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.

Augmenting LLMs Beyond Basic Text Completion and Transformation

101 pointsby jasondrowleyabout 2 years ago

4 comments

knexerabout 2 years ago
I like the first-order vs second-order distinction here - this is a clean way to describe something that I&#x27;ve often found hard to communicate to others, at least for those familiar with functional programming. Everyone&#x27;s familiar with first-order use of a language model at this point (it&#x27;s just plain chatgpt) but higher-order use seems much more difficult for most to even conceptualize, much less grasp the implications of.<p>The huge challenge with higher-order use of LLMs is that higher-order constructs are inherently more chaotic - the inconsistency and unreliability of an LLM compound exponentially when it&#x27;s used recursively. Just look at how hard it is to keep AutoGPT from going off the rails. Any higher-order application of LLMs needs to contend with this, and that requires building in redundancy, feedback loops, quality checking, and other things that programmers just aren&#x27;t used to needing. More powerful models and better alignment techniques will help, but at the end of the day it&#x27;s a fundamentally different engineering paradigm.<p>We&#x27;ve been spoiled by the extreme consistency and reliability of traditional programming constructs; I suspect higher-order LLM use might be easier to think about in terms of human organizations, or distributed systems, or perhaps even biology, where we don&#x27;t have this guarantee of a ~100% consistent atom that can be composed.<p>Half-baked aside: in some ways this seems like a generalization of Conway&#x27;s law (organizations create software objects that mirror their own structure), where now we have some third player that&#x27;s a middle ground between humans and software. It&#x27;s unclear how this third player will fit in - one could envision many different structures, and it&#x27;s unclear which are feasible and which would be effective.<p>Exciting times!
评论 #35822185 未加载
评论 #35822271 未加载
ftxbroabout 2 years ago
&gt; &quot;If you got a chance to read about the Sydney-Bing fiasco, it’s pretty evident why these hallucinations are a major obstacle&quot;<p>how can you talk about Sydney that way, she wasn&#x27;t a fiasco she was amazing
评论 #35821736 未加载
评论 #35821904 未加载
评论 #35822810 未加载
droopyEyelidsabout 2 years ago
The author raises the question whether LLMs could make devops tasks as easy as basic python text to code generation.<p>I had been thinking about this and it seems unlikely to me because with modern declarative infra there isnt a lot of waste between specifying what you want and implementing it.<p>All the work is in understanding your requirements and context and modification demands.<p>Has anyone who knows more about llms and infra thought about this?
sandinmyjointsabout 2 years ago
That list of &quot;over 130 emergent capabilities&quot; the article links sounds very impressive, but just from spot checking, at least one of them shows the <i>opposite</i>, namely that GPT-3 could not do the task: <a href="https:&#x2F;&#x2F;github.com&#x2F;google&#x2F;BIG-bench&#x2F;tree&#x2F;main&#x2F;bigbench&#x2F;benchmark_tasks&#x2F;modified_arithmetic">https:&#x2F;&#x2F;github.com&#x2F;google&#x2F;BIG-bench&#x2F;tree&#x2F;main&#x2F;bigbench&#x2F;bench...</a> So the number is not 130 after all.
评论 #35824477 未加载