Large Language Models (LLMs)

• The Bottom Line: Large Language Models are powerful AI systems that can generate human-like text, representing both a revolutionary technology and a potential investment bubble that demands extreme caution from a value investor.
• Key Takeaways:
• What it is: An LLM is a type of artificial intelligence trained on a colossal amount of text data, allowing it to understand, summarize, generate, and predict text in a conversational way.
• Why it matters: This technology is poised to disrupt nearly every industry, creating opportunities for massive productivity gains but also fueling the kind of speculative mania that value investors must avoid. It's a test of distinguishing a true long-term moat from fleeting hype.
• How to use it: A savvy investor uses LLMs as a powerful research assistant to speed up due diligence, but never as a replacement for their own critical judgment or as a reason to abandon the core principles of value investing.

Imagine an intern who has read every book in the Library of Congress, every article on Wikipedia, and billions of other websites. This intern doesn't truly understand the concepts in the way a human does, but they have developed an almost supernatural ability to recognize patterns in language. If you give them the beginning of a sentence, they can make an incredibly accurate prediction about which word should come next, and the word after that, and so on, until they've written a complete, coherent paragraph. That, in essence, is a Large Language Model (LLM). Think of it as autocomplete on a god-like scale. When your phone suggests the next word in a text message, it's using a tiny, simple model. An LLM like OpenAI's GPT-4 or Google's Gemini does the same thing, but it's been trained on an unimaginably vast ocean of data and powered by supercomputers. This allows it to do much more than just complete a sentence. It can:

• Answer complex questions in a conversational tone.
• Summarize a 100-page report into a few bullet points.
• Write computer code.
• Translate languages.
• Draft marketing copy, emails, and even poetry.

The key insight is that LLMs are masters of syntax and probability, not of truth or consciousness. They are pattern-matching engines. This distinction is absolutely critical for an investor. An LLM can sound incredibly intelligent and authoritative, but it has no concept of what is true or false. It is simply generating the most statistically likely sequence of words based on the data it was trained on. This power and its inherent limitations create a fascinating and dangerous landscape for investors.

“What the wise do in the beginning, fools do in the end.” - Warren Buffett
¹⁾

For a value investor, the emergence of LLMs isn't just another tech trend; it's a seismic shift that must be viewed through the unwavering lens of business fundamentals, competitive advantages, and, most importantly, price. We must separate the revolutionary nature of the technology from the investment proposition of the companies involved. This requires dissecting the impact into two categories: opportunities and threats.

• Productivity as a Moat-Widener: The most tangible, near-term value of LLMs will likely be found not in the flashy model-builders, but in established, high-quality businesses that use the technology to become more efficient. Imagine a software company that uses an LLM to handle 50% of its customer support queries, slashing costs. Or an insurance firm that uses AI to process claims in minutes instead of days, improving customer satisfaction and freeing up human capital for more complex tasks. For these companies, the LLM isn't the business; it's a tool to widen their existing moat by lowering costs and improving service. This is a classic value investing play: finding great businesses getting even better.
• The “Picks and Shovels” Play: During the gold rush, the most consistent fortunes were made not by the prospectors, but by the merchants selling picks, shovels, and blue jeans. In the age of AI, the equivalent is the companies that provide the essential, non-negotiable infrastructure. This includes designers of specialized semiconductors (like GPUs), manufacturers of those chips, and builders of data centers. While these companies can also become overvalued, their role in the ecosystem is fundamental and less speculative than betting on which specific AI application will win.
• Enhanced Research Capabilities: For the individual investor, LLMs can be a powerful tool to level the playing field. They can help you digest a dense annual report, summarize an earnings call transcript, or get up to speed on the competitive dynamics of a new industry. This allows you to spend more of your time on the most important part of analysis: critical thinking.

• The Hype Cycle and Mr_Market's Mania: LLMs have ignited a speculative frenzy reminiscent of the dot-com bubble of the late 1990s. Any company that mentions “AI” in a press release sees its stock price soar, often with no connection to underlying revenue or profit. Mr_Market is in a euphoric mood, and he is pricing these stocks not on current reality, but on a perfect, rosy vision of the future. A value investor's job is to ignore this siren song and remain anchored to intrinsic_value. Paying 100 times sales for a money-losing company is not investing; it is pure speculation.
• Intense and Capital-Destroying Competition: The race to build the most powerful LLM is being fought by some of the largest companies in the world: Microsoft, Google, Amazon, and Meta. They are pouring tens of billions of dollars into research, talent, and computing power. This kind of arms race, while producing amazing technology, is often terrible for shareholder returns. The end result could be a commoditized market where the models themselves are cheap or free, and profits are scarce.
• The Competence Trap: Benjamin Graham warned us to stay within our circle_of_competence. The inner workings of LLMs are incredibly complex. It is difficult for a non-expert to judge which company has a genuine technological edge versus one that just has better marketing. This opacity makes it very easy to make a catastrophic investment mistake based on a compelling but flawed story.
• Erosion of Existing Moats: While LLMs can widen moats, they can also destroy them. A company whose business is based on organizing and selling access to information could see its entire model upended by an LLM that can provide similar answers for free. As an investor, you must now ask of every company in your portfolio: “How vulnerable is this business to disruption from generative AI?”

As a concept, LLMs are not something you calculate, but a framework you apply to your investment process. This can be broken down into two primary methods: using LLMs as a tool, and analyzing companies within the LLM ecosystem.

Think of an LLM as a brilliant but sometimes unreliable intern. Your job is to give it tasks that play to its strengths while diligently fact-checking its output.

1. Step 1: Information Synthesis. Feed the LLM the transcript of a CEO's latest interview or the “Management's Discussion & Analysis” section of a 10-K report. Ask it to:
   • “Summarize the key strategic priorities mentioned in this text.”
   • “Identify the main risks highlighted by management.”
   • “Extract all mentions of capital allocation plans (e.g., buybacks, dividends, acquisitions).”
2. Step 2: Competitive Landscape Mapping. When analyzing a company, like John Deere, you could ask:
   • “Who are the primary global competitors to John Deere in the agricultural equipment market?”
   • “What are the main product categories for both John Deere and its competitor AGCO?”
3. Step 3: Concept Explanation. If you're reading about a semiconductor company and encounter a term you don't understand, ask the LLM:
   • “Explain what 'EUV lithography' is in simple terms and why it's important for chip manufacturing.”
4. Step 4: CRITICAL VERIFICATION. This is the most important step. NEVER trust any specific number, date, or factual claim from an LLM without verifying it from a primary source (e.g., the company's official SEC filings). LLMs are known to “hallucinate”—that is, to confidently state incorrect information. Use it for summarization and brainstorming, not for hard facts.

Instead of trying to pick the one “winning” LLM, analyze the entire ecosystem, just as you would any other industry.

Your “result” from this analysis is not a number, but an investment thesis.

• A strong thesis might focus on a company in the Application Layer, like “Durable Software Co.,” which has a 20-year track record of profitability and is now using AI to reduce churn and increase prices, leading to predictable growth in free cash flow.
• A weak or speculative thesis would be, “AI-Hype Inc. is going to be the next big thing.” This is a story, not an analysis. It lacks any anchor to fundamental value.

From a value investor's viewpoint, the ideal outcome is to find a wonderful, understandable business whose long-term earnings power is being enhanced by LLMs, but where this improvement is not yet fully reflected in the stock price. This creates a margin_of_safety.

Let's compare two fictional companies to illustrate the value investing approach to the LLM trend.

• Company A: “NeuralNet Dreams Inc.” A two-year-old startup that has developed a new LLM. They are not yet profitable, burning through cash at a rate of $50 million per quarter. The media loves them, and their stock has risen 1,000% in the past year on the story of their “revolutionary” technology. Their valuation is $30 billion, despite having only $10 million in annual revenue.
• Company B: “Reliable Accounting Software Corp.” A 25-year-old company that provides essential accounting software to small businesses. They have a wide moat built on high switching costs, a trusted brand, and a massive user base. They have grown profits at a steady 8% per year for the last decade. Recently, they integrated an LLM-powered feature that automates expense reporting and tax preparation for their customers. They are charging 15% more for this new tier, and early adoption is strong. The stock trades at a reasonable 18 times earnings.

The Analysis: A speculator or trend-follower is immediately drawn to NeuralNet Dreams. The story is exciting, the growth potential seems infinite, and they are afraid of missing out. A value investor, however, sees massive red flags. The valuation is detached from reality. There is no history of profitability. The competitive landscape is brutal. It is impossible to confidently project future cash flows, and therefore impossible to determine the company's intrinsic_value. There is no margin_of_safety; instead, there is a “margin of danger.” The value investor is drawn to Reliable Accounting Software. This is a business they can understand. It has a proven track record and a durable competitive advantage. The LLM is not a speculative bet; it is a logical, value-adding extension of their core product. The investor can reasonably project how this new feature will increase revenue and profits, calculate an updated intrinsic value, and determine if the current stock price offers a sufficient margin of safety. They are investing in an evolution of a great business, not gambling on a revolution.

• Transformative Economic Potential: The ability of LLMs to automate tasks and augment human intelligence is real. This will create genuine economic value and productivity gains, which will ultimately benefit the shareholders of well-positioned companies.
• Moat Fortification: For companies with existing competitive advantages (e.g., proprietary data, large user bases), LLMs can be a powerful tool to make their moats even wider and more formidable.
• Powerful Research Tool: When used correctly, LLMs can democratize investment research, allowing individual investors to process information more efficiently than ever before.

• Valuation Risk: This is the single greatest danger. The narrative surrounding LLMs is so compelling that investors are willing to pay prices that assume decades of perfect, uninterrupted, exponential growth. Any hiccup can lead to a catastrophic collapse in the stock price. This is a classic violation of the margin_of_safety principle.
• The “Black Box” Problem: The complexity of these models makes them fall outside the circle_of_competence for most investors. It is difficult to independently assess the quality or defensibility of a company's technology.
• Commoditization: The core technology of LLMs may become a cheap, widely available utility, much like cloud computing. If this happens, the enormous profits that investors are currently pricing in may never materialize for the model builders. The value will flow to the companies that use the utility most creatively.
• “Hallucination” and Unreliability: As a research tool, an LLM's tendency to invent facts is a critical flaw. An investor who blindly trusts its output for financial data is setting themselves up for disaster. Always verify.

• durable_competitive_advantage: The central question is whether LLMs create new moats or just temporary technological leads.
• circle_of_competence: Acknowledging the complexity of AI is key to avoiding costly mistakes.
• margin_of_safety: The most important principle for avoiding permanent loss of capital when investing in a hyped-up sector.
• speculation: Contrasting the act of investing in a solid business using AI with gambling on an unproven AI story.
• mr_market: Understanding how market sentiment can swing from euphoria to despair, especially with new technologies.
• intrinsic_value: The necessary anchor to prevent being swept away by exciting narratives and soaring stock prices.
• return_on_invested_capital: A crucial metric for evaluating companies in the capital-intensive AI arms race.