eBay's new DeepAffinity paper shows a small, fine-tuned model beating general-purpose LLMs at forecasting long-term shopper preferences, a lesson in specialization over scale.
A new benchmark shows published fraud-detection results may not survive contact with real transaction data — and what that means for fintech and DeFi ML systems.
Most search teams wonder whether they matched the query. Not necessarily the correct metric it would seem
A new Bosch study shows RL cutting truck dispatch costs up to 23% — but only within the demand patterns it was trained on. Here's why classic OR solvers still hold the line everywhere else.
A new paper shows that 28-69% of accounts flagged as "at risk" by typical churn models were never actually churning - the model was just confusing seasonality with decline.
ReasonCast gates every promo or holiday signal through a relevance check before it touches the forecast - and the biggest win came from knowing when to leave the forecast alone.