Goldman Sachs Exchanges
2026-05-11 · Hosted by Allison Nathan · Goldman Sachs
Executive Summary
Goldman Sachs’ Allison Nathan and George Lee speak with Osman Ali, global co-head of Quantitative Investment Strategies (QIS), about how generative AI is reshaping investing. Osman argues the technology is evolving long-standing sentiment-analysis techniques while simultaneously making markets less efficient as widespread use of similar models creates herd behavior, crowding, and predictable mispricings — counter-consensus to the assumption that AI will tighten price discovery. He also stresses that data, technology and human experience together are what preserves an edge in a zero-sum game.
Key Stories & Changes
1. AI Is an Evolution, Not a Revolution, in Quant Investing
Goldman’s QIS team (~100 people globally) has a 37-year track record dating to the late 1980s.
The team has been doing sentiment analysis since around 2008 using earlier “bag of words” classifiers; LLMs are the latest step-up, not a discontinuity.
LLMs allow fine-tuned, language-specific extraction (e.g., Japanese-language management disclosures) that wasn’t possible 10-20 years ago.
Osman: more than 50% of what drives a stock’s return over the next 12 months is not the fundamentals but market sentiment, themes, and trend exposure — making language models particularly valuable.
2. AI May Make Markets Less Efficient, Not More
Inefficient corners (small caps, merger arb) may become more efficient as LLMs surface mispricings.
But in mainstream markets, widely available LLMs are expected to produce herd behavior because models tend to give the same answer to the same question, driving investors into the same securities.
Osman’s counter-consensus view: this will create more alpha opportunity, not less, because crowding drives prices away from fundamental value and creates predictable reversion patterns.
3. The Edge Equation: Data + Technology + Experience
Investing remains a zero-sum game; not everyone can outperform.
Three required ingredients: proprietary data (Goldman cites 37 years of curated data), scalable technology, and human context/experience to know what questions to ask.
Team size has stayed roughly flat despite AI adoption — “absolutely not smaller,” net-net slightly larger.
4. Market Structure Is More Complex
Modern equity markets mix passive investors (price-indifferent), retail (sentiment-driven), hedgers, and active alpha-seekers.
This cocktail means clearing prices are increasingly driven by technicals rather than fundamentals, creating predictable inefficiencies for data-driven investors to exploit.
QIS analyzes 15,000 stocks every single day.
Trends Identified
1. Technicals Over Fundamentals
Osman acknowledges drivers of stock returns have skewed toward market technicals and away from underlying business fundamentals over the past decade. Sentiment, theme exposure, and crowding now matter more — and LLMs are accelerating that shift by giving the average investor more uniform views.
2. Democratization Without Equalization
Even as powerful models become widely available, they don’t level the playing field. The edge migrates to whoever holds proprietary data, scaled infrastructure, and the contextual investing know-how to ask the right questions. Egalitarian tools, asymmetric outcomes.
3. The Rise of Autonomous Investing Machines
Headlines about fully autonomous AI-driven investing point to a new asset class. Osman’s view: this amplifies herd behavior because identical models give identical answers, creating mispricings exploitable by investors who model the investor psyche itself.
4. Career Implications
Future investing careers favor a combination of data-science fluency AND experience/context — Osman’s advice for those entering the field is to seek organizations that take data, technology, and experience seriously together, not one in isolation. —-
Sentiment Analysis
Overall Market Sentiment: Cautiously Constructive / Analytical
The conversation is forward-looking and analytical rather than directional on near-term markets. Osman is bullish on the opportunity set for sophisticated quant investors and confident in the persistence of human-machine combinations.
Risk Factors Highlighted
Herd behavior from AI uniformity: Widely used models give similar answers, causing crowding and prices to detach from fundamentals.
Zero-sum constraint: Mathematically impossible for everyone to outperform; democratized tools don’t change this.
Data cost: Proprietary data exists but is “may not be cheap” — capital intensity remains a barrier.
Complexity of drivers: With many forces (passive, retail, hedgers) driving price action, distilling signal from noise is harder.
Predictable reversion as a risk to the herd: Investors crowding into identical AI-driven trades face reversion risk modeled by sophisticated players.
This episode was covered in today’s The Market Signal — 2026-05-11, a cross-source synthesis of multiple podcast reports.