Intelligent Alpha CEO Doug Clinton says the firm began as a test of whether ChatGPT could beat the S&P 500, then became a fund run by frontier models.
Doug Clinton says knowledge graphs will generate alpha for 12 to 18 months before commoditizing, and that firms should accept imperfection for speed because the paradigm shifts within months.
Doug Clinton says closed-source GPT models consistently lead on investment tasks, and Khe Hy says smaller funds will likely wait for a 20% performance gap before switching to open-source models.
AI Portfolio Management Rewards Speed, Proprietary Data and Human Taste
Intelligent Alpha lets frontier AI models analyze stocks and manage portfolios. Few stocks beat the index, so a median AI process will not capture the right tail by itself. The edge comes from a hybrid design, proprietary data and human taste, and it fades fast, so firms must keep rebuilding and avoid attachment to old systems.
Takeaways
Hybrid fundamental and quant design beats naive prompting. A plain request for 10 stocks mostly returns the largest S&P 500 names. Intelligent Alpha holds a couple hundred to 500 positions with a few conviction bets sized at several points. Human researchers add data and push large bets.
Proprietary data is the hardest, most valuable input. Doug splits data into public data, third-party consensus and LLM-augmented estimates, and proprietary feet-on-the-street research. Agents running channel checks are the long-term frontier. Current models rate about B plus as analysts and lack top human creativity.
Knowledge graphs are alpha for twelve to eighteen months. Doug expects well-built knowledge graphs to commoditize after that window. EBITDA can mean fifteen things to fifteen people, so firms must teach models their definitions. Human taste lifts the work from B plus to A, and many PMs cannot articulate their definitions.
Speed beats perfection, and sunk costs are dangerous. Whatever generates alpha today stops working once others find it, and the paradigm shifts within months. Intelligent Alpha has dropped systems within a month or two, and prefers building or modifying open source tools to control its own destiny.
Closed-source models lead on investment tasks. The IA 500 benchmark has ten top models each pick about 200 stocks, ensembled into roughly 500 holdings that beat the S&P 500 over the past year. Doug sees GPT leading and expects a lasting gap with open-weight models.
Large firms struggle more with organization than technology. Compliance and data sovereignty block big asset managers, yet are easier to solve technically than organizationally. Billion-dollar funds often have ten to fifteen investment staff, so in-house builds are heavy. Khe says firms need an AI-pilled leader willing to take governance risk.
Model portfolios inherit stale knowledge and bias. Brett's Fable test returned a concentrated AI portfolio, including 35% in three large platform names, 20% in Palantir and Spotify, and 10% in cash. Its knowledge ends around April 2026, so Doug would feed it current earnings, consensus and transcripts.
Advice for practitioners
Look six to eighteen months ahead and prepare for capabilities that will be unlocked.
Tap a name to go there · tap a line to see why they are linked
On this episode 3
Show 1
Learn the models by playing with a fun project first.
Large firms should operate like startups, since techniques Brett taught in fall 2025 are already obsolete.