Writer’s papers tested models with and without memory, finding that personalization can amplify sycophancy and cut accuracy by as much as 71% in some tasks.
The reported figures show why investors are paying attention: scale arrives before proof is complete, and the gap between promise and delivery will be judged through deployment, margins and regulatory reaction.
For readers, the practical point is to separate the headline valuation from the business mechanism. The winners will be the projects that convert AI capability into repeatable work, not just a larger launch narrative.
Sources：TechCrunch、CocoLoop、The Register、WRITER、Bitcoinworld.