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。

Turing Complete Transformers: Two Transformers Are More Powerful Than One

190 点作者 georgehill超过 1 年前

13 条评论

qsort超过 1 年前
These reviews are <i>brutal</i>. It&#x27;s basically science-speak for &quot;the paper is utter trash&quot;.<p>&quot;The main claim [...] is both somewhat obvious and previously already stated&quot;<p>&quot;Many pieces of writing are overly assertive and inaccurate.&quot;<p>&quot;I do not think it deserves spending half a page demonstrating that {0^n 1^n} is not in the regular language.&quot;
评论 #38922862 未加载
评论 #38920787 未加载
评论 #38921945 未加载
评论 #38923354 未加载
评论 #38922593 未加载
Imnimo超过 1 年前
The reviews are pretty harsh, but after reading the paper, I feel they may be too generous. Somehow this paper spends several pages proving things that are both trivial and irrelevant, and spends zero words explaining the architecture of their model. This is borderline crackpot territory.
评论 #38921447 未加载
评论 #38923267 未加载
davesque超过 1 年前
Something I noticed when skimming the paper that is also called out by one of the reviewers:<p>&quot;Moreover, in the submission the authors considered only Transformers with input and output of bounded lengths, which are quite strange since Turing machines do not pose constraints on the tape length. If the length is constrained in Transformers, they clearly do not match Turing machines.&quot;
评论 #38921097 未加载
Tainnor超过 1 年前
This response by the authors to the reviews to me looks like they didn&#x27;t really understand the objection:<p>&gt; First, it&#x27;s worth noting that different reviewers sometimes gave opposite critiques of the paper, e.g. Reviewer erh8: The conclusion in this paper is questionable... It contradicts to [1], which shows that Transformers are Turing-complete Reviewer bz3o: The main claim that transformers with finite attention span are not computationally universal is both somewhat obvious and previously already stated<p>If I&#x27;m reading the reviews correctly, the claim by both reviewers was that transformers <i>are</i> actually Turing complete, but one reviewer added that they&#x27;re &quot;obviously&quot; not Turing complete if you restrict their memory a priori (which I would agree is obvious). So there isn&#x27;t really a contradiction between the reviews.<p>From briefly skimming the paper, this does look indeed to me like researchers which aren&#x27;t really familiar with theoretical CS trying to handwave their way into something that looks ground-breaking. But while you absolutely can get away with vague-ish description in a more experimental part of CS, you absolutely can&#x27;t get away with it in computability theory - that field is rigorous, and basically maths.
评论 #38925155 未加载
DoubleDerper超过 1 年前
Hear me out. Try three transformers.
评论 #38920728 未加载
评论 #38920706 未加载
junipertea超过 1 年前
Is the discussion about the paper, or about how it was unilaterally rejected?
评论 #38920949 未加载
评论 #38921214 未加载
morthaur超过 1 年前
&gt; Let’s consider a simple problem that finite automata cannot solve: recognizing the language L = {anbn|n ≥ 0}. This language consists of strings with n ’a’s followed by n ’b’s. &gt; A regular expression or finite automaton can recognize strings in this language up to a certain length. For example, the regular expression a∗b∗ can recognize strings in L up to length 2.<p>That regex makes no guarantee that the number of a&#x27;s matches the number of b&#x27;s, which doesn&#x27;t match their language definition. I think they wanted (ab)*, which does, and can match any string in their language.
评论 #38932600 未加载
bee_rider超过 1 年前
Out of curiosity, what do people think of the comments to the reviewers by the authors?<p>I was pretty surprised to see them challenge the reviewers. Maybe open review is different, but I was trained to try find ways to defer to the reviewers or, basically, placate them if possible. It look like the authors have tried to argue back, for example finding a contradiction between the reviews… it seems like a risky strategy to me. Then again I haven’t ever received feedback this negative, thank goodness.
评论 #38922578 未加载
评论 #38922890 未加载
评论 #38925795 未加载
fmbishu超过 1 年前
Still going to Review the paper, but what about NTM-like transformers if turing is to be attained(external memory)
K0balt超过 1 年前
Looks like things are about to get interesting, between this work and iterative, self directed AI REPL type architecture.
m3kw9超过 1 年前
Why not as many as possible like how gpus have shader processors
fhackenberger超过 1 年前
Paper written by one transformer?!
krackers超过 1 年前
Dupe of <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=38917829">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=38917829</a> ?
评论 #38920864 未加载