Reading

The cost structure of agent applications differs from single-turn question answering. A shopping assistant completing one recommendation may undergo multiple steps—searching, reading results, narrowing scope, verifying constraints, and submitting an answer—in a multi-step loop, with each step constituting an independent billed call to a frontier model. At scale, this per-token billing loop makes deployment unviable. The solution offered by the AWS blog is sequence-level knowledge distillation (SeqKD): have the frontier teacher model generate complete multi-step trajectories, use a benchmark grader to filter for trajectories that actually solve the task, and then perform supervised fine-tuning on the small model using these trajectories. The teacher only needs to output text, requiring no weights, logits, or reinforcement learning infrastructure, so closed-source API models can also serve as teachers.

The most practically valuable finding in this blog is an ablation experiment: given the same 426 training problems and the same filtering pipeline, the only difference was whether the teacher's reasoning text was retained in the training objective. Removing reasoning and keeping only tool calls dropped the average absolute success rate from 0.541 to 0.415, a difference of about 13 percentage points. The authors explain that reasoning is not decorative narration but a causal chain connecting observations to next actions. Agents need to decide what to search for, how to filter, and when to terminate at each step, decisions driven by reasoning. If the student learns only the sequence of tool calls without the underlying reasoning, it achieves only partial transfer—preserving reasoning is the key to bringing the student close to the teacher's level.

However, SeqKD has a clear capability ceiling. The blog notes that what determines whether distillation succeeds is not the parameter gap between student and teacher, but whether the task itself is within the student's capability range. The teacher GLM-5.3 has approximately 40 billion active parameters, while the student Qwen3.5-4B has only 4 billion—a tenfold gap—but the shopping task is simple enough for a 4B model, so the student almost fully recovers the teacher's quality. Conversely, if the task exceeds the student's original capability, distillation will only produce long reasoning chains that are correctly formatted but contain wrong answers; trajectories lengthen while accuracy remains flat. The authors' recommended response is: upon seeing this symptom, stop adding data, either switch to a larger student or reduce task difficulty. Furthermore, SeqKD plateaus at the hardest tail of multi-step loops because the student cannot see the states entered after its own mistakes; breaking through this bottleneck requires on-policy methods, making SeqKD itself merely the first stage.

From a cost structure perspective, the significance of distillation extends beyond simply shrinking the model. Frontier APIs charge per token, requiring payment for every step of every loop, with costs growing linearly with traffic. The distilled 4B student is deployed on self-hosted endpoints, billed by instance hour, where increased traffic incurs no extra charge until scaling out is required. The authors do not provide specific per-task cost figures because those depend on your request patterns, sequence lengths, and concurrency levels, requiring self-measurement. Meanwhile, IFEval shows the student's general instruction-following capability dropped by about 6 to 8 percentage points, which is the expected cost of specialization—if the drop is significantly larger, it indicates the distillation narrowed the model too severely.

Since 2000, a total of 557 commercial banks in the United States have failed, with only 28 of them closing on any day other than Friday. Friday is the preferred closure day for the Federal Deposit Insurance Corporation (FDIC), and acquirers are typically arranged before the failure occurs. This means that modern bank failures are rarely triggered by market forces such as deposit runs, but instead follow supervisory closure decisions. For comparison, between 1865 and 1935, before the establishment of the federal deposit insurance system, bank failure dates were distributed almost uniformly across the days of the week—the researchers believe that at that time, depositors and shareholders usually determined the timing of the closure themselves.

But there is an easily overlooked feedback loop here: the basis for supervisors to close banks is book capital (the CET1 ratio), and book capital itself is partly produced by supervisory examinations. The researchers found that the probability of a bank revising its previously submitted financial statements (call reports) sharply increases around on-site supervisory examinations; over the five years preceding a bank's failure, its equity-to-assets ratio typically declines by 8 percentage points, nearly 3 of which come from revisions triggered by examination-induced loss recognition. In other words, supervisors force banks to write potential losses into their statements through examinations, and then use those worsened figures to decide on closure. In the CAMELS rating system, the capital sub-component is mechanically related to book capital, while the management sub-component allows for supervisory discretion—for well-capitalized banks, the composite rating is closer to the management rating; for severely undercapitalized banks, the composite rating is almost equivalent to the capital rating.

The researchers further validated the causal direction using a natural experiment during the 2008 financial crisis: under identical financial conditions, banks assigned to 'stricter' examiners recognized more losses and reported lower book capital, thereby increasing their probability of failure. Stricter supervision indeed reduced the FDIC's resolution costs, but also led to a short-term contraction in credit supply. This constitutes a core trade-off: strict supervision can curb speculative 'gambling for resurrection' behavior and prevent zombie banks from misallocating credit, but it may do so at the cost of further tightening supply when credit conditions are already weak; lenient supervision may preserve credit but raise ultimate resolution costs. The deposit insurance system weakens depositor discipline, making this trade-off even more prominent.

The researchers explicitly state that their analysis is entirely empirical and they make no normative judgments regarding the 'optimal degree of regulatory strictness.' The applicable boundaries of the conclusions are also worth noting: the natural experiment sample is limited to the 2008 crisis period, and the definition of 'strict supervisors' relies on examiners who assign worse CAMELS ratings for the same financial condition—a difference that may partially reflect subjective judgment rather than objective standards. For investors, the value of this research lies in providing an actionable analytical framework: when observing an abnormally high frequency of a bank's financial statement revisions, this is not only a signal of deteriorating asset quality, but also possibly a process where supervisors are actively creating the conditions for closure.

In an investment letter released on November 2, 2026, a16z partners Raghu Raghuram, Martin Casado, and Jamie used a quote from former Netscape CEO Jim Barksdale as their starting point: "The main thing is to keep the main thing the main thing." They argue that most current debates in the robotics industry revolve around training data, model architecture, and hardware types, but there is only one main line—getting robots to complete human tasks. Tutor Intelligence is a company designed around this main line, which is also the core reason they led its Series B round.

Most robotics companies follow a path of building a general model first, then finding application scenarios. Partners point out that this requires massive upfront data collection and capital investment just to build even a basic model. Tutor Intelligence took the opposite approach: identifying real needs and specific scenarios first, deploying hardware into production environments before having a 'robot brain,' and having human teleoperators (whom they call 'tutors') remotely control robots to complete work. Partners stated that these operators were so skilled that customers derived real value from them and could not distinguish between work done by employees and work done by tutors.

Every real-world deployment generates data. Tutor reverse-builds models from this deployment data, enabling robots to gradually gain autonomous capabilities. Partners note that every robot put to work makes the next one smarter, cheaper, and easier to deploy. The data for model training is no longer generic samples pre-collected in laboratories, but operational data truly needed in customers' actual production environments.

In terms of business model, Tutor charges by the hour rather than following the traditional robotics industry practice of multi-million-dollar upfront procurement and implementation fees. Partners believe this shifts capital expenditure to operating expense, unlocking the market for small and medium-sized manufacturing enterprises—where much of American manufacturing exists—and also serving some of the largest logistics companies. According to the investment letter, Tutor's robots are already operating in some of the largest manufacturing and logistics networks in the United States, with plans to have contracts covering over 1,000 robots by the end of 2027; in the past 12 months, nearly one-third of customers who received Tutor robots were existing clients installing additional robots; deployment implementation time is as low as 8 hours.

Tutor currently has two robots: Cassie, which has a single arm and handles case picking and palletizing/de-palletizing workflows; and Sonny, which has two arms and greater dexterity, serving item-by-item picking in e-commerce fulfillment. Both robots share the same model, which is part of Tutor's vertical integration across models, hardware, and data collection. Partners believe this full-stack capability enables Tutor to evolve toward more generally capable robots.

It should be noted that all the above operational data—deployment time, customer repurchase ratio, target number of contracted robots, and claims that operators are indistinguishable from humans—are sourced from self-reports by Tutor Intelligence and restatements by a16z as the lead investor, without independent verification. The investment letter itself includes a disclaimer stating that a16z has not independently verified the information therein. The core of the partners' argument lies in the logic of path selection: whether deploying first and modeling later can continuously produce superior data quality and commercial speed compared to modeling first and deploying later still needs to be tested in larger-scale deployments.

On November 2, 2026, in an interview on The Verge Decoder podcast, Matt Strauss, chairman of the NBCUniversal Media Group, rarely publicly disclosed the specific methods the company uses to evaluate content rights investments. His core formula is not complex: Customer Lifetime Value equals Average Revenue Per User (ARPU) multiplied by the number of months a user remains retained before churning. However, Strauss emphasized that the real difficulty lies in embedding this formula into a portfolio framework—treating every piece of content as a distinct asset within an investment portfolio, separately evaluating whether it drives acquisition, engagement, frequency, or retention, without requiring every individual title to be independently profitable.

He compared this approach to a mutual fund: "Some content may be loss leaders, but it adds value elsewhere in the portfolio." This was precisely the decision logic NBCU applied when its WWE contract came up for renewal. Six years ago when Peacock launched, the team led by Strauss chose to bring in WWE for three reasons: monthly premium live events establish the habit of opening the app; the massive library provides content depth; and the addressable market for WWE was calculated to have sufficient penetration potential. The key judgment was whether the lifetime value of subscribers acquired through WWE could cover the cost of the rights.

When renewal fees increased significantly, NBCU did not feel forced to follow suit. Strauss said they calculated a different question: if we do not renew, how many subscribers can we retain? The answer was the majority—because WWE viewers had already been habituated to other content; they enjoy the Fast & Furious franchise, Yellowstone, and other programs on the platform. Consequently, NBCU abandoned the high-cost renewal, retaining only a much smaller package of WWE programming, and reallocated the saved funds to other parts of the portfolio.

This case reveals Strauss's overall thinking on content expenditure: it is not about "is this title worth the price," but rather "what role does this title play in the portfolio, and where will users go if it is removed." He admitted that the fact that WWE fans like Fast & Furious was initially an assumption that needed testing and verification—"We have assumptions, but we must test these hypotheses."

Strauss put forward a judgment as early as 2019: the streaming market would move from "great unbundling" to "great rebundling," ultimately replicating the economic structure of cable television—dual revenue streams of advertising plus subscription, broad content for all household members, and bundled pricing. He said at the time that Peacock was not built to win the streaming war, but to build growth after the war. By 2026, the market reality he described closely matched this prediction: almost all streaming services have launched ad-supported tiers, 80 percent of Peacock's subscribers are on the ad tier, and the proportion of advertising revenue coming from streaming and sports has risen from about 30 percent six years ago to nearly 60 percent.

As Anthropic prepares for its initial public offering (IPO), one of the most attention-grabbing figures is not revenue, but "$40 trillion." In an analysis published on November 2, 2026, Kim Song-hyun, editor at The Core (South Korea), noted that in the S-1 registration statement filed with the U.S. Securities and Exchange Commission (SEC), Anthropic stated that the economic value of work performed by global knowledge workers exceeded $40 trillion last year. While Anthropic did not explicitly label this figure as a "Total Addressable Market" (TAM), the article argues that it effectively transforms the value of productivity gains AI could deliver into a business opportunity within investment narratives.

What truly warrants close examination here is the shift in market definition. Historically, companies have typically defined their markets based on what they sell—for example, estimating the spending scale around current products and services. Anthropic’s approach differs: it takes human work that AI can replace or augment as its starting point. The economic value of knowledge labor—including development, research, accounting, legal services, and office work—is included in the potential market. AI companies are no longer just selling model API calls or software seats, but rather execution capabilities that enter enterprise workflows.

The Core compares this shift to previous public listings. Companies like Snowflake estimated markets largely based on existing product spending; Airbnb and Uber incorporated the economic activity connected by their services into their market definitions; and in discussions related to its listing, SpaceX presented a market opportunity where AI accounted for a significant portion. Anthropic’s $40 trillion figure appears more aggressive because it uses the "value of human labor" itself as the market boundary. The larger the boundary, the grander the story may seem, but it also risks obscuring a key distinction: the creation of economic value does not equal a company’s ability to capture revenue.

The article illustrates this distinction with a simple example: if AI helps a company save 10 billion South Korean won in costs annually, the price the company pays for AI services may be far lower than 10 billion won. A significant portion of those saved costs might become profits for the adopting company or be passed back to customers. Therefore, even if AI has a substantial cumulative impact on the global economy, how much revenue Anthropic captures depends on pricing, competition, customer budgets, and switching costs. The Core also cites IDC forecasts suggesting that investments in AI solutions and services could generate a cumulative impact of $22.5 trillion on the global economy by 2031, but this remains a macroeconomic productivity metric, not Anthropic’s order book.

From the operational numbers disclosed in the S-1 analysis, Anthropic’s growth is indeed rapid. The article states that Anthropic’s 2025 revenue was $4.585 billion, while Q1 2026 revenue reached $4.731 billion, already surpassing the full-year 2025 total. Q2 2026 revenue was $11.555 billion, bringing the first-half total to $16.286 billion—approximately 14 times the same period last year. Meanwhile, the company still reported an operating loss of approximately $693 million in the first half, but the loss rate dropped significantly from about 184% in the prior year’s first half to roughly 4%. In Q2, GAAP operating profit was approximately $974 million, with an operating margin of about 8.4%.

Alphabet delivered a seemingly contradictory performance report for the third quarter: total revenue reached $127.6 billion, up approximately 25% year-over-year; Google Cloud revenue doubled to $30.7 billion, with an operating margin rising to 36%. However, on the day following the earnings release, the stock price plummeted by approximately 13%, erasing about $500 billion in market value in a single day—the largest single-day drop since its IPO. The market was not punishing growth itself, but rather two structural issues occurring simultaneously: AI infrastructure investment caused free cash flow to turn negative for the first time, and the monetization capability of the AI search product itself is lower than the traditional search it is replacing.

First, looking at the cash flow level. Alphabet generated $42.3 billion in operating cash flow in the third quarter, but capital expenditures surged to approximately $51.4 billion during the same period, resulting in a free cash flow deficit of $9.07 billion, exceeding market expectations. The company maintained its full-year capital expenditure guidance of $195 billion to $205 billion, with the vast majority directed toward data centers, servers, networks, and self-developed TPU chips. Investments are expected to remain high next year as well. This is a typical 'invest-first, harvest-later' structure: revenue is growing, but cash outflows precede revenue generation.

What truly unsettled the market was the second layer of issues. In the third quarter, Google Search advertising revenue was approximately $63.3 billion, up 12% year-over-year, with the growth rate slowing compared to the previous quarter. Chief Business Officer Philipp Schindler acknowledged on the earnings call that searches generated in AI Mode still have lower monetization levels than traditional search. This statement transformed a long-standing concern among investors into a fact confirmed by management: Google is redefining its core advertising product with AI, and the new product's ability to generate profit has not yet caught up.

The mechanism is not complex. The monetization model of traditional search is highly mature—users input keywords, relevant ads are displayed at the top of the results page, and search terms with clear purchase intent hold extremely high value for advertisers. AI search, however, provides an integrated, complete answer. There is currently no clear solution for where to place ads, how many to show, or how to do so without damaging user experience. More critically, AI answers themselves require computing power consumption, which further compresses the profit margin per search.

This creates a dilemma: failing to expand AI search means users will be diverted to conversational products like ChatGPT and Claude; expanding too quickly directly erodes the advertising revenue with the highest profit margins. Editor Kim Yoo-min of The Core summarized this situation as the paradox of AI search—defensive expansion and self-cannibalization are happening simultaneously.

The reason the market reacted so differently to Alphabet and Microsoft lies in the different pressure structures they bear. Microsoft is also investing heavily in AI data centers, but its third-quarter capital expenditures combined with finance leases totaled $45.2 billion, below market expectations, and free cash flow remained positive at $18.6 billion. More importantly, Microsoft does not have a core revenue line as weighty as Alphabet's search advertising that is being directly rewritten by its own AI products.

The American online community Reddit is closing a door it has maintained for approximately twenty years. Steve Huffman, CEO of Reddit (United States), announced that the platform will pilot mandatory login requirements in select countries outside the United States over the coming months. Posts that were previously viewable without registration will now require users to create an account. A global rollout timeline and the interface displayed to non-logged-in users who click on post links have not yet been determined. Huffman acknowledged that this represents a significant personal shift for someone who has long valued an open internet, but he emphasized that it is a necessary decision for the platform and its business.

The primary force driving this decision is data collection by AI companies. Reddit hosts a vast accumulation of user-generated product reviews, life experiences, and professional discussions, which hold immense value for AI firms seeking to learn from authentic human language and experiences. The issue lies in the fact that some enterprises and automated tools scrape posts on a large scale without permission. Reddit has filed lawsuits against Anthropic and Perplexity, alleging unauthorized access to data, and has restricted access to Old Reddit. Huffman stated that most unauthorized data access occurs when users are not logged in; limiting access for non-members makes it more difficult for automated scrapers to operate—a judgment made by the company itself, though its actual effectiveness remains to be seen.

However, Reddit does not intend to sever its data relationship with AI companies entirely. In 2024, Reddit signed contracts granting data access rights to OpenAI and Google for post data. The core distinction in strategy is that unauthorized scraping is blocked, while data supply under formal contracts continues. User conversations that were once freely available to everyone are transforming into commercial assets that AI companies must pay to acquire.

This transition coincides with changes in Reddit's revenue structure. Financial reports for the third quarter of 2026 showed a significant decline in traffic originating from search engines, particularly among non-logged-in users and general website visitors. Huffman revealed that since October, average weekly visits driven by search engines have decreased by over one million. The spread of AI search services may be a contributing factor—previously, users would search on Google and then directly visit Reddit posts to find answers; now, AI synthesizes information from multiple websites and provides answers directly within search results or chat interfaces, reducing the necessity of visiting original pages. However, Reddit itself has not attributed the traffic decline solely to AI search.

A more critical economic reality is that Reddit's revenue does not depend uniformly on all visitors. According to the company, app users and daily active logged-in users contribute over 80% of overall monetization. Compared to anonymous visitors, logged-in users are easier to target with precise advertising and are more likely to return to the platform repeatedly. For Reddit, the commercial value of maintaining a core group of logged-in users far outweighs simply expanding the total number of visitors. Mandatory login therefore serves two goals simultaneously: preventing unauthorized data outflow and converting more users into monetizable logged-in identities.

A key challenge in applying reinforcement learning to the real world is directly utilizing pre-collected datasets without interacting with the environment. However, standard off-policy reinforcement learning methods often suffer from value overestimation due to distribution shift between the dataset and the learned policy, particularly when handling complex and multimodal data distributions. To address this issue, researchers from the University of California, Berkeley, and Google Research proposed the "Conservative Q-Learning" (CQL) algorithm. The core idea of this algorithm is to learn a conservative Q-function such that the expected value of the policy under this Q-function is lower than its true value, thereby preventing overly optimistic estimates caused by out-of-distribution actions.

CQL achieves this goal by adding a simple Q-value regularization term to the standard Bellman error objective. This method not only theoretically proves to produce a lower bound on the current policy's value but can also be integrated into policy learning processes with theoretical improvement guarantees. In practical applications, CQL requires adding fewer than 20 lines of code to existing deep Q-learning and actor-critic implementations. Experiments show that in both discrete and continuous control domains, CQL significantly outperforms existing offline reinforcement learning methods, typically improving final returns by 2 to 5 times, with particularly outstanding performance when learning from complex and multimodal data distributions. For researchers and engineers looking to learn effective policies from static data without interacting with the environment, the paper "Conservative Q-Learning for Offline Reinforcement Learning," published in 2020, provides a robust and easy-to-implement solution.

In the 1977 Berkshire Hathaway shareholder letter, Warren Buffett delves into the correct standards for measuring management's economic performance. He criticized most companies for simply defining "record" earnings as a new high in earnings per share (EPS). Buffett pointed out that because companies typically increase their equity base year after year, if a company's equity capital increases by 10% while EPS grows by only 5%, such management performance is not particularly noteworthy. He argued that a more appropriate measure is the return on equity capital (Return on Equity). In 1977, Berkshire's operating return on beginning-of-year equity capital was 19%; although EPS grew by 37%, the beginning capital also increased by 24%, making the EPS growth appear less impressive than it seemed on the surface. This viewpoint emphasizes that investors should focus on the efficiency of capital usage rather than merely the growth of absolute profit figures.

Additionally, Buffett elaborated on the philosophy of insurance business operations in the letter, particularly regarding the importance of "tailwinds" versus "headwinds." He noted that in the insurance industry, even if some major mistakes are made, overall performance can still be quite satisfactory; whereas in the textile industry, even exceptionally excellent management yields only meager profits on average. This reveals a core investment principle: choosing businesses in favorable industry environments (tailwinds) is more critical than relying solely on superior management capabilities. He cited an example where the entire industry raised prices significantly in 1976 to offset disastrous underwriting results from 1974-75, making 1977 a beneficiary year. However, he also warned that as inflation and social inflation cause costs to rise by approximately 1% monthly, if rates cannot increase at the same pace, underwriting profits will shrink. Therefore, identifying turning points in industry cycles and adjusting strategies accordingly is crucial.

Finally, Buffett discussed the advantages of indirectly holding quality businesses through stocks. Using his investment in Capital Cities Communications as an example, he pointed out that directly purchasing the assets owned by the company might cost twice as much as the stock market investment and would not provide the same level of management. Through non-controlling stock investments, Berkshire could obtain the operational results brought by excellent management while avoiding the burden of control responsibilities. This strategy of "non-controlling interests" allows investors to enjoy long-term value created by outstanding management without interfering in daily operations. This concept challenges the traditional assumption that control inevitably leads to better outcomes, demonstrating that minority equity investments can be an effective way to obtain high-quality commercial returns at the right price.

For instance, suppose an investor is evaluating two technology companies. Company A has had an average annual EPS growth of 15% over the past five years, but its net asset scale expanded by 20%; Company B has an average annual EPS growth of 8%, but its net assets expanded by only 2%. According to Buffett's logic, despite Company A's faster EPS growth, Company B has higher capital efficiency, and its improvement in return on equity better reflects the true economic performance of management. Investors should be wary of companies that drive EPS growth by issuing large amounts of new shares or accumulating retained earnings, and instead look for enterprises that generate higher returns on a smaller capital base.