Reading

The US-based financial podcast 'Excess Returns' released an episode titled 'The AI Gap Is Closing' on September 20, 2026, featuring Jason Hsu, founder and Chief Investment Officer (CIO) of Rayliant Global Advisors and co-founder of Research Affiliates. According to the publisher's show notes, Hsu discusses the current state of the US-China AI race, the economic logic behind AI spending, and what market concentration means for investors.

Specific topics listed include why Hsu believes AI safety requires cooperation between the US and China, how close Chinese AI models are to US developers, China's electricity infrastructure and the competitive threat from open-source AI, where AI profits could accrue across hardware, energy, models, and applications, and how chip restrictions are encouraging China to develop domestic capabilities. The episode also touches on opportunities and challenges for factor investors in China due to retail trading, as well as Chinese technology companies, dividend-paying state enterprises, and US-China trade.

For readers interested in AI industry structure and market risk, this episode specifically addresses the 'AI spending arms race' and the concentration risk facing S&P 500 investors. Hsu also shares insights on momentum crashes, value cycles, and how Rayliant uses machine learning to combine factors for building multifactor portfolios. While no full transcript is provided, timestamps indicate these discussions occur roughly between minutes 33 and 41.

This recommendation is based on the publisher's listed topics and guest identity, aiming to provide professional investors with an entry point to understand AI industry dynamics and market structural risks from a quantitative perspective. Please note that views expressed represent the guest's personal stance and do not constitute investment advice.

David Rosenberg, founder of the prominent macro strategy firm Rosenberg Research, recently returned to the US financial podcast 'Excess Returns' for a conversation with host Matt Zeigler. Recorded on September 16, 2026, before the Federal Reserve's policy announcement, the episode features Rosenberg explaining his bullish case for Treasury bonds, why he expects inflation and economic growth to slow, and the risks he sees in an AI-driven stock market, according to the publisher's show notes.

The discussion covers consumer spending, Federal Reserve policy, gold, international stocks, and how to translate economic outlooks into diversified portfolios. Notably, Rosenberg argues that AI exposure extends beyond technology stocks into utilities, industrials, and other sectors. He also views higher oil prices as a tax on consumers rather than evidence of sustained broad-based inflation, and discusses how slowing wage growth and falling savings shape consumer spending. The episode further explores opportunities in healthcare, consumer staples, pipelines, European stocks, and Asia, alongside his model portfolio's allocation to equities, bonds, cash, and commodities.

For readers interested in macro asset allocation and AI industry investment risks, this episode offers a window into how a seasoned strategist balances aggressive AI themes with defensive bond positions in a high-valuation environment. While no full transcript is provided, timestamps indicate detailed discussions on 'AI concentration risk beyond technology stocks' (starting at 22:53) and 'Betting against the inflation consensus' (starting at 36:43). Note that this recommendation is based on the publisher's listed topics and does not include unverified specific conclusions or predictions from the guest.

In the early days of natural language processing, words were typically treated as independent atomic units lacking semantic associations. In their 2013 paper "Efficient Estimation of Word Representations in Vector Space," Tomas Mikolov and colleagues proposed a new approach: using a simple model architecture to quickly learn high-quality continuous word vectors from corpora consisting of billions of words. This paper, authored by researchers at Google Inc. (United States), demonstrates that computational efficiency and representation quality can be achieved simultaneously.

Although traditional Neural Network Language Models (NNLM) could generate word vectors, their computational cost was extremely high, making it difficult to scale to super-large datasets. The new model architecture proposed by the Mikolov team removed expensive hidden layers, directly utilizing linear relationships between input and output layers to predict context or center words. This simplification allowed the model to learn vectors with good semantic relationships from a corpus of 1.6 billion words in less than one day. The paper notes that these vectors not only bring similar words closer together in space but also capture linear patterns such as "king - man + woman = queen."

For today's readers, the key to understanding this paper lies in how it breaks the myth that "complex models are required for good results." It showcases engineering wisdom in trade-offs: by sacrificing some model complexity (such as removing non-linear hidden layers), feasibility in processing massive data is gained. This idea directly drove the popularization of subsequent Word2Vec technology and became one of the cornerstones of pre-training stages in modern large language models. AI practitioners and NLP beginners are advised to read this paper, "Efficient Estimation of Word Representations in Vector Space" (2013), focusing on its analysis of computational complexity regarding model architecture and the design of the new log-linear model, to understand the turning point where word vector technology moved from theory to large-scale application.

In the 2012 letter to shareholders, Warren Buffett began by noting that although Berkshire achieved total gains of $24.1 billion, book value per share grew by 14.4%, slightly lagging behind the S&P 500. He candidly admitted this was the ninth time in 48 years that Berkshire had underperformed the index, emphasizing that such relative underperformance typically occurs during strong markets.

Buffett reaffirmed the core logic of his performance metric: management's responsibility is to increase "intrinsic business value" at a faster rate than the S&P 500, while book value is merely a proxy indicator that significantly understates true value. If this cannot be achieved, investors would be better off simply purchasing low-cost index funds. This perspective decouples investment performance from short-term stock price fluctuations, returning focus to the long-term accumulation of corporate fundamentals.

Addressing the regret over failing to complete a large acquisition in 2012, Buffett detailed the subsequent H.J. Heinz transaction as a case study. Berkshire partnered with Jorge Paulo Lemann, with each party contributing approximately $4 billion to purchase common equity in the holding company. Additionally, Berkshire separately invested $8 billion in preferred stock, which carried a 9% dividend rate and included warrants to purchase 5% of the holding company's common stock at a nominal price.

This transaction structure illustrates Buffett's unique mechanism for capital allocation: using preferred stock to secure fixed high yields and protect against downside risk, while utilizing warrants to retain upside potential. This design not only locks in cash flow returns but also embeds options for management incentives and ultimate control within the joint venture structure, reflecting his deep understanding of applying complex financial instruments in industrial acquisitions.

Suppose a modern investor faces a mature consumer brand with slowing growth. Looking solely at the price-to-earnings ratio might suggest it lacks appeal. However, referencing the thinking in this letter, an investor could focus on whether the enterprise possesses a moat capable of generating stable cash flows, and whether there is an opportunity to participate in its transformation or privatization through structured financing similar to preferred stock plus options, thereby gaining a margin of safety and potential explosive power in a low-valuation range.

The Berkshire Hathaway Shareholder Letter 2012 was written by Warren E. Buffett and published on March 1, 2013 (corresponding to fiscal year 2012). It is suitable for investors who wish to deeply understand the definition of "intrinsic value" in value investing and learn how to balance risk and return through structured transactions. It is recommended to start with the sections regarding performance benchmarks and the details of the Heinz acquisition.