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。

My objection(s) to the "LLMs are just next-token predictors" take

3 点作者 hummuscience5 个月前

1 comment

openrisk5 个月前
There is a fundamental fallacy here: equating mildly complex algorithms the fundamental workings of which are actually designed and well understood - even if not trivially explainable, to very complex biological and brain processes that are still very superficially understood.<p>This reveals at best a paucity of imagination. There is no reason to imagine that all complex systems irrespective of fundamental structure and spanning vastly different degrees of complexity are &quot;emerging&quot; anything like the same behaviors.<p>The explainability of LLM&#x27;s and so-called &quot;deep learning&quot; will be liberated when it disentangles itself from vague and forced anthropomorphism but the marketing allure is alas to strong.