TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

The Scaling Hypothesis (2021)

97 点作者 oli5679将近 3 年前

5 条评论

light_hue_1将近 3 年前
As an ML researcher, simple back of the envelope math shows you that the scaling hypothesis is totally wrong. We will never reach human-level AI by simply scaling what we have today.<p>GPT-3 has 10^11 parameters and needs 10^14 bytes of training data. Averaged performance on a bunch of benchmarks is 40-50% depending on what kind of prompts you provide: <a href="https:&#x2F;&#x2F;res.cloudinary.com&#x2F;dyd911kmh&#x2F;image&#x2F;upload&#x2F;f_auto,q_auto:best&#x2F;v1598020447&#x2F;gpt3-3_krvb14.png" rel="nofollow">https:&#x2F;&#x2F;res.cloudinary.com&#x2F;dyd911kmh&#x2F;image&#x2F;upload&#x2F;f_auto,q_a...</a> 10x fewer parameters drops your performance by about 10%.<p>If you just linearly extrapolate that graph, and ML doesn&#x27;t generally scale linearly, models tend to peter out eventually, you&#x27;re talking about needing models that are 10^6 or more larger with a similar increase in training data. This is.. starting to be impractical.<p>That&#x27;s 10^17 or more parameters and 10^20 or more data. And that&#x27;s assuming the models actually continue to learn.<p>This is also extrapolating with an average. Datasets in machine learning are not difficulty calibrated at all. We have no idea how to measure difficulty. So this extrapolation is being driven by the easier datasets, and it won&#x27;t saturate the hard ones. For example, GPT-3 makes a lot of systematic errors and there are plenty of benchmarks where it just isn&#x27;t very good regardless of how many parameters it has.<p>Our understanding of what intelligence is in the first place is the biggest hurdle here. This is why we can&#x27;t benchmark systems. Why we can&#x27;t come up with a benchmark, where performance on that benchmark means we&#x27;re x% of the way toward an intelligent system. As systems get better, our benchmarks and datasets get better to keep up with them. So just saying we&#x27;re going to saturate performance on today&#x27;s benchmarks with some model that has 10^17 parameters just doesn&#x27;t mean much at all.<p>We have no guarantee and no reason to expect that doing well on today&#x27;s benchmarks, even if we invested trillions of dollars would matter in the grand scheme of things.<p>Doesn&#x27;t mean these models can&#x27;t be useful. But there&#x27;s plenty more to do before we can just say &quot;take what we have and invest $1T to scale it up and we&#x27;ll be good to go&quot;.
评论 #31567276 未加载
评论 #31567110 未加载
评论 #31565040 未加载
评论 #31566669 未加载
Jack000将近 3 年前
I think the main disconnect between large language models and what people expect from &quot;AI&quot; is the fact that LMs learn meaningful representations whereas most people associate intelligence with meaningful behaviors. All examples of &quot;natural&quot; intelligence are embodied agents so it&#x27;s difficult to imagine any other kind of intelligence.<p>The large transformer models we have now are really the low hanging fruit. When you have a lot of data and compute, the easiest way to scale is to train a supervised model with ground truth labels. In contrast it&#x27;s much harder to train an RL agent, where you have to design the environment and keep track of its state.<p>Language modelling is a problem where you can get arbitrarily close to perfect but never actually achieve it. Without some kind of grounding in vision&#x2F;proprioception there will always be gaps in understanding. When they start scaling GATO-like models I think we&#x27;ll be a lot closer to human-like intelligence.
评论 #31566176 未加载
评论 #31568548 未加载
评论 #31565868 未加载
ZeroGravitas将近 3 年前
&gt; The scaling hypothesis regards the blessings of scale as the secret of AGI: intelligence is ‘just’ simple neural units &amp; learning algorithms applied to diverse experiences at a (currently) unreachable scale. As increasing computational resources permit running such algorithms at the necessary scale, the neural networks will get ever more intelligent.<p>I think this is what I believe i.e. that animals and humans are just evolved machines, with no divine spark. Not sure that I agree that it&#x27;s unpopular or that it allows you to make decades long predictions on progress that&#x27;ll come true.<p>I&#x27;m also not sure why you&#x27;d want to do this if smaller models are meeting your needs. Feels a bit like the future of flight being predicted as a man with flapping wings rather than a jet engine.
评论 #31569016 未加载
评论 #31568761 未加载
rossdavidh将近 3 年前
On the one hand, it is clearly true that much of the requirements for neural networks (in silicon) to achieve intelligence of a general sort, is to simply be big enough.<p>On the other hand, anyone acquainted with the facts of biological brain size vs. intelligence, can see that this cannot be all that is going on. Bison are not clearly more intelligent than crows, not even when the part of the brain involved in operating the body parts and processing raw sensory data is excluded. Something other than scale must be involved, even if scale is a necessary condition.
评论 #31565662 未加载
评论 #31564736 未加载
评论 #31564661 未加载
评论 #31564794 未加载
RodgerTheGreat将近 3 年前
Even if neural nets continue to come up with adequate solutions to challenging problems (sometimes, hopefully, maybe), at the end of the day the best-case scenario is a model that cannot be audited, holistically understood, or trusted with decisions of any gravity.<p>If beating the strongest human Go players with a self-trained ML model was detonating a fission bomb, pervasive automation with zero accountability is a nuclear winter.
评论 #31565567 未加载
评论 #31565628 未加载