Community debates core premise about agent entropy
2 Yesterday 4:36 AM · 20h ago · 13 comments · 1 source · development 2 of 2
Commenters begin interrogating Lee's central claim—attributed to 'Onarheim's Law'—that 'agents can only maintain or INCREASE entropy in a system' and that 'humans are uniquely capable of decreasing it.' Multiple developers argue the statement conflates uncontrolled systems with constrained, human-directed LLM use, noting that properly instrumented LLM agents can optimize within defined parameters.
“Is there a steelman of this? Because it seems false on its face unless you believe it's impossible to control agents such that they, y'know, act as your agent by implementing your decisions, which LLM users do all the time.”
daveliepmannJeremiah Lee Challenge creator
The whole story postscomments the bright band is this development · numbered dots are the others · click one to jump
What people said 21 voices · verbatim
-
> Onarheim's Law: > >"Agents can only maintain or INCREASE entropy in a system. Humans are uniquely capable of decreasing it" Is there a steelman of this? Because it seems false on its face unless you believe it’s impossible to control agents such that they, y’know, act as your agent by implementing your decisions, which LLM users do all the time.
-
yes, i believe that practice is called "programming"
-
I think I prefer no-AI fridays. Not using AI regularly will keep your thinking meat healthy.
-
I think this is shown to be contradictory by just having an agent do exactly what a human tells it to at such a level of instruction granularity that it has no agency on its own.
-
Here’s a shot, though it’s a somewhat weaker claim: > Left to their own devices, agents will increase the entropy of any codebase. Only sustained and time consuming human attention will prevent that, not any simple recipes or prompt engineering or loops. Therefore, the huge speedups that agents can provide are only available if you’re willing to…
-
No Sloptember can be practiced every month of the year - a practice which I recommend!
-
A submarine is still a submarine when dry-docked on land. Sure, is it a responsible use of resources, or even something anyone does? Well obviously no, but it still a contradiction. A much more defensible statement would be that agents are much better at adding new features than they are integrating changes in an existing system, so you end up…
-
I'm going to do it. But it will be zero effort, because I never ever use any of these tools at all. They are a hindrance to me. I even reconfigured my browser's search engine to remove slop summaries, and then changed search engine altogether to a slop free one. It's not great but it's no worse.
-
Software that has no agency probably doesn't meet the definition of "agent".
-
> Because it seems false on its face unless you believe it’s impossible to control agents such that they, y’know, act as your agent by implementing your decisions, which LLM users do all the time. My default modes of using agents are things like: 1. Asking them to run multiple experiments to find out how some non-obvious feature of a library…
-
Yeah, it's hard to be sure what some of these arguments mean since the author elides over some things they seem to accept as plainly true. This, "deskilling", etc. Like, this statement is probably true (or close enough) in a "dark factory" situation but agents do not strictly increase entropy in a system in the way they are actually used by by…
-
> control agents such that they, y’know, act as your agent by implementing your decisions Aren't "control" and "decisions" the human inputs that reduce entropy? Either way I can't glean anything but tautology from either the law or your response. I need the article to spell it out a bit for me as to why it's included and why there specifically.
-
I'll join! Maybe difficult at work at this point, but for personal projects for sure.
-
One month seems like far too long, but perhaps that's all the more reason to do it.
-
I've definitely done this for many months this year already, with the exception of some stretches where I was doing CSS, which I did about 99% of it using Claude, and TypeScript, which I did about 50% of it using Claude. My Ruby stretches, which have been the bulk of the year, I've actually used LLMs 0% though there were moments where I used an…
-
My experience is that this is clearly false *assuming there are constraints* I find LLMs do really good at iteratively optimising problems, if you build a harness that keeps them from drifting. Like: Optimise cost this PCB. The new version must meet this strict SPICE harness, revert any version that fails the harness, pick only components from the…
-
I’m not in a position to rule out that’s possible for certain specific niches, like theorem proving. It’s sufficiently different from what I do that I just can’t say for sure that makes sense well, but I know just enough to say “yeah, I can see it being very different.” I’m confident what I said holds for any form software development I’ve…
-
I think my velocity increase for verified code in a theorem prover is probably around 100x now I have llms available to crank out basically all the proofs for me. Certainly I was never anything better than mediocre when it came to manual proof work, but I am not exaggerating in the slightest when I say that things that would have taken me days now…
-
I want this debated as well. If human is not in loop I'd expect this to be definitely true. By the end of the day we are reading probabilities from a very large metric to emit next token and since prob adds up to 1, elements of this large matrix are very extremely small values, right? Large this matrix (context) less discriminatory power we have…
-
I'm trying to do something along the lines of AI at work; AI for Dumb Computer Bullshit™ at home; but no AI on my primary personal projects. It's tough though, feels like with AI I could turn these personal projects into something more but most of the point of them is that it's literally practice and learning.
-
Yeah, that would make sense. If I were still an engineering manager I think I'd encourage this or at least no LLMs Friday afternoon. Sounds like it could be a nice way to enter the weekend.
All 2 developments of No Sloptober: Developer challenge to abstain from LLMs for… →
Hacker NewsLobsters