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。

Ask HN: When will we hit a limit on LLM performance?

5 点作者 ksj211412 个月前
All the AI founders (e.g., Dario Amodei) seem to believe that we&#x27;re nowhere near the end of seeing performance improvements in LLMs as they are trained on more data (i.e., LLM scaling laws) - at least that&#x27;s what they say publicly, but they obviously have skin in the game. Curious what knowledgeable people think who are not incentivized to make optimistic public statements?<p>What I really want to know is, assuming capital &#x2F; compute is not a constraint, will be continue to see order of magnitude improvements in LLMs, or is there some kind of &quot;technological&quot; limit you think exists?

2 条评论

henry_pulver12 个月前
As far as I (ex-ML researcher) know, the main technological case that LLM performance will hit a limit is due to the amount of text data available to train on is limited. The ways these scaling laws work is they require 10x or 100x quantity of data to see major improvements.<p>This isn&#x27;t necessarily going to limit it though. It&#x27;s possible there are clever approaches to leverage much more data. This could either be through AI-generated data, other modalities (e.g. video) or another approach altogether.<p>This is quite a good accessible post on both sides of this discussion: <a href="https:&#x2F;&#x2F;www.dwarkeshpatel.com&#x2F;p&#x2F;will-scaling-work" rel="nofollow">https:&#x2F;&#x2F;www.dwarkeshpatel.com&#x2F;p&#x2F;will-scaling-work</a>
评论 #40685391 未加载
smartician12 个月前
Research seems to suggest we need exponential training data volume increases to see meaningful performance gains: <a href="https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2404.04125" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;2404.04125</a><p>Personally I think we&#x27;ve already hit a ceiling.
评论 #40686845 未加载