powerful model and we are working to make it generally available. we do not think it is a good strategy to keep powerful models to a chosen few. given its cyber capabilities, we need a little big longer to do do this safely. but hopefully not too long!↗ @thsottiaux · 请求账号 @thsottiaux · 2026-08-07 22:09:44 UTC · 原创原帖：Somewhere on your phone you have the closest thing to magic we have shipped. It does things for you. All day if you want. Go try it.↗ @thsottiaux · 请求账号 @thsottiaux · 2026-08-07 21:46:56 UTC · 引用原帖：I feel Theo is in need of a reset 👀↗ @aleabitoreddit · 请求账号 @aleabitoreddit · 2026-08-07 20:21:30 UTC · 原创原帖：There we go, White House finally invests in more breath in critical minerals/materials. Amazing policy move, as a TLDR: - $WWR receives $25M (graphite) - $SRL (ASX) receives $400M (Scandium) - $FEAM receives $8M (Boron) - $HREE receives $4.8M (magnet rare earths) out of the public companies. With more private investments from $150 million into Niron Magnetics or $85 million into Standard Bauxite. It's literally spare change to the US gov, for ENORMOUS amount of downstream applications. More should be done with funding amounts to accelerate development and derisk supply chains (don't own any of the above, just support the policy move), but great announcement.↗ @Brad_Setser · 请求账号 @Brad_Setser · 2026-08-07 19:26:42 UTC · 原创原帖：It is rather shocking -- but there is no sign China's assault on the global auto market is slowing On track to vehicle exports of over 12m/ $200b this year, and passenger exports of 10-11m 1/↗ @Brad_Setser · 请求账号 @Brad_Setser · 2026-08-07 18:51:55 UTC · 引用原帖：Not really true, as Mark Sobel has noted -- The Exchange Stabilization Fund has $24b in rolling one day bills (basically cash equivalent), and it can borrow dollars against its SDRS if needed at the Fed see the balance sheet below↗ @Brad_Setser · 请求账号 @Brad_Setser · 2026-08-07 18:31:29 UTC · 引用原帖：Thanks. I realize that some of my answers were a bit technical, and really directed at those who live inside the currency markets. I hope others also find it useful.↗ @Brad_Setser · 请求账号 @Brad_Setser · 2026-08-07 17:48:07 UTC · 引用原帖：the Chinese July trade numbers were very strong; seems like export volumes continued to grow at a double digit pace; net exports are still propping up China's growth↗ @sama · 请求账号 @sama · 2026-08-07 16:29:18 UTC · 原创原帖：congrats to oklo for achieving criticality! (less than a year after groundbreaking)↗ @aleabitoreddit · 请求账号 @aleabitoreddit · 2026-08-07 12:55:47 UTC · 引用原帖：Some more TLDR takeaways from $AAOI ER: - They also flagged DSP and TIA bottlenecks, which is a positive industry read through for names like $MXL, then $SMTC and $MTSI. - For revenue growth, they're expecting more than $500 million revenue Q4, which is absurd growth. For reference: Q2: $191.9M Q3: ~$255M-$290M Q4: $500M+ By mid-2027: ~$1.41B quarterly revenue run-rate Always impressed by the revenue ramp from some of these optical players.↗ @michaelxpettis · 请求账号 @michaelxpettis · 2026-08-07 10:56:56 UTC · 原创原帖：According to a Pew survey, younger Amercians are far more favorably disposed towards China than older Americans.↗ @aleabitoreddit · 请求账号 @aleabitoreddit · 2026-08-07 08:50:01 UTC · 原创原帖：I usually make fun of sellside, but Rosenblatt has pretty goated channel checks on optical names, and their reports are one of the few I like talking about. $AAOI, $LITE, $SIVE, and the laser party has started to recover around the est. timing windows. eg. last month: "Multiple short sellers told them they will likely close their positions... late July and early August". They might have missed that they believed $AMD would be $AAOI's first CPO customer... since AAOI said they would be missing from first-gen CPO (maybe future gen?) But lot of their talking points about China CPO laser positioning (eg. 2-3 years behind) got corroborated word-for-word from AAOI earnings today.↗ @aleabitoreddit · 请求账号 @aleabitoreddit · 2026-08-07 08:27:45 UTC · 原创原帖：Unitree, China's humanoid leader, is set to IPO this month at a $9B valuation! (~$6.2B originally per March filing). Derivative markets are pricing open at an implied $36.04B MC, roughly ~4X listing price. $CCXI / Agility Robotics, the first US publicly listed pure play humanoid leader, is also going public later in Q4, at a ~$2.5B pre-money valuation. It will be interesting to see how markets price in America vs. Chinese Humanoid leaders.↗ @Brad_Setser · 请求账号 @Brad_Setser · 2026-08-07 08:26:39 UTC · 引用原帖：China's export boom continues. If export and import prices in June were both unchanged, export volumes are up 14% y/y while import volumes are up 2-3% 1/↗ @jukan05 · 请求账号 @jukan05 · 2026-08-07 07:26:42 UTC · 原创原帖：SK hynix: We are actively considering additional shareholder return measures to enhance shareholder value, with the details to be finalized and announced during the third quarter. $SKHY↗ @aleabitoreddit · 请求账号 @aleabitoreddit · 2026-08-07 07:26:37 UTC · 原创原帖：Earlier today, optical interconnect company Lumilens "emerges from stealth" at a $5.5B valuation, after raising $700m. Just as a recap from $POET / $SIVE Linkedin OSINT mapping I shared earlier, Lumilens was supplying a top 3 hyperscaler customer. So we got confirmation of that a month later, along with the material figure that Lumilens has a "multi-billion-dollar customer agreement". And just for reference, $POET has existing contracts with Lumilens ($50m purchase order, up to $500m) + contracts with $SIVE (depending on product lines). I actually took tiny positions in POET again after reading this news, since it finally gives them visibility into extremely well funded hyperscaler suppliers (on top of Poet's really large balance sheet). But TLDR: > Lumilens around similar (or even higher) valuations as Lightmatter/Ayar, bigger than markets thought with hyperscaler customers. > nice read through upstream on $POET (derisking) / $SIVE. > Great for optical valuations in general, seeing a company go to $5.5B in 2 years.↗ @michaelxpettis · 请求账号 @michaelxpettis · 2026-08-07 06:22:01 UTC · 原创原帖：Xinhua: "China's gold market saw divergent demand trends in the first half of 2026. Gold consumption was up 1.2%, with jewelry consumption falling 33.88% to 132.13 tonnes and gold bar and coin consumption surging 28.42% to 339.34 tonnes." This makes it pretty clear that Chinese gold imports belong more in the capital account than in the trade account.↗ @michaelxpettis · 请求账号 @michaelxpettis · 2026-08-07 06:13:16 UTC · 原创原帖：It is not easy to talk about what happened recently in Ceuta without getting yelled at, but this discussion by two Carnegie scholars seems both calm and disinterested.↗ @thsottiaux · 请求账号 @thsottiaux · 2026-08-07 06:12:28 UTC · 原创原帖：I meant to say "ghiblify", but I said "gimlify"↗ @aleabitoreddit · 请求账号 @aleabitoreddit · 2026-08-07 05:46:59 UTC · 原创原帖：Wow, there's gem after gem in $AAOI earnings for $SIVE + other laser player readthrough. 1. AAOI not meaningfully participating in CPO first-gen deployments. "We just can’t make enough of them [CPO lasers] to be involved in their current first-generation [CPO] deployments because there’s just not enough capacity. We have to prioritize our ability to make lasers for our own transceivers first" So first-gen CPO laser deployments, might even be narrowed down further for Western players like $SIVE, $LITE, $AVGO, and $COHR. And independent CPO CW laser capacity became more valuable if $AAOI had to turn away more customers... Also remember Trendforce was talking about $AMD singing CW LTAs? $COHR / $LITE have capacity signed with $NVDA ... I thought it would be $AAOI, but they might be out of the equation. $MTSI also doesn't look meaningful with early CPO participation (eg. no mention in ER aside from NPO, and removal from Ayar website). So I wonder who players like $AMD is going to go with for Helios (eg. Sivers + Ayar more likely candidate now)? Btw, this is not bearish AAOI because they have too much demand for optical transceiver business. Just more bullish on the existing few qualified CPO laser names that have capacity. 2. Demand imbalance and bottleneck for InP lasers / optical transceivers. AOI's CEO stated kinda supported that when they said: "The customer demand is 20%-40% higher" than expanded built out capacity. "We are getting this kind of demand from several big customers almost every week. Lasers are the biggest bottleneck right now for the transceiver business" 3. China being years away from having CPO DWDM specification lasers. I covered this earlier when looking at CPO competition from channel checks. But AAOI confirming that China is "easily at least two, three years or even longer from having CPO lasers is incredible tread through on defensibility Western laser positioning in the CPO laser chokepoint. TLDR: - High demand imbalance for CW lasers and optical transceivers. - $AAOI not in first-gen CPO due to capacity constraints (not exactly bearish AAOI because they have too much demand for their optical transciver business, but even better news for the few independent players with capacity coming online like $SIVE) - China years behind in CPO lasers.↗ @michaelxpettis · 请求账号 @michaelxpettis · 2026-08-07 05:44:55 UTC · 原创原帖：Li Daokui: "The overall judgment is that new drivers of growth have not yet been created, while the old growth engines of recent decades have already faded." "Over the past 20 years," he continues, "China’s rapid economic expansion was driven primarily not by household consumption—whose share of GDP was roughly 35 to 38 percent, excluding services consumption—but by two major engines. The first was the largest infrastructure buildout in human history. The second was the broad upswing in real estate." Many people are talking about this July 12 speech by Li Daokui, not because his analysis is especially new (many Chinese economists have been saying similar things privately, just as many foreign economists have been saying them publicly), but because he said it in such a public forum. Even as he insists several times that "China's future is very bright," as he probably must, the speech nonetheless suggests that we may be seeing a gradual shift in the policymaking consensus. To me, the most important part of the speech is his willingness to state clearly something that I don't think Beijing wants to hear. There is a hope among policymakers that over the next few years Beijing will manage a transition in which the healthy part of the K-shaped economy (advanced technology and high-value-added manufacturing) becomes sufficiently strong relative to the weak part (everything else) to pull the whole economy forward. In my opinion, this was always unlikely, even in a world willing to absorb China's attempt to expand its already-large global share of advanced technology and high-value-added manufacturing. In today's world of rising trade tensions, however, I think it is all but impossible. Only the United States, the European Union, and Japan are large enough and advanced enough to support China's manufacturing ambitions through their net demand, and all three are increasingly closing their markets. Li seems to agree that this "K-shaped dream" is unrealistic. He says: "Once we start talking about K-shaped divergence, it is as though we are placing our hopes on the upward arm of the K and expecting it to pull the entire Chinese economy forward. That actually leads us away from the real problem. The upper arm of the K cannot pull up the entire economic base. On this issue, we must maintain a strong sense of urgency." While I agree strongly with Li's analysis of the problem, however, I disagree with his solution, which is to use the central government balance sheet to clean up local government balance sheets so that they can return to their old mission of pumping up economic activity. "The crux of the problem," he says, "is the blockage, and the blockage lies in local-government debt. The answer is therefore obvious: the central government, and indeed the public sector as a whole, should expand its debt level." My problem with this proposal is that it assumes local governments invested badly for reasons that can be corrected. It further assumes that future behavior can be changed, perhaps by requiring local governments to promise to reform: "Eligible local governments could apply to the central government and sign a commitment promising to regulate future borrowing and refrain from excessive debt issuance, in exchange for replacing their existing high-interest debt." But that is not what happened, nor is it what will happen. Beijing has set excessively high GDP growth targets for decades. Because it could not get consumption growth to accelerate sufficiently, this meant that it also could not allow investment growth to decelerate. With the highest investment rates in history, achieving these growth targets required an enormous pipeline of economically viable investment projects. The problem is that, unlike the period before the end of the 2000s, China is no longer sufficiently underinvested across the board for huge increases in investment to generate enough viable projects and sustain real growth. China has had excess capacity in property, infrastructure, and manufacturing for many years. That necessarily means the returns on new investment have been very low or even negative. And because investment remained so high, debt had to surge – which, of course, it did. Not only has this problem not improved in recent years, it has become progressively worse. That is why local-government debt has risen so much faster since the early 2010s than the real value of provincial economies. That is why so many provinces are struggling under their debt burdens. It is not that local governments were careless or irresponsible when they made these investments. Rather, the GDP growth targets themselves forced them to continue investing, whether or not additional investment was economically justified. The inevitable result was an ever-growing volume of non-productive investment. Because provinces are now burdened by these debts, Li proposes cleaning up their balance sheets by transferring the debt onto the central government balance sheet. He is right that the central government still has a relatively clean balance sheet. It is, in fact, the last clean balance sheet left in China. But that is precisely why I think it should not be compromised in an effort to sustain GDP growth targets that have become increasingly unrealistic for an economy facing the kinds of structural problems Li himself describes. As long as Beijing continues to set excessively high GDP growth targets, there is no way to prevent debt from continuing to surge. My concern is that Li's proposal would simply cause China to lose the last clean balance sheet it still possesses and, with it, its ability to moderate and manage the pace of what is likely to be an inevitable – and inevitably difficult – adjustment.↗ @thsottiaux · 请求账号 @thsottiaux · 2026-08-07 05:23:31 UTC · 原创原帖：You can just ask Codex with GPT-5.6 Sol the wildest things and it will just do it. I talk to it for 5 minutes straight with things that just seem to require weeks of work, get up to get something in the fridge, pet the dog, come back and ... it's done.↗ @jukan05 · 请求账号 @jukan05 · 2026-08-07 04:53:55 UTC · 原创原帖：FT: BYTEDANCE IS IN THE PRE-TRAINING STAGE OF A MODEL WITH UP TO 10 TRILLION PARAMETERS, APPROACHING MYTHOS IN SCALE.↗ @jukan05 · 请求账号 @jukan05 · 2026-08-07 03:26:44 UTC · 原创原帖：This is insane. GCP is basically a money-printing machine. No wonder Google executives are choosing GCP over Gemini.↗ @Brad_Setser · 请求账号 @Brad_Setser · 2026-08-07 01:01:14 UTC · 引用原帖：Completely agree with Goldman's China/ Asia team -- Need two tools to move China closer to domestic and external balance. A stronger currency on the external side, and more fiscal support on the domestic side 1/↗ @Brad_Setser · 请求账号 @Brad_Setser · 2026-08-07 00:29:06 UTC · 引用原帖：Sander and I have always supported policies to strengthen European demand growth, for what it is worth. Much easier for those policies to raise German growth if Germany isn't losing a percentage point of growth due to net exports (rising real imports, falling real exports)↗ @Claudia_Sahm · 请求账号 @Claudia_Sahm · 2026-08-06 23:57:12 UTC · 原创原帖：#PuffiSays I’m tired of this week … me too, almost over.↗ @aleabitoreddit · 请求账号 @aleabitoreddit · 2026-08-06 23:41:38 UTC · 原创原帖：$MTSI earnings transcript: "Customers are coming to us with urgency due to the general supply shortage of indium phosphide DFB lasers" Emphasis on "Urgency", CW "DFB lasers", "coming to us". CW DFB names, such as like $SIVE and $LITE should be happy to hear about this laser bottleneck getting validated.↗ @jukan05 · 请求账号 @jukan05 · 2026-08-06 23:15:02 UTC · 原创原帖：Ibiden (largest substrate supplier for intel EMIB) — earnings call Q&A Q) What is the development status and technical difficulty of EMIB-T, and when does mass production start? A) We understand that one or two competitors are also developing the relevant technology. Ibiden has largely completed the technology-establishment phase and is about to enter the yield-improvement phase ahead of full-scale mass production, which we believe puts us ahead of competitors. The technical difficulty lies in the bonding of the silicon bridge and in the core formation process. The Gama plant will be brought online in stages starting at the end of FY2027 and will contribute to earnings from FY2028. If demand materializes before then, we plan to accommodate it using existing facilities. ---------------------------------------- Reads like a de facto confirmation of 2028 EMIB-T ramp. $INTC↗ @aleabitoreddit · 请求账号 @aleabitoreddit · 2026-08-06 22:54:46 UTC · 原创原帖：Just some TLDR notes on $AAOI earnings: - Expects full qualification of 1.6T products by their hyperscaler customer within next couple of weeks (helpful revenue ramp #2, timelines) - Continues to believe that AOI will have the largest AI DC transceiver production capacity in the US (reiterating ambitions during a time when their competitors might get banned) - Expects facilities toward InP capacity to come online in early 2027 (timeline FYI in terms of further ramp) - Total capacity is approaching 200,000/u per month, EOY 650,000/u per month of 800g/1.6t. EOY 2027, 930,000/u (this is the ramp i want to see) - "increase our manufacturing capacity for our external light source or ELSFP. That's for co-packaged optics or CPO". we anticipate ramping production later this year and into 2027, ultimately culminating in about 400,000 pieces per month in 2028 (need some time to model this into revenue) - "As we have mentioned before, we've been manufacturing lasers internally for many years. This has allowed us to avoid some of the shortages that have affected others in the industry" (vertical integration bull case during CW/EML laser shortages) - We believe that in the future, CPO will continue to drive increased demand for high-power lasers (thesis validation on CPO sector) - "to our long-term objective of returning non-GAAP gross margins to around 40%" - "We ended the second quarter with $508.8 million in total cash equivalents" I need to double check if the ATM finished or not - "our ability to deliver revenue in general, and specifically when it comes to 800G products, is limited by our production capacity right now" "If we could produce more, we could ship more right now" Demand > Supply validation. - "Most of the increased capacity will be in U.S. Even so, let me say that, like I keep emphasizing, that is not good enough for the customer demand. The customer demand is 20%-40% higher." Unholy photonics demand validation across the whole sector, read through for $LITE, $SIVE / $JBL, and others is amazing. - "Not in the next two, three years, especially the demand is so big. Okay? Even combined AI, $LITE, $COHR, $AVGO all together, it's still very tough to meet the customer demand in the next few years" More optical sector demand validation. - CPO Timelines: "If you're talking about really high volume manufacturer [for CPO market], I would say more like the late Q3 next year" and "We have been working very close with at least five customer" If you care about current earnings (which I'm not really looking at closely) Revenue: $191M vs. $190m EPS: $.06 vs. $.02 TLDR: Extraordinary demand across the laser + optical sector read through. Kinda supporting Lumentum CEO statement that laser shortage is worse than memory shortages. 2027 capacity ramp on track. To map inflection period with timelines, would be around early next year, as stated in their previous earnings call. AAOI has the customers now. Limitation is making enough lasers and transceivers.↗ @Brad_Setser · 请求账号 @Brad_Setser · 2026-08-06 21:41:55 UTC · 引用原帖：Maybe "keep calm and pay the cost of carry on the yen ... " Rate differential is still negative↗ @Claudia_Sahm · 请求账号 @Claudia_Sahm · 2026-08-06 21:02:54 UTC · 引用原帖：I agree with KGB Kitty. meow.↗ @sama · 请求账号 @sama · 2026-08-06 19:56:24 UTC · 引用原帖：5.6 Sol much better in chat now and unlimited text chat for free users!↗ @thsottiaux · 请求账号 @thsottiaux · 2026-08-06 18:29:02 UTC · 引用原帖：Agent Plugins. A standard for (most) of your agents out there, including Codex and ChatGPT.↗ @Brad_Setser · 请求账号 @Brad_Setser · 2026-08-06 16:48:06 UTC · 引用原帖：Among other things, I tried to note that the Japanese government is a very big part of the "market" -- MoF and GPIF are the biggest components of Japan's net intevement position (along with Toyota and the MNEs) 1/↗ @jukan05 · 请求账号 @jukan05 · 2026-08-06 14:56:10 UTC · 引用原帖：Tesla is posting job openings for memory process engineers for its Terafab. @GavinSBaker said that Elon is currently prioritizing DRAM above all else for the Terafab.↗

YouTube 更新

今日最值得看

Dwarkesh Patel：8 Predictions for the Era of Continual Learning

节目讨论持续学习时代的八项预测。预测属于嘉宾或节目的分析框架，适合与实际技术、成本与部署数据交叉对照。

Two Minute Papers：DeepMind Just Changed How AI Sees The World

节目介绍 DeepMind 相关进展及其对 AI 感知的含义。视频围绕“DeepMind Just Changed How AI Sees The World”展开，以下仅转述频道选题，不把标题或节目判断当作独立核验结论。观看时可重点留意其证据来源、演示条件与反例。

Matt Wolfe：AI News: The AI Stories Everyone’s Freaking Out About

节目汇总近期引发讨论的 AI 新闻。它适合作为线索入口；涉及产品能力、融资或风险的具体判断仍应回到一手资料复核。

其他频道更新

AI ExplainedAI is getting a little out of control
节目讨论 AI 发展的速度与外溢影响。视频围绕“AI is getting a little out of control”展开，以下仅转述频道选题，不把标题或节目判断当作独立核验结论。观看时可重点留意其证据来源、演示条件与反例。↗