Altman says OpenAI can't please everyone and must balance AI safety with development奥尔特曼称 OpenAI 没法让所有人满意,必须平衡 AI 安全与发展
According to IT Home on 19 November, at the DealBook Summit held on Wednesday, 18 November, US time, OpenAI CEO Sam Altman said that his company is "balancing" between "safety" and "...
IT之家 11 月 19 日消息,美国当地时间 11 月 18 日周三召开的 DealBook 峰会上, OpenAI 首席执行官萨姆 · 奥尔特曼(Sam Altman)表示,其公司正在“安全”与“…
Key Points要点速览
According to IT Home on 19 November, at the DealBook Summit held on Wednesday, 18 November, US time, OpenAI CEO Sam Altman stated that his company is seeking a balance between "safety" and "development" in AI. Altman and OpenAI President Greg Brockman participated in the event together, stating that risks are inevitable in the process of advancing AI development, and therefore the company must make trade-offs between prioritising safety and continuing innovation. In the interview, Altman candidly admitted that it is impossible for OpenAI to "satisfy everyone". He said that if the company were to "put the pedal to the metal" in pursuit of technological breakthroughs, the public would worry about AI getting out of control; but if it were to overemphasise safety, it could also slow down product and technological progress. Altman emphasised that OpenAI adopts an "iterative deployment" strategy, which involves rolling out technology in stages, observing risks in real-world environments, and continuously making improvements. In his view, one of the most important ways to understand AI risks is not to conduct development in isolation, completely detached from reality, but rather to subject the technology to real-world testing. Altman also acknowledged that as the capabilities of models rapidly improve, the challenges of safety, alignment, and monitoring are becoming significantly more difficult. OpenAI President and co-founder Greg Brockman said that as model capabilities have significantly improved, safety, security, and "alignment" (i.e., AI behaviour that is in line with human intentions) have become issues that must be taken seriously. He said that this is not just a change for OpenAI as a company, but a cultural shift that the entire AI industry is undergoing. OpenAI has signed voluntary agreements with other AI companies, committing to conducting risk assessments and collaborating with third-party assessment organisations. The company said this week that it is bringing in independent security researchers from organisations such as METR and Apollo Research to conduct security assessments of its models. Altman also warned that the current fierce competition among AI companies is detrimental to safety. He stated that companies can only understand their own safety standards and find it difficult to fully grasp the development status of their competitors; this "race in the fog" may lead to insufficient investment in safety. He called on the industry to establish common safety standards and promote greater transparency. Altman explicitly rejected some of the discussions in Silicon Valley about AI gradually taking over decision-making power, and even potentially forming "superintelligence". He said that humans must continue to control the future and should not gradually hand over decision-making power to AI. Brockman also believes that the more powerful AI becomes, the greater the responsibility to ensure that it is aligned with human goals and is safe and controllable. On the business side, Altman revealed that OpenAI still plans to go public at some point. Previously, OpenAI had postponed its IPO plans until next year to prioritise security issues. Earlier this month, the company completed a $11 billion (approximately RMB 73.489 billion at the current exchange rate) funding round, reaching a valuation of $1.25 trillion (approximately RMB 8.35 trillion at the current exchange rate).
IT之家 11 月 19 日消息,美国当地时间 11 月 18 日周三召开的 DealBook 峰会上, OpenAI 首席执行官萨姆 · 奥尔特曼(Sam Altman)表示,其公司正在“安全”与“发展”寻找 AI 平衡点。 奥尔特曼和 OpenAI 总裁格雷格 · 布罗克曼(Greg Brockman)共同参与本次活动,他们表示在推进 AI 发展的进程中不可避免地会遇到风险,因此公司必须在优先保障安全和持续创新之间做出取舍。 奥尔特曼在访谈中坦言,OpenAI 不可能“让所有人都满意”。他称,如果公司只顾“踩满油门”追求技术突破,公众会担心人工智能失控;但若过度强调安全,也可能延缓产品和技术进步。 奥尔特曼强调 OpenAI 采取“迭代部署”策略,即分阶段推出技术、在真实环境中观察风险并不断改进。在他看来,理解 AI 风险的重要方式之一,不能完全脱离实际进行封闭开发,更重要的是让技术接受现实检验。 奥尔特曼也承认,随着模型能力迅速增强,安全、对齐和监控的难度正在显著提高。OpenAI 总裁、联合创始人格雷格 · 布罗克曼表示,随着模型能力显著增强,安全性、保障性和“对齐”(即 AI 行为符合人类意图)已成为必须认真对待的问题。他称,这不只是 OpenAI 一家公司的转变,更是整个 AI 行业正在经历的文化转变。 OpenAI 已与其他 AI 公司签署自愿协议,承诺开展风险评估,并与第三方评估机构合作。该公司本周表示,正在引入 METR、Apollo Research 等机构的独立安全研究人员,对模型进行安全评估。 奥尔特曼还警告,当前 AI 公司之间的激烈竞争不利于安全。他表示,各家公司只能了解自身的安全标准,难以充分掌握竞争对手的开发情况;这种“迷雾中的竞赛”可能使安全投入不足。他呼吁行业建立共同的安全标准,并推动更大程度的透明度。 针对硅谷部分关于 AI 逐步接管决策权、甚至可能形成“超级智能”的讨论,奥尔特曼明确表示反对。他说,人类必须继续掌控未来,不应将决策权逐步交给 AI;布罗克曼也认为,AI 能力越强,保障其与人类目标一致、安全可控的责任就越大。 在商业层面,奥尔特曼透露 OpenAI 仍计划在某个时点上市。此前,OpenAI 已将 IPO 计划推迟至明年,以优先处理安全问题;本月早些时候,公司完成 110 亿美元 (IT之家注:现汇率约合 734.89 亿元人民币) 融资,估值达 1.25 万亿美元 (现汇率约合 8.35 万亿元人民币) 。
想马上用起来?去「AI工具」栏挑一个直接下载,或在「AI教程」里跟着图文步骤做一遍。
去 AI工具 → 看 AI教程 →阅读与点赞数据保存在你的浏览器本地,欢迎留下你的想法。
今日正能量学一点,用一点;今天种下的种子,会长成明天的能力。去免费下载专区 →广告
评论 文明发言,让讨论更有价值