Marchés Financiers

December 15, 2025, 3:22 PM EST. Stanford researchers warn against an AI bubble bursting with hype while acknowledging real progress. The outlook for AI in 2026 is more realistic: modest gains in many tasks, with both efficiencies and new frictions as the ecosystem matures. In medicine, a self-supervised revolution could slash labeling costs, enabling high-quality training on large, private datasets while preserving patient privacy. This may trigger a new ChatGPT moment for biomedical AI, with foundation models that improve diagnosis and broaden access to rare diseases. Yet risks persist: environmental costs from compute, and the need to prevent misdirection and deskilling. The overarching message: progress will be data-driven, incremental, and deeply dependent on empirical studies that reveal what AI can and cannot do.