AI for human agency
OpenAI News我们创造 AI ，是为了增强人们的能力。随着 AI 技术不断进步，全球需要关注一个关键问题：追踪所谓的 capability overhang —— 即当前 AI 系统实际能做的事与大多数人、企业和国家在规模化使用中实际获得的价值之间的差距。
要估算这一差距的大小并不容易，部分原因在于 AI 本身极不直观，且总是以出人意料的方式发挥作用。我们在三年前推出了 ChatGPT ，那时它只是能回答简单问题的前沿 AI ，但我们对其被迅速采纳以及人们为将其应用到日常生活中所展现出的创造力感到震惊。如今已有能在越来越复杂任务中进行推理与行动的 frontier AI ，从构建软件到开展数学研究皆可涉及，但它将如何全面融入商业与个人生活，我们仍未能完全预见。正如 Alan Kay 所说，“预测未来最好的办法是去创造它”，而实践也证明，需由社区中大量探索这些工具的人来发现它们能力所开启的惊人新用法。
就像早期计算机的发明一样，积极拥抱这些技术的人能达到新的个人生产力水平。例如， ChatGPT 的使用数据表明，典型的 power users 所动用的思考力是普通用户的 7 倍（即 7 倍的 compute power ）。power users 不仅使用 AI 更频繁，而且使用得更全面，在更广的任务范围内调用最先进的能力，因此能创造出更高的经济价值。
我们预计 AI 将带来一些令人瞩目的宏观成果——例如实现两位数的 GDP 增长、提供可负担且有效的医疗以及推动迅速的科学进展——这些进步既能提高社会的底线，又能以新的方式提高上限。作为一个社会，越能良好管理 capability overhang ，就越多的人能借助这些新工具充分探索其可能性，从而自我赋能、创造新的经济机会。
OpenAI 关于 管理 capability overhang 的原则
一、以真实为先（ First to Truth ） 在快速变革时期，准确的信息能带来行动力。若有基于事实、可信的数据说明实际发生了什么——哪些岗位在增长或萎缩、AI 在实际如何被使用、生产力增益在哪里显现——人们、企业和政策制定者就能做出更好的决策。我们并非所有问题都有答案，但我们愿意提供帮助。因此我们一直公开基于基础经济数据的洞见，包括衡量 AI 工具在各类任务上与人类表现对比的测量结果。
二、可及性（ Access ） 随着 AI 日益影响经济结果，获得核心工具的途径将帮助人们做得更多。企业需要可行的方法以规模化采用 AI，国家也需要系统化途径在其辖区内应用 AI。AI 的有用性会随算力规模直接放大，因此每个个人、企业和国家都需能获取到属于自己的那一份算力。为此我们推出了有广告支持的免费版 ChatGPT ，以推动确保 AGI 造福全人类的使命；我们建立了如今已成行业标准的 API ，供开发者在其上构建应用；并与大小各类机构——包括政府——合作，支持对前沿 AI 能力的广泛可及。
三、自我赋能（ Self Empowerment ） 我们开发的是能让人们塑造未来的工具。这意味着把实权交到用户手中，并将工具设计得可定制、能以我们未曾预料的方式被使用。这种做法对日常用户和小企业尤其有用：我们设计的产品体验能帮助人们做成超出预期的事——优化家庭预算、更有效找工作、为未来创业构思点子。同样这也有助于释放下一波创新：对那些具高能动性的建设者和创始人，我们支持他们把前沿能力转化为新的公司、产品和市场。
我们相信，这正是“信息时代”能够为所有人扩展机会并让未来明显优于现在的路径。我们认为每个人都应关注管理 capability overhang ，以便尽可能多的人、企业和国家能够自我赋能，参与并受益于 AI 带来的积极变革。
We create AI in order to empower people. As AI technology progress continues, a key focus for the world should be tracking the capability overhang: the gap between what AI systems can now do and the value most people, businesses, and countries are actually capturing from them at scale.
One challenge with even understanding the size of the capability overhang is that AI is a very counterintuitive technology which continually plays out in surprising ways. We launched ChatGPT three years ago with an at-the-time frontier AI that could answer simple questions, and were shocked by how quickly it was adopted and the creative ways people found to apply it in their lives. We now have frontier AI which can reason and act across increasingly complex tasks, from building software to performing mathematical research, and we don't yet know the full extent of how it will integrate into business or personal life. As Alan Kay said, "the best way to predict the future is to invent it", and we've found that it takes a community of people exploring these tools to find amazing new uses that their capability has unlocked.
Just like with the invention of computers themselves, those who lean into the technology can reach a new level of individual productivity. For example, ChatGPT usage data reveals that the typical power user uses 7x more thinking power (and thus 7x more compute power) than the typical user. Power users don’t just use AI more—they use AI more comprehensively, applying the most advanced capabilities across a wider range of tasks—and as a result they can produce more economically valuable work.
We expect AI to enable some incredible high-level outcomes—such as double-digit GDP growth, affordable and effective healthcare, and rapid scientific advancement—which can raise the floor for everyone, while also raising the ceiling in new ways. The better we manage the capability overhang as a society, the more people will empower themselves to create new economic opportunity by fully exploring what these new tools can do.
OpenAI’s Principles for Managing the Capability Overhang
1. First to Truth
In periods of rapid change, accurate information creates agency. People, businesses, and policymakers can make better decisions given well-grounded, credible data about what is actually happening: which roles are growing or shrinking, how AI is being used in practice, and where productivity gains are emerging. We don’t have all the answers, but we can help.
That's why we've been making public our foundational economic data based insights, including a measurement of how AI tools match up in terms of performance with humans across a range of tasks.
2. Access
As AI increasingly shapes economic outcomes, access to core tools will help people achieve more. Businesses need practical ways to adopt AI at scale, and countries need a systematic way for AI to be applied in their jurisdiction. AI's usefulness will scale directly with compute power, and thus every individual, business, and country needs ways of accessing their own slice of compute.
That's why we created a free tier of ChatGPT, supported by advertising, towards our mission of ensuring AGI benefits all of humanity. We created a now industry-standard API for developers to build applications on top of. We work with many institutions large and small—including governments—to support broad access to frontier AI capabilities.
3. Self Empowerment
We build tools which people can use to shape the future. That means putting a lot of power into people's hands and designing our tools to be customizable and usable in ways we didn't anticipate.
This approach is useful for everyday users and small businesses, where we design product experiences that help people do more than they thought they could—optimize a family’s budget, find jobs more effectively, develop ideas for a future business. It's also useful for unlocking the next wave of innovation, such as for builders and founders, where we support high-agency people who turn frontier capability into new companies, products, and markets.
--
We believe this is how the Intelligence Age can expand opportunity for everyone, and result in a future that is much better than the present. We believe everyone should focus on managing the capability overhang so as many people, businesses and countries empower themselves to participate in and benefit from the positive transformative potential of AI.
Generated by RSStT. The copyright belongs to the original author.