Bridgewater's Greg Jensen on AI, Inflation and What Markets Are Getting Wrong

Episode Summary

Executive Summary: The episode centers on Bridgewater co-CIO Greg Jensen’s view that AI—especially large language models—will not simply replace investing, but can dramatically augment it by generating theories, helping reason over complex models, and then being checked by statistical methods and human judgment. The discussion also covers why AI struggles with hallucinations, market regime shifts, and reflexive/adversarial markets, plus Jensen’s macro view that inflation and rates may stay stickier than markets expect.

Main Topics: AI vs. machine learning vs. quant investing (Priority: 5/5): Jensen distinguishes legacy expert systems, statistical machine learning, and modern large language models, arguing that Bridgewater’s approach combines human intuition, algorithms, and newer AI tools rather than relying on pure prediction from data alone. How Bridgewater plans to use AI (Priority: 5/5): Bridgewater is building an AI-focused venture/lab with dedicated staff and a future AI-run fund, aiming to reinvent parts of its investment process using LLMs to generate theories, statistical models to test them, and reasoning layers to reduce errors. Limits of LLMs in investing (Priority: 5/5): Jensen says off-the-shelf LLMs are weak at precision and can hallucinate, making them unsuitable as direct stock-pickers; they are better used as theory generators and explainers when paired with rigorous validation. Reflexivity, adversarial markets, and data limitations (Priority: 4/5): The conversation emphasizes that markets differ from games like chess because the environment changes as participants learn, data is limited, and models can fail when they don’t anticipate how their own actions alter outcomes. AI and changing work inside finance (Priority: 4/5): Jensen argues AI will reshape analyst roles by making people more like flexible generalists, changing coding and knowledge-work workflows, and boosting output rather than simply eliminating jobs immediately. Macro outlook: inflation, rates, and recession risk (Priority: 5/5): Jensen gives Bridgewater’s updated macro view: inflation is proving stickier, the Fed may need to stay tighter for longer, growth is slowing more gradually than expected, and markets are too optimistic about an easy return to target inflation. Examples from poker, chess, Go, and Zillow (Priority: 4/5): The discussion uses AI successes and failures in chess, Go, poker, and Zillow’s housing model to illustrate both the power of machine learning and its blind spots when facing complex, changing, or adversarial settings.

Key Arguments: Bridgewater’s edge historically came from translating human intuition into durable, testable algorithms and datasets, not from blind statistical pattern-matching. Large language models are valuable because they approximate reasoning and can scale theory generation, but they are unreliable as standalone decision-makers. Statistical models are precise about the past but weak about the future; LLMs are the opposite, so combining them can offset each other’s weaknesses. AI models can fail in adversarial environments because they do not inherently understand underlying principles or account for how markets respond to their own behavior. The best use of AI in investing is as a force multiplier: generate hypotheses, challenge them, test them, and loop in human oversight. Market regime shifts are hard for both humans and models to detect, so AI out of the box will not reliably identify turning points. The post-COVID macro environment changed balance sheets, savings behavior, and fiscal impulses enough to make higher rates less immediately contractionary than in past cycles. Markets are overly optimistic about how quickly inflation will return to target and how easily the Fed can engineer a soft landing.

Data Points: Bridgewater AI lab size: 17 people - Jensen says 17 people are working on the AI reinvention effort, with him leading it. Dedicated AI staff: 16 others are 100% dedicated - He notes that 16 team members besides him are fully focused on reinventing Bridgewater with machine learning. GPT-3.5 benchmark at Bridgewater: First-year investment associate level - He says ChatGPT 3.5 could answer Bridgewater’s internal investment-associate tests at about the level of a first-year IA. GPT-4 benchmark: Significantly better than average first-year investment associate - He says GPT-4 performed materially better than GPT-3.5 on conceptual tests. AI performance band: Around the 80th percentile on many tasks - Jensen describes current LLMs as impressive but not perfect, roughly 80th percentile across diverse exams and tasks. Fed funds rate: About 5.5% - Tracy references the policy rate as evidence of how far tightening has gone. Kroger CEO AI mentions: Eight times - Tracy notes a Kroger earnings call where the CEO mentioned AI repeatedly to illustrate how broadly the term is being used. Podcast report length: Five minutes or less - Promotional copy for Bloomberg’s Stock Movers describes the format of those reports. Bridgewater history: Since the 1990s / 2012 / 2016-17 - Jensen references Bridgewater’s long evolution with expert systems, Dave Ferrucci’s 2012-era work, and OpenAI involvement around 2016-17.

Pivotal Quotes: "If somebody is going to use large language models to pick stocks, I think that's hopeless." — Greg Jensen: He explains why raw LLM outputs are too imprecise to be used as direct stock-picking engines. "You have to use it correctly and not misuse it in order to try to generate that answer." — Greg Jensen: On regime-change prediction, he says AI can help only if it is embedded in a broader analytical process. "The best use of AI in investing is a force multiplier." — Joe Weisenthal: Joe summarizes the episode’s central takeaway about AI’s role in finance.

Implications: AI is likely to reshape investing as a co-pilot for research, theory generation, and model testing—not a turnkey stock picker. Firms that combine LLMs, statistics, and human judgment may gain a major edge, while markets may remain more inflationary and policy-constrained than investors expect.