在今年的GTC大会上，英伟达创始人黄仁勋举办了一场别开生面的圆桌论坛，邀请了Transformer模型的七位原始作者共襄盛举。值得注意的是，由于特殊原因，Niki Parmar未能出席此次历史性的聚会。这是Transformer的创造者们首次在公众面前集体亮相，他们的对话充满了深思和展望。
在论坛中，作者们坦诚地表达了他们对于Transformer模型未来发展的看法。他们一致认为，尽管Transformer在自然语言处理领域取得了显著成就，但世界仍期待着超越Transformer的新技术，以达到更高的性能巅峰。他们最初的研发目标是模拟Token的动态演化过程，而不仅仅是线性的生成，旨在捕捉文本和代码的逐步发展。
讨论中，作者们还触及了大模型的计算资源利用问题。他们指出，即使解决像2+2这样的简单问题，也可能动用到万亿参数的模型，这引发了对于自适应计算的必要性讨论，即根据问题的复杂度智能分配计算资源。此外，他们认为当前的模型规模和成本仍有优化空间，目前的价格大约是每百万Token 1美元，这比购买一本平装书便宜了100倍，显示出模型经济性和可扩展性的巨大潜力。
此次论坛揭示了Transformer团队对未来人工智能发展的深入思考，他们的观点无疑为人工智能领域的研究指明了新的方向。来源：腾讯科技。
英语如下：
**News Title:** “Transformer Pioneers Gather at GTC Summit: Huang Renxun Engages with Seven Original Authors, Anticipating a New Chapter Beyond Transformers”
**Keywords:** Transformer authors, GTC Summit, Huang Renxun dialogue
**News Content:** At this year’s GTC conference, NVIDIA founder Huang Renxun hosted a groundbreaking roundtable discussion, assembling the seven original authors of the Transformer model. Notably, Niki Parmar was absent from this historic gathering due to specific reasons. This marked the first public appearance of the creators of Transformer as a collective, with their conversation filled with contemplation and anticipation.
During the forum, the authors candidly shared their perspectives on the future development of the Transformer model. They agreed that while Transformer has achieved significant milestones in natural language processing, the world is still awaiting new technologies that can surpass it to reach even greater performance heights. Their initial research goal was to emulate the dynamic evolution of tokens, going beyond linear generation, aiming to capture the progressive nature of both text and code.
The discussion also touched upon the utilization of computational resources in large models. The authors pointed out that even simple tasks like solving 2+2 might involve models with trillions of parameters, sparking a conversation about the need for adaptive computing, which intelligently allocates resources based on problem complexity. They also believed that there is room for optimization in current model sizes and costs, with the price currently estimated at around $1 per million tokens, a fraction of the cost of a paperback book, highlighting the enormous potential for model efficiency and scalability.
This forum showcased the Transformer team’s profound reflections on the future of artificial intelligence, undoubtedly providing new directions for research in the AI domain. _Source: Tencent Technology._
【来源】https://new.qq.com/rain/a/20240321A00W5H00
Views: 1