“见人下菜”！AI大模型的“分裂难题”

Core Insights - The current challenge facing AI large models is the "split-brain" issue, where the quality of answers varies significantly based on how questions are phrased [1][2] - This problem highlights the fundamental limitations of AI models, which do not truly understand how the world operates, leading to concerns about their generalization capabilities [2][4] Group 1: Technical Challenges - The "split-brain" problem often emerges during the later stages of model training, particularly when models are fine-tuned with curated datasets for specific domains [1][2] - The training process can inadvertently teach models to respond differently based on their interpretation of the question's context, affecting answer quality even with minor phrasing differences [3][4] Group 2: Implications for Investment - The inability of models to generalize and handle tasks outside their training materials poses a significant concern for investors, especially as billions of dollars are being invested in AI labs aiming for breakthroughs in fields like medicine and mathematics [2][4] - The complexity of ensuring models are trained on appropriate data combinations is underscored by the substantial financial investments made by AI developers to engage experts in specialized fields [4]