IT之家 9 月 15 日消息，在接受 a16z 采访时，OpenAI 联合创始人格雷格 · 布罗克曼（Greg Brockman）表示，人类已经迈入通用人工智能时代。
IT之家注：通用人工智能（Artificial General Intelligence，AGI）通常指能在广泛领域处理复杂认知任务，并能迁移、学习和完成新任务的 AI。
OpenAI 在《OpenAI Charter》中给出的定义为 AGI 是“高度自主、且在大多数具有经济价值的工作上超越人类的平台”。这一定义包含两道门槛：一是平台可自主完成较完整的工作链路；二是在多数经济活动中，其表现超过人类。
在采访中，布罗克曼表示随着 GPT-6 Astra 等模型具备长时程自主执行与电脑操作能力，AI 已跨入可被称为“AGI 时代”的门槛。AGI 不再只是将来的概念，而是正在现实中浮现；接下来的关键难题将逐步从“何时达成 AGI”转向如何使用和部署越来越强大的 AI 平台。
对此布罗克曼表示 OpenAI 抽调了 25% 的生产工程师，暂停其原有项目，转入平台防御和安全架构升级工作。团队使用 Astra 查找内部平台中的漏洞，并称已修复多项严重难题。
另外，布罗克曼称，团队在排查中发现了新难题，但随后难题发现进入“饱和”。他表示，据 OpenAI 所知，团队已基本找到 Astra 有能力识别的全部 P0 级难题，以及其可发现的全部关键难题。
他指出，这一结果并不代表平台不存在其他漏洞。随着新模型浮现，团队将启动新一轮排查，并把新一代网络安全能力部署到内部平台中，以寻找此前未被识别的漏洞。
本文来源：https://www.ithome.com/1/002/341.htm
若有版权问题，请来信告知：13951431913#139.com 或在右方提交评论。
Home > Intelligent Age > Artificial Intelligence
Thanks to IT之家 reader Stephen_233 for the tip!
IT之家, September 15 — In an interview with a16z, OpenAI co-founder Greg Brockman stated that humanity has entered the era of general artificial intelligence.
IT之家 Note: Artificial General Intelligence (AGI) typically refers to AI that can process complex cognitive tasks across a wide range of domains, with the ability to transfer, learn, and perform new tasks.
In the 《OpenAI Charter》, AGI is defined as “a highly autonomous system that outperforms humans in the majority of economically valuable work.” This definition involves two thresholds: the system must be able to autonomously complete relatively complete work chains, and in most economic activities, its performance must surpass that of humans.
In the interview, Brockman stated that with models like GPT-6 Astra demonstrating long-duration autonomous execution and computer operation capabilities, AI has crossed the threshold into what can be called the “AGI era.” AGI is no longer just a future concept but is emerging in reality; the key challenge going forward will shift from “when to achieve AGI” to how to use and deploy increasingly powerful AI platforms.
Regarding this, Brockman revealed that OpenAI has diverted 25% of its production engineers, halting their original projects to focus on platform defense and security architecture upgrading. The team used Astra to search for vulnerabilities in the internal platform and claimed to have resolved multiple severe issues.
Additionally, Brockman noted that the team discovered new problems during the investigation but subsequently entered a state of “saturation” for problem discovery. He explained that, to the best of OpenAI’s knowledge, the team has essentially found all the P0-level issues that Astra is capable of identifying, as well as all the critical vulnerabilities it can detect.
He pointed out that this result does not mean the platform is free of other vulnerabilities. As new models emerge, the team will initiate a new round of inspections and deploy the latest cybersecurity capabilities to the internal platform to search for previously unidentified vulnerabilities.