Hacker Newsnew | past | comments | ask | show | jobs | submit | rbren's commentslogin

This should be labeled humor or taken down. I did laugh though

If you're interested in driving coding agents with code, check out the OpenHands Software Agent SDK [1]

We need to define agents in code, and drive them through semi-deterministic workflows. Kick subtasks off to agents where appropriate, but do things like gather context and deal with agent output deterministically.

This is a massive boost in accuracy, cost efficiency, AND speed. Stop using tokens to do the deterministic parts of the task!

[1] https://github.com/OpenHands/software-agent-sdk


"conversation.send_message("Write 3 facts about the current project into FACTS.txt.")"

why tf would i ever need this


Some people seem to be convinced by logical reframings, like "if you jump into a woodchipper you die, but if 50% of people jump into the woodchipper they all survive"

A logical reframing is not equivalent though! We know everyone else gets the same frame, and most of the problem is predicting what other people will do when presented with this particular two-button frame.


Considerations:

* many people (at least toddlers, people with dementia) are going to press blue roughly by accident. See the lizardman constant

* other people will not want to be responsible for any deaths and will press blue out of a sense of moral imperative

* many other people are going to take this into account and vote blue out of hopes we can save everyone

You should vote blue.


The first point is interesting. You could fork the question over this and have a few variants:

1.) The pure form where the button presses and restricted to legal agents (i.e. people with credible legal standing over their choices). 2.) The mixed form with the caveat listed here inclusive of all humans whether they are even physically capable of pushing a button. 3.) you could also go for a more expansive scenario that takes 2 to the extreme and includes animals as well.

1.) gets to the game theoretic form of the question. 2 muddies things, and 3 sets up a case for blue since the non agentic voters asymptote to 50-50 and a slim edge is morally preferable to killing half.


You don't even have to go that far from the original question. If instead of the entire world being a single game, if you have hundreds of millions of sub-games where 9 random people are placed within, what should you do?

Surely some of those groups are going to be filled with selfish red pickers. Should the kind coordinating players still go blue? All the red pickers are going to lie that blue is sensible. I suspect that more coordinators will die in this way than the always blue pickers if every coordinating player went red.

So now the full-world version only has the law of large numbers on their side, but they have no way of knowing just what percentage of the population is a selfish red picker. Going for team blue is the much riskier option that can yield catastrophe.


Why would a red picker ever lie about it? If I can get all 8 of my fellow players to pick red then we’re all safe. If it’s a button I’ll just break the blue button or wire it to red.


A selfish player will claim that they will coordinate with the group, and then vote red in private. A coordinating player will pick what the group chooses, whether that be red or blue. You are talking about a coordinating player here. Yes, in this case if all players agree to red, it's obvious you should all pick red. It's completely safe.


With 3, especially if the animals outnumber humans, you’d first want to do some research into animal psychology to see whether red or blue has an edge for animals.


That is true in isolation, but the reason we study problems like this one is to try to gain insight into our society (or our minds) and in our society, toddlers and people with dementia have guardians that make important decisions for them. Consequently, even after your comment, I'm still struggling to see how this toy problem or game sheds any light on anything I care about. Contrast that with prisoner's dilemma, Newcomb's problem or the ultimatum game, which sheds a lot of light.

But this is HN, so people are going to discuss it just because it is fun to discuss it.


Puzzles like this are based on assumptions like all participants are rational adults with their full faculties.


This one explicitly is not.


It’s a made up toy problem. It exists for fun. The stated problem has some implicit assumptions. But you can rejigger the rules and assumptions to tweak the incentives and ethics. That’s the whole point. You could take the puzzle and apply it to a band of pirates held in a jail. That might make the outcome more obvious. Or you could imagine what would happen if the voting order were sequential. These are all just different formalisms that are fun to speculate over, but the rules can be interpreted many ways.


Yes, but those are different thought experiments from this one.


Forgetting how to code is not a bad thing. Forgetting how to make software is.


Lots of negativity towards k8s in here. It's always funny to me when $WILDLY_POPULAR_TECH gets ripped apart like this, as though no one has ever had a positive experience with it. I've seen similar pile-ons for React, microservices, git, PHP, JavaScript, cloud services, really anything that's been adopted at scale.


It’s only natural that seeing frequent complaints mostly happens for tech that has high adoption. Stuff that nobody uses doesn’t get many complaints.


HN has had a hate boner for K8s for as long as I can remember.

In my experience, K8s is a million times better than legacy shit it is usually replacing. The Herokus, the Ansible soup, the Chef/Puppet soup before that etc. The legacy infra that was held together by glue and sweat that everybody was afraid to touch.


As SRE, totally agree. Most companies I've been at where we implement K8S, which is around 30-50 VMs, ends up building their own, shittier Kubernetes. This blog post: https://www.macchaffee.com/blog/2024/you-have-built-a-kubern... is a favorite of mine.


"But it's my legacy shit and I understand it and I haven't taken the time to learn something new because that's scary so I'll just piss on anything attempting to replace it."

Human nature, really.


people love to make posts like "I used k8s to host my personal website from my homelab and it was too complicated!"


Good reminder to choose model-agnostic tooling!


Strong article! I’ve been using the engine/car analogy for a while now.

If you want to play with the basic building blocks of coding agents, check out https://github.com/OpenHands/software-agent-sdk


I still think every developer should be building their own IDE

https://github.com/rbren/personal-ai-devbox


I still strongly believe every developer should be vibecoding their own cowork/openclaw/devin

Here are the prompts I use for my AI environment, though it's changed a bunch since the last snapshot

https://github.com/rbren/personal-ai-devbox


Neat! I might give it a try.

What do you mean by interfaces in "These interfaces can do literally anything on the host machine. You're responsible for your own security"?

Also, your backdooring image links to a 404.


The prompts contain e.g. a terminal UI, which gives you root access to the machine. If someone can access that UI and its backend, the can do whatever they want! So make sure to put it behind a firewall or basic auth or something else.


Thanks for the link. You mention security; is the _average_ developer safer going with OpenClaw?


It's probably just as hard to secure OpenClaw as this, but you'll find better tutorials for securing OpenClaw


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: