--- author: Dwarkesh Patel date: '2024-06-13' guest: '' layout: post.njk source: https://www.youtube.com/watch?v=wZ0ToxtXz5g speaker: Dwarkesh Patel tags: - agi - intelligence - arc-benchmark - memorization - brute-force title: AI智能的本质：从Arc基准到‘暴力破解’的思考 summary: 本次讨论探讨了AI模型若能解决Arc基准测试（达到人类80%水平）是否标志着AGI的到来。发言者质疑了基准测试的有效性，认为其可能被规模化数据“作弊”或“暴力破解”。文章深入剖析了智能的本质，将其比作路径寻找算法，并挑战了AI“记忆”与人类“推理”的二元划分，认为两者可能处于同一连续光谱上，且人类学习过程本身也包含大量算法记忆。 insight: '' draft: true series: '' category: ai-ml area: tech-engineering project: [] people: [] companies_orgs: [] products_models: - arc media_books: [] status: evergreen --- ### AI智能的本质：从Arc基准到“暴力破解”的思考 讨论的核心在于，如果一个多模态模型能在一年内解决“Arc”基准测试，并达到人类平均水平的80%，这是否意味着通用人工智能（AGI）的出现？对此，演讲者提出了一种设想：期望看到一个LLM类型的模型能够以80%的准确率解决Arc，但其训练数据不应是明确为预测该测试集而准备的。然而，Arc基准测试的意义在于，它每次都呈现新的图谱和智能挑战，使得作弊变得不可能。如果Arc是一个完美无瑕的基准，那么在测试集内是无法被预测的。Arc测试已发布四年多，至今仍能抵抗“死记硬略”式的解决方案，在一定程度上经受住了时间的考验。但演讲者也指出，如果通过手工创建数十万个Arc任务，再通过程序化生成变体，最终可能产生数亿个任务，那么训练数据与测试集之间很可能会产生足够的重叠，从而能够获得极高的分数。这引出了一个观点：拥有足够的规模，总能“作弊”。
Original English so suppose that it's the case that in a year a multimodal model can solve Arc let's say get 80% whatever the average human would get then AGI quite possibly yes I think if you if you start so honestly what I would like to see is uh an llm type model solving Arc at like 80% but only trained on information that is not um explicitly trying to anticipate what it's going to be in the AR test set but isn't the isn't the whole point of Arc that you can't sort of it's a new chart of type of intelligence every single time is the point so if Arc were perfect Flawless Benchmark it would be impossible to anticipate within the test set and you know Arc was released uh more than four years ago and so far it's been resistant to memorization so I think it has uh to some extent passed a test of time but I don't think it's perfect I think if you try to make by hand uh hundreds of thousands of AR tasks and then you try to multiply them uh uh by programmatically generating variations and then you end up with maybe hundreds of millions of tasks uh just by brute forcing the task space there will be enough overlap between what you're train on and what's in the test set that you can actually score very highly so you know with enough scale you can always cheat
### 智能的路径寻找与“暴力破解”的界限 演讲者进一步探讨了智能的本质，提出了一种比喻：可以将智能视为一种在未来情境空间中寻找路径的算法。这类似于实时战略（RTS）游戏开发中的概念，玩家在一个地图上拥有部分信息（存在战争迷雾），无法完全了解未知区域或过去区域的现状。智能的目标是，一旦设定了目标，就能找到通往该目标的最佳路径。然而，这种路径寻找受限于已知信息，无法预见未知或变化。如果能完全掌握地图的所有信息，路径寻找问题可以通过死记硬略所有可能的路径来解决，即通过纯粹的记忆来完成。但在现实生活中，由于未来充满不确定性，生命是不断变化的，因此纯粹的记忆或死记硬略是不现实的。演讲者暗示，如果世界的“分布”是静态的（例如，在一种“病态的分布”下），那么通过“暴力破解”所有可能的行为空间来“强制获得”智能是可能的。
Original English if you can do this for every single thing that supposedly requires intelligence then what good is intelligence apparently you can just Brute Force intelligence if if the world if your life were athetic distribution uh then sure you could just bruteforce the space of possible behaviors could like you know the way we think about intelligence there are several metaphors SEL actives but one of them is you can think of intelligence as a past finding algorithm in future situation space like I don't know if you're familiar with game development like RTS game development but you have a map right and and you have it's like a 2d 2D 2D map and um you have partial information about it like there is some fog of War on your map there are areas that you haven't explored yet you know nothing about them and then there are areas that you've explored but um you only know how they were like in the past you don't know how they like today and um and and now instead of thinking about Tod I think about the space of possible future situations that you might encounter and how they're connected to each other intelligence is a past finding algorithm so once you set a goal it will tell you uh how to get there Optimum um but of course it's it's constrained by the information you have uh it it cannot pass fine in an area that you know nothing about it cannot also anticipate uh uh changes and um the the the thing is if you had complete information about the map uh then you could solve the pathf finding Problem by simply memorizing every possible path every mapping from uh point A to point B uh you could you could solve the problem with pure memory but where the reason you cannot do that in real life is because you don't actually know what's going to happen in the future life is Ever Changing
### 人类与AI的学习光谱：记忆还是推理？ 演讲者对“记忆”与“推理”的界限提出了质疑，认为这种区分在人类和AI之间可能过于简化。当一个孩子学习代数或微积分时，我们不会说他们“记住了”微积分，而是说他们“学会了”微积分。然而，当AI解决类似的数学问题时，这种能力却常被贴上“记忆”的标签。演讲者认为，人类学习过程，包括掌握加法算法、代数或微积分，本质上也是在学习和应用算法或程序，这与AI的学习过程在根本上是相似的。如果AI通过学习大量技能和技术，即使它们能自动化几乎任何任务，只要分布是静态的，就可能被归类为“记忆”。演讲者进一步推测，如果远程工作技能可以通过记录屏幕并训练模型来合成数据，从而导致大量远程工人失业，而AI创造了巨大的经济价值，那么这种情形下的AI能力是否仍应被视为“记忆”？核心观点是，人类和AI在处理信息和解决问题的方式上可能共享一个连续的光谱，而非截然不同的类别。
Original English I feel like you're using words like memorization which we never use for human children if if like your kid learns to do algebra and then like now learns to do calculus you wouldn't say they memorized Calculus if they can just solve any arbitrary algebraic problem you wouldn't say like they've memorized algebra they say they've learned algebra humans are never really doing pure memorization or pure reasoning but that's only because you're semantically labeling when the human does the skill it's a memorization when the exact same skill is done by the llm as you can measure by these bench marks and you can just like plug in any sort of math problem humans are doing the exact same as LM is doing which is just for instance I know if you learn to add numbers you're memorizing an algorithm you're memorizing a program and then you you can reapply it you are not synthesizing on the Fly uh the addition program so obviously at some point some human had to figure out how to do addition but like the way a kid learns it is not that they sort of out from the accents of SE Theory how to do addition I think what you learn in school is mostly memorization right so my claim is that listen these models are vastly underparameterized relative to how many flops or how many parameters you have in the human brain and so yeah they're they're not going to be like coming up with new theorems like the smartest humans can but most humans can't do that either um what most humans do it sounds like a similar to what you're calling memorization which is memorizing skills or memorizing um you know uh techniques that you've learned and so it sounds like it's compatible in your tell me if this is wrong is it compatible in your world if like all the remote workers are gone but they're doing skills which we can potentially make synthetic data of so we record everybody's screen and every single remote worker screen we sort of understand the skills they're performing there and now we've trained a model that can do all this all the remote workers are unemployed we're generating trillions of dollars of economic activity from Mii uh remote workers in that world is are we still in the memorization regime so sure uh with memorization you can automate almost anything as long as it's it's a static distribution as long as you don't have to deal with change