AI is the New Consensus

Everyone’s racing to summarize listed company filings in seconds. Congratulations. So is every other fund on the planet.

Here’s the uncomfortable truth: AI that processes public information isn’t giving you an edge. It’s giving you consensus.

Public information is becoming an alpha desert.

Large language models are, by design, consensus machines. During training, they learn to predict the most likely next token based on patterns across billions of documents, effectively distilling human knowledge toward its statistical mean. This process systematically reduces variance: outlier perspectives, contrarian views, and unconventional reasoning are dampened in favour of what is most frequently represented in the training data.

In investing terms, this is the textbook definition of consensus thinking. But alpha requires being non-consensus and right. You can’t generate excess returns by thinking what everyone else thinks. LLMs trained on everyone’s collective output are architected to do precisely that.

This pattern of technology reducing investment edge has repeated for 25 years:

EDGAR made filings accessible → edge gone

Bloomberg made search instant → edge gone

Reg FD killed selective disclosure → edge gone

LLMs can now interpret an entire earnings season in minutes → edge gone

Every time information and technology becomes universally accessible, it stops being a source of differentiation.

The long term investors who outperformed after each wave weren’t the ones who read faster. They were the ones who had built something proprietary that couldn’t be downloaded, scraped, or summarised.

Your management meetings. Your site visits. Your industry calls.

Your non-consensus investment thesis.

That’s where alpha lives. Not in a faster public transcript summary.

—