Next-Token Myopia 🤔🤬😡 - сборник AI антипаттернов собираю и всякие деградации / галлюцинации . Можете скормить своим агентам и поделиться - было ли у вас такое или нет еще: Next-Token Myopia refers to a critical limitation in Autoregressive Large Language Models (LLMs), where the model’s training objective forces it to focus strictly on predicting the immediate next token instead of optimizing for long-range planning, coherence, or structural outcomes. While Next-Token Prediction (NTP) has powered massive breakthroughs in artificial intelligence, this "short-sightedness" prevents models from naturally executing complex reasoning tasks that require implicit planning