三,生萬物

incredible outperformance on goal 3, even though it took awhile:https://t.co/5GpYOlN8HB

— Sam Altman (@sama) September 17, 2024

OpenAI technical goals

June 20, 2016

OpenAI’s mission is to build safe AI, and ensure AI’s benefits are as widely and evenly distributed as possible.

Illustration: Ruby Chen

We’re trying to build AI as part of a larger community, and we want to share our plans and capabilities along the way. We’re also working to solidify our organization’s governance structure and will share our thoughts on that later this year.

Our metric

Defining a metric for intelligence is tricky, but we need one to measure our progress and focus our research. We’re thus building a living metric which measures how well an agent can achieve its user’s intended goal in a wide range of environments.

Goal 1: Measure our progress

The metric will consist of a variety of OpenAI Gym(opens in a new window) environments with a unified action and observation space(opens in a new window) (so a single agent can run across all of them), including games, robotics, and language-based tasks. Our implementation will evolve over time, and we’ll keep the community updated along the way.

Our research

A significant fraction of our research bandwidth is being spent on fundamental research. We’ll always be developing and testing new ideas, especially those that don’t fit neatly into our current worldview. This is important—our current ideas will not be enough to achieve our long-term goal.

We’ve also formed teams around specific projects. The intention isn’t just to solve these problems, but to develop general learning algorithms in the process. These algorithms will, in turn, help us build agents that are more capable according to our metric. These projects are:

Goal 2: Build a household robot

We’re working to enable a physical robot (off-the-shelf; not manufactured by OpenAI) to perform basic housework. There are existing(opens in a new window) techniques(opens in a new window) for specific tasks, but we believe that learning algorithms can eventually be made reliable(opens in a new window) enough(opens in a new window) to create a general-purpose robot. More generally, robotics is a good testbed for many challenges in AI.

Goal 3: Build an agent with useful natural language understanding

We plan to build an agent that can perform a complex task specified by language, and ask for clarification about the task if it’s ambiguous. Today, there are promising algorithms for supervised language tasks such as question(opens in a new window) answering(opens in a new window), syntactic(opens in a new window) parsing(opens in a new window) and machine(opens in a new window) translation(opens in a new window) but there aren’t any for more advanced linguistic goals, such as the ability to carry a conversation, the ability to fully understand a document, and the ability to follow complex instructions in natural language. We expect to develop new learning algorithms and paradigms to tackle these problems.

Goal 4: Solve a wide variety of games using a single agent

We aim to train an agent capable enough to solve any game in our initial metric. Games are virtual mini-worlds that are very diverse, and learning to play games quickly and well will require significant advances in generative models and reinforcement learning(opens in a new window). (We are inspired by the pioneering work of DeepMind(opens in a new window), who have produced impressive(opens in a new window) results(opens in a new window) in this area in the past few years.)

Our projects and fundamental research all have shared cores, so progress on any is likely to benefit the others. Each captures a different aspect of goal-solving, and was chosen for its potential to significantly move our metric.

We’re just getting started on these projects, and the details may change as we gain additional data. We also expect to add new projects over time.

Authors

Ilya Sutskever, Greg Brockman, Sam Altman, Elon Musk

Last edited by @suen 2024-09-17T23:32:57Z

@Creative 使用典雅中文翻譯,同時注意信、達。

OpenAI 技術目標

二零一六年六月二十日

OpenAI 的使命是構建安全的 AI,並確保 AI 的益處儘可能廣泛且公平地分配。

(插圖:Ruby Chen)

我們正在努力構建 AI,並将其视为一個更大的社群的一部分,我們希望在此過程中分享我們的計劃和能力。我們也在努力鞏固本組織的治理結構,並將在今年晚些時候分享我們在這方面的想法。

我們的衡量指標

為智慧定義一個衡量指標是棘手的,但我們需要一個指標來衡量我們的進展並集中我們的研究方向。因此,我們正在構建一個動態的指標,用於衡量一個代理在廣泛的環境中實現其使用者預期目標的能力。

目標一:衡量我們的進展

該指標將包含各種具有統一動作和觀察空間的 OpenAI Gym 環境(以便單個代理可以在所有環境中運行),包括遊戲、機器人和基於語言的任務。我們的實作將隨著時間的推移而發展,我們將隨時向社群更新。

我們的研究

我們很大一部分研究資源都投入到了基礎研究中。我們將始終開發和測試新的想法,尤其是那些不完全符合我們當前世界觀的想法。這很重要——我們當前的想法不足以實現我們的長期目標。

我們還圍繞特定項目組建了團隊。目的不僅是解決這些問題,而且是在此過程中開發通用的學習演算法。這些演算法反過來將幫助我們根據我們的指標構建更強大的代理。這些項目是:

目標二:構建家用機器人

我們正在努力使一個實體機器人(現成的;非 OpenAI 製造)能夠執行基本的家務勞動。目前已有一些用於特定任務的 現有技術 ,但我們相信,學習演算法最終可以變得 足夠可靠 ,從而創造出一個通用的機器人。更廣泛地說,機器人是 AI 許多挑戰的良好試驗平台。

目標三:構建具有實用自然語言理解能力的代理

我們計劃構建一個能夠執行由語言指定的複雜任務的代理,並在任務不明確時請求澄清。如今,對於諸如 問題 解答語法 解析機器 翻譯 等監督式語言任務,已經有一些很有前景的演算法,但對於更高級的語言目標,例如進行對話的能力、完全理解文檔的能力以及遵循自然語言中的複雜指令的能力,還沒有任何演算法。我們期望開發新的學習演算法和範式來解決這些問題。

目標四:使用單個代理解決各種遊戲

我們的目標是訓練一個足夠強大的代理,使其能夠解決我們初始指標中的任何遊戲。遊戲是極其多樣化的虛擬微型世界,學習快速而有效地玩遊戲將需要在 生成模型強化學習 方面取得重大進展。(我們受到了 DeepMind 先驅工作的啟發,他們在過去幾年中在這一領域取得了 令人印象深刻 的成果 。)

我們的項目和基礎研究都具有共同的核心,因此任何一個項目的進展都可能使其他項目受益。每個項目都抓住了目標解決的不同方面,並因其具有顯著提升我們指標的潛力而被選中。

我們才剛剛開始這些項目,隨著我們獲得更多數據,細節可能會發生變化。我們也期望隨著時間的推移增加新的項目。

作者

Ilya Sutskever, Greg Brockman, Sam Altman, Elon Musk