世界模型而非语言模型或将引发科学革命

今日内容一览：[00:29] 世界模型而非语言模型或将引发科学革命[01:33] OpenAI 对轻量级 GPT-5.6 模型大幅降价 80%[02:25] NASA 罗曼空间望远镜预定于 8 月底发射---[00:04] Welcome to today's tech podcast. Today we cover three major stories: Google DeepMind researchers analyze why world models rather than language models might spark future scientific revolutions, OpenAI aggressively slashes prices by 80 percent for its lightweight GPT-5.6 model, and NASA's flagship Roman Space Telescope prepares for launch nine months ahead of schedule.欢迎收听今天的科技播客。今天我们将带来三条重磅科技动态：谷歌 DeepMind 研究人员探讨为何世界模型而非语言模型才可能引发未来的科学革命、OpenAI 对其轻量级 GPT-5.6 模型大举降价 80%，以及 NASA 旗舰级罗曼空间望远镜提前 9 个月完成准备并将于 8 月底发射。---[00:29] World Models, Not Language Models, May Spark Scientific Revolutions. Can language models spark a scientific revolution? A new position paper from Google DeepMind researcher Tom Zahavy argues they cannot. Zahavy bases his argument on Albert Einstein's view of scientific discovery. Einstein believed science progresses through a sensory intuition (直觉；直觉感知。) that leaps to new foundational (基础的；根本的。) assumptions, or axioms, followed by logical deduction. While current language models excel at statistical pattern recognition and formal logic, they struggle with creative leapmaking when no existing template exists. When Einstein formulated general relativity, he relied on an intuitive mental simulation of a falling observer in an elevator rather than simple data fitting. Zahavy suggests that action-controllable world models, which allow AI agents to run interactive experiments in simulated physical environments, might provide the necessary feedback loop for genuine scientific breakthroughs.世界模型而非语言模型或将引发科学革命。语言模型能够引发科学革命吗？谷歌 DeepMind 研究员 Tom Zahavy 发表的一篇观点论文认为不能。Zahavy 的论据建立在爱因斯坦对科学发现的看法之上。爱因斯坦认为，科学的发展依赖于通过感官直觉跃迁至新的基础假设（即公理），随后再进行逻辑推演。尽管当前的语言模型在统计模式识别和形式逻辑方面表现出色，但在没有现成模板的情况下，它们难以完成创造性的理论飞跃。当爱因斯坦提出广义相对论时，他依赖的是对电梯中自由落体观察者的直觉思维模拟，而非单纯的数据拟合。Zahavy 指出，允许 AI 智能体在模拟物理环境中进行交互式实验的“动作可控世界模型”，或许能为真正的科学突破提供必要的反馈回路。（来源：The Decoder）---[01:33] OpenAI Slashes Prices by 80 Percent for Lightweight GPT-5.6 Model. OpenAI has slashed API prices for its GPT-5.6 model lineup (产品阵容；系列。). Effective July 30, the affordable GPT-5.6 Luna model saw an 80 percent price cut, dropping to twenty cents per million input tokens and one dollar twenty cents per million output tokens. The mid-tier Terra model received a 20 percent price reduction. OpenAI stated that tasks costing one dollar a year ago now run for about six cents on Luna while executing nearly nine times faster. The price cuts were made possible by GPT-5.6 Sol optimizing OpenAI's GPU software and improving token generation speeds. Intense price pressure from low-cost Chinese AI developers and competitor offerings also drove this aggressive (激进的；极具拓展性的。) pricing strategy.OpenAI 对轻量级 GPT-5.6 模型大幅降价 80%。OpenAI 对其 GPT-5.6 模型阵营进行了大幅降价。自 7 月 30 日起，平价模型 GPT-5.6 Luna 降价 80%，输入价格降至每百万 token 0.20 美元，输出价格降至 1.20 美元。中端模型 Terra 的价格也下调了 20%。OpenAI 表示，一年前需要花费 1 美元的任务，现在在 Luna 上运行只需约 6 美分，且运行速度快了近 9 倍。此次降价主要归功于 GPT-5.6 Sol 对 OpenAI 自身 GPU 软件的优化以及代币生成速度的提升。来自中国低成本 AI 开发者和竞争对手产品的激烈价格压力也是促成这一激进定价策略的重要因素。（来源：The Decoder）---[02:25] NASA's Roman Space Telescope Target Launch Set for Late August. NASA's next flagship observatory (天文台；空间观测站。), the Nancy Grace Roman Space Telescope, is scheduled to launch on August 30 aboard a SpaceX Falcon Heavy rocket from Kennedy Space Center. The mission is progressing nine months ahead of its official launch readiness date. Project managers credited strict adherence to cost limits and meticulous (细致的；周密的。) operational planning for delivering the complex spacecraft early. Engineers recently finished fueling the telescope with propellant needed for positioning at the Earth-Sun L-2 Lagrange point. Once operational, Roman will conduct wide-area infrared sky surveys to investigate dark energy and search for exoplanets, complimenting ground observatories.NASA 罗曼空间望远镜预定于 8 月底发射。NASA 的下一代旗舰观测站——奈西·格蕾丝·罗曼空间望远镜预定于 8 月 30 日搭乘 SpaceX 猎鹰重型火箭从肯尼迪航天中心发射。该任务比官方预定的发射就绪日期提前了 9 个月。项目经理将这台复杂航天器的提前交付归功于对成本限制的严格遵守和周密的行动规划。工程师近期已完成望远镜的推进剂加注，这些燃料将用于将其定位在日地 L-2 拉格朗日点。投入运行后，罗曼望远镜将开展大面积红外天空巡天观测，以研究暗能量并寻找系外行星，与地面观测站形成有效互补。（来源：SpaceNews）---[03:13] 📚 今日单词：discrepancy- /dɪˈskrepənsi/- noun [03:17] 词组/例句- - Following the logic of the argument, it would have done what the astronomers of the era did: invented an extra planet to account for the small discrepancy, rather than rethinking space and time. 按照这种逻辑推演，AI 只会像那个时代的天文学家那样，发明一颗额外的行星来解释这一点微小的差异，而不是重新思考空间与时间。---[03:44] That wraps up today's tech briefing. From the boundaries of AI scientific discovery to intense price wars and outer space exploration, technology continues to move at a breakneck pace. Which of today's topics intrigued you the most? Do you think AI world models will eventually make groundbreaking scientific discoveries? Let us know your thoughts in the comments, and thank you for listening!以上就是今天科技播客的全部内容。从 AI 科学发现的边界到激烈的价格战，再到深入外太空的探险，科技发展的脚步从未停歇。今天探讨的话题中，哪一个最引发你的思考？你认为 AI 世界模型最终能做出突破性的科学发现吗？欢迎在评论区分享你的看法，感谢收听，我们下期再见！前往小宇宙评论区与主播互动