arXiv cs.AI
Aug 12, 2026
How Organizations Use AI: Evidence from ChatGPT
Using privacy-preserving links between ChatGPT Enterprise account records, worker roles, task classifications, and public-company financial data through March 2026, the authors analyze a six-month sample covering more than 1,500 organizations and 17 million messages. They report that adoption is concentrated among larger, more valuable, R&D- and SG&A-intensive public companies, while use inside adopting firms spans functions and seniority and is most intense among early-career workers.
- The authors report a six-month-horizon sample of more than 1,500 organizations and 17 million messages.
- The authors report that adoption among U.S. public companies is concentrated in larger, more valuable, and more R&D- and SG&A-intensive firms.
- The authors report broad task and seniority coverage, with especially high usage intensity among early-career workers.
Why it mattersThis is unusually granular evidence on enterprise AI diffusion and workflow breadth, helping investors distinguish seat expansion and usage intensity from broad claims about adoption. The evidence is observational, tied to ChatGPT Enterprise customers, and cannot by itself establish productivity or financial returns.
arXiv q-fin
Aug 12, 2026
Regime-Gated Residual Mixture-of-Experts for Cross-Sectional Volatility Forecasting
The authors test five-day realized-volatility forecasts across 1,027 U.S. equities with a rolling walk-forward design and a separate Japanese panel. Their proposed model routes residual corrections through regime-aware mixture-of-experts gates; they report better accuracy, training stability, and Value-at-Risk calibration than a capacity-matched MLP, while directly appending regime variables to the forecast input degrades results.
- The authors report that their regime-gated residual mixture-of-experts model outperforms a capacity-matched MLP in the U.S. study.
- The authors report similar gains on a Japanese equity panel and improved Value-at-Risk calibration.
- The authors report that direct use of regime variables in the forecasting input reduces performance and stability.
Why it mattersThe study points to routing architecture—not just model size—as a potentially important design choice for nonstationary risk models. The result is a preprint-level claim from the authors and needs independent replication, broader benchmark comparisons, and transaction-cost-aware downstream testing before investment use.