TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

The cost to train an AI system is improving at 50x the pace of Moore’s Law

254 pointsby kayzaalmost 5 years ago

13 comments

solidasparagusalmost 5 years ago
Resnet-50 with DawnBench settings is a very poor choice for illustrating this trend. The main technique driving this reduction in cost-to-train has been finding arcane, fast training schedules. This sounds good until you realize its a type of sleight of hand where finding that schedule takes tens of thousands of dollars (usually more) that isn&#x27;t counted in cost-to-train, but is a real-world cost you would experience if you want to train models.<p>However, I think the overall trend this article talks about is accurate. There has been an increased focus on cost-to-train and you can see that with models like EfficientNet where NAS is used to optimize both accuracy and model size jointly.
评论 #23742426 未加载
calebkaiseralmost 5 years ago
This is an odd framing.<p>Training has become much more accessible, due to a variety of things (ASICs, offerings from public clouds, innovations on the data science side). Comparing it to Moore&#x27;s Law doesn&#x27;t make any sense to me, though.<p>Moore&#x27;s Law is an observation on the pace of increase of a tightly scoped thing, the number of transistors.<p>The cost of training a model is not a single &quot;thing,&quot; it&#x27;s a cumulative effect of many things, including things as fluid as cloud pricing.<p>Completely possible that I&#x27;m missing something obvious, though.
评论 #23740807 未加载
评论 #23740163 未加载
评论 #23741046 未加载
评论 #23740255 未加载
评论 #23744675 未加载
评论 #23741125 未加载
lukevpalmost 5 years ago
What are some domains that a solo developer could build something commercially compelling to capture some of this $37 trillion? Are there any workflows or tools or efficiencies that could be easily realized as a commercial offering that would not require massive man hours to implement?
评论 #23741293 未加载
评论 #23741694 未加载
评论 #23741544 未加载
评论 #23741625 未加载
评论 #23743493 未加载
评论 #23742304 未加载
anonualmost 5 years ago
Ark Invest are the creators of the ARKK [1] and ARKW ETFs that have become retail darlings, mainly because they&#x27;re heavily invested in TSLA.<p>They pride themselves on this type of fundamental, bottom up analysis on the market.<p>It&#x27;s fine.. I don&#x27;t know if I agree with using Moore&#x27;s law which is fundamentally about hardware, with the cost to run a &quot;system&quot; which is a combination of customized hardware and new software techniques<p>[1] <a href="https:&#x2F;&#x2F;pages.etflogic.io&#x2F;?ticker=ARKK" rel="nofollow">https:&#x2F;&#x2F;pages.etflogic.io&#x2F;?ticker=ARKK</a>
评论 #23740244 未加载
gchamonlivealmost 5 years ago
I remember this article from 2018: <a href="https:&#x2F;&#x2F;medium.com&#x2F;the-mission&#x2F;why-building-your-own-deep-learning-computer-is-10x-cheaper-than-aws-b1c91b55ce8c" rel="nofollow">https:&#x2F;&#x2F;medium.com&#x2F;the-mission&#x2F;why-building-your-own-deep-le...</a><p>Hackernews discussion for the article: <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=18063893" rel="nofollow">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=18063893</a><p>It really is interesting how this is changing the dynamics of neural network training. Now it is affordable to train a useful network on the cloud, whereas 2 years ago that would be reserved to companies with either bigger investments or an already consolidated product.
评论 #23739869 未加载
评论 #23740951 未加载
ersieesalmost 5 years ago
I would really like a thorough analysis on how expensive it is to multiply large matrices, which is the most expensive part of a transformer training for example according to the profiler. Is there some Moore’s law or similar trend?
评论 #23740519 未加载
mellosoulsalmost 5 years ago
It is regrettable if an equivalent to the self-fulfilling prophecy of Moore&#x27;s &quot;Law&quot; (originally an astute observation and forecast, but not remotely a law) became a driver&#x2F;limiter in this field as well, even more so if it&#x27;s a straight transplant for soundbite reasons rather than through any impartial and thoughtful analysis.
评论 #23740061 未加载
gxxalmost 5 years ago
The cost to collect the huge amounts of needed to train meaningful models is surely not growing at this rate.
gentleman11almost 5 years ago
Despite nvidia vaguely prohibiting users from using their desktop cards for machine learning in any sort of data center-like or server-like capacity. Hopefully AMDs ml support &#x2F; OpenCl will continue improving
评论 #23739733 未加载
sktguhaalmost 5 years ago
Does it mean that the cost to train something like gpt3 by OpenAI will reduce from 12 million dollars to less next year ? If so how much will it reduce to ?
m3kw9almost 5 years ago
It was probably because very inefficient to begin with.
评论 #23739946 未加载
bra-ketalmost 5 years ago
&quot;AI&quot; is not really appropriate name for what it is
seek3r00almost 5 years ago
tl;dr: Training learners is becoming cheaper every year, thanks to big tech companies pushing hardware and software.