Key Takeaways
- • Numerai, a US hedge fund that crowdsources stock prediction models from data scientists worldwide, reportedly released stock names for a dataset it had kept encrypted for roughly 10 years, according to the primary source dated October 16, 2026.
- • The new dataset, called 'Unity,' reportedly contains 1,040 features alongside real stock identifiers, allowing users to track securities by industry and region and to add their own external inputs such as financial statements and news.
- • The source reports that connections via AI agents to Numerai exceeded 30,000 sessions over the prior 30 days, which the company cited as the rationale for opening the data to agent-driven research.
- • The author describes a personal, low-cost experiment using an AI coding tool to test whether scoring earnings press release text improves stock prediction accuracy, with the emphasis on identifying approaches that do not work.
Summary
Numerai, described in the primary source as an unusual US hedge fund, builds its investment process around stock price prediction models submitted by data scientists worldwide, bundling the strongest predictions into its managed funds. The article notes that JP Morgan previously allocated a $500 million framework to the firm. Until now, the stock identifiers in the participant dataset had been encrypted, so outside researchers could run machine learning without knowing which securities each row represented.
On October 16, 2026, according to the primary source, Numerai released the new 'Unity' dataset with real stock names. The dataset reportedly includes 1,040 features. With identified securities, users can track stocks over time, categorize them by industry and region, and combine externally collected information such as financial statements and news as their own features. The source quotes the company's rationale that, in the age of agents, showing data is preferable to hiding it, and reports that AI-agent connections exceeded 30,000 sessions in the last 30 days.
The author, writing a first-person account, tested a single hypothesis: that an AI model reading and scoring earnings press release text would improve stock prediction accuracy. The setup involved opening a Numerai account, launching Claude Code, and submitting one instruction prompt. The article states that its stated purpose is to document what does not work, including failures and placebo verification, rather than to offer a method for profiting from Numerai. The author states that the piece is not an invitation to invest in specific stocks or services. Quantitative results from the experiment were not included in the portion of the article provided.
On October 16, 2026, according to the primary source, Numerai released the new 'Unity' dataset with real stock names. The dataset reportedly includes 1,040 features. With identified securities, users can track stocks over time, categorize them by industry and region, and combine externally collected information such as financial statements and news as their own features. The source quotes the company's rationale that, in the age of agents, showing data is preferable to hiding it, and reports that AI-agent connections exceeded 30,000 sessions in the last 30 days.
The author, writing a first-person account, tested a single hypothesis: that an AI model reading and scoring earnings press release text would improve stock prediction accuracy. The setup involved opening a Numerai account, launching Claude Code, and submitting one instruction prompt. The article states that its stated purpose is to document what does not work, including failures and placebo verification, rather than to offer a method for profiting from Numerai. The author states that the piece is not an invitation to invest in specific stocks or services. Quantitative results from the experiment were not included in the portion of the article provided.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.