探索与利用的权衡
Search documents
「我受够了Transformer」:其作者Llion Jones称AI领域已僵化,正错失下一个突破
3 6 Ke· 2025-10-26 23:24
Core Insights - Llion Jones, co-author of the influential paper "Attention is All You Need," expressed his fatigue with the Transformer architecture at the recent TED AI conference, highlighting a stagnation in AI research due to an over-reliance on a single framework [2][20]. Group 1: Current State of AI Research - Despite unprecedented investment and talent influx in AI, the field has become narrow-minded, potentially overlooking the next major breakthrough [2][8]. - Researchers are under pressure to publish quickly and avoid being "scooped," leading to a preference for safe, easily publishable projects over high-risk, transformative ideas [8][11]. - Jones noted that the current environment is reminiscent of the period before the introduction of the Transformer, where researchers were focused on minor improvements to RNNs, missing out on significant innovations [11][16]. Group 2: The Role of Freedom in Innovation - Jones emphasized that the Transformer was born from a free and organic research environment, contrasting sharply with today's pressure-laden atmosphere [12][14]. - He suggested that fostering an exploratory research environment, where researchers can take risks without the fear of immediate results, is crucial for future breakthroughs [13][19]. - At Sakana AI, Jones aims to recreate the conditions that led to the creation of the Transformer, minimizing competitive pressures and encouraging innovative thinking [14][15]. Group 3: Implications for Future AI Development - Jones warned that the success of the Transformer might be hindering the search for better technologies, as the current capabilities discourage exploration of alternatives [16][20]. - He called for a shift in the incentive structures within the AI research community to prioritize collaboration and shared discoveries over competition [18][19]. - The ongoing debate about the limitations of simply scaling Transformer models suggests that architectural innovation is necessary for continued progress in AI [19][20].