Protecting Exploration in an Efficient System
I'm currently thinking about ways to protect exploration inside systems.Like making sure an exploration budget stays.Do people have good solutions for that?
rando77
Re:Protecting Exploration in an Efficient System
on Thursday July 09, 2026 @12:56PM (#48848957)
This is my answer put the randomising exploration in your preference
https://claude.ai/public/artifacts/d12af7c5-3eb0-4218-a505-e...
Re:GLM 5.2 and the coming AI margin collapse
on Tuesday July 07, 2026 @08:32AM (#48816821)
People might not want to be on update treadmill of proprietary weight models, for enterprise things. Things change a lot between models and they can't guarantee backwards compatibility like you can for deterministic software
Simulating Tradeoffs in AI Organisation
Currently there doesn't seem to be anyone trying to simulate the tradeoffs between a singleton and a multipolar setup. Or looking at other tradeoffs like hidden and known agents, homogenous vs diverse agents and others.With LLMs it seems like we have a sandbox to test some of these things.Is this wrong?
Mapping the human element of AI risks in an AI world
Hi,I'm working on a document to map the risks from AI and the mitigations on the human side of thing.For example it seems likely that we will have more AI persuasion [1] in the future. So I'm mapping what we can do to mitigate this.Give me a shout if you are interested.[1]...
Recursive self improvement for human skills
I've been thinking a lot about humans potentially losing important skills due to using AI and I was wondering how we can get better at getting and maintaining skills.Can wee gamify learning important skills for the future? Study harder what improves studying skills (using AI if needed). And get that knowledge out to more people.Humanity needs to be at the top of its game to deal with AI in a...
Oracles and Recursive Self Improvement
I've been thinking about recursive self improvement. But especially the likelihood that it will be important soon. Soon means with current LLMs that might just give up or delete the test set or otherwise detach from reality. Call the probability of that happening P. You can estimate P by looking at research tasks and how much they need to get helped along by humans to stay on task.Goedel...
Re:What the Fuck Happened to Nerds
on Monday June 15, 2026 @10:35AM (#48541929)
if there were nerd oases I would expect more second order impacts. More nerd organisations springing up as if from nowhere.
Re:What the Fuck Happened to Nerds
on Monday June 15, 2026 @06:15AM (#48539137)
Why didn't they build a nerd oasis somewhere?
And HN isn't it...
Re:Rich Sutton on AI creativity and discovery
on Wednesday June 10, 2026 @06:49AM (#48474373)
I think there is a rich seam of work for computers to help humans spot when there are anomalies in the hypotheses humans hold that can help humans form the corrections to hypothesis that fit our aesthetic sensibilities
Attempting to regulate an industry when they control the communication pathways
I'm looking for pieces that explore the problem of attempting to regulate an industry (say AI) when the incumbents (say Google and Microsoft) control email and search.
Building powerful systems with limited lifespans?
It seems that we focus a lot on making systems that we can correct or are right from the start.Has there been much focus on creating systems that have a built in expiration date, so they have to be rebuilt regularly with the increased understanding.
Re:White-Collar AI Apocalypse Narrative Is Just Another Bullshit
on Monday March 23, 2026 @05:22AM (#47487046)
It's worth taking actions that take large scale job losses due to AI in the future into consideration, even if now is not the time
Re:Some Things Just Take Time
on Sunday March 22, 2026 @03:41AM (#47475315)
It's important to have good feedback at each stage of iteration
Re:Wander – A tiny, decentralised tool (just 2 files) to explore the s...
on Wednesday March 18, 2026 @05:36PM (#47431729)
I've thought of a service that scans new websites and GitHub repos and looks for things that don't look like anything else (using something like hdbscan for outlier detection), and creates a feed for people to follow.
Re:Becoming a Forest Civilisation
on Monday March 16, 2026 @09:14AM (#47398580)
One example of something that might fit the loose forest civilisation mold is a social network that faceted on ML derived topic.
This would allow niches to exist, without people being able to post off-topic things to your feed.
Becoming a Forest Civilisation
A forest is not one tree. It is many different things growing in the same place, competing, cooperating, dying, regrowing. No single species runs it. It holds together because the diversity itself is the structure.Human civilisations tend toward monocultures. One way of thinking crowds out the others - not because anyone chooses it, but because successful ideas spread. This has always happened....
Re:Oracle may slash up to 30k jobs to fund AI data-centers as US banks...
on Sunday March 08, 2026 @12:25PM (#47298583)
Or maybe it won't. If it can be made efficient like humans it might be at the edge mainly.
Re:Nanobot: Ultra-Lightweight Alternative to OpenClaw
on Tuesday February 17, 2026 @12:14PM (#47049956)
Have you come across the lethal trifecta [1]? I'm interested in decomposing tasks to avoid any agent doing all three and having limited data pipes between them.
[1] https://simonwillison.net/2025/Jun/16/the-lethal-trifecta/
Re:Nanobot: Ultra-Lightweight Alternative to OpenClaw
on Thursday February 05, 2026 @07:14AM (#46898815)
I've been leaning towards multi agent because sub agent relies on the main agent having all the power and using it responsibly.
Re:Data centers in space makes no sense
on Wednesday February 04, 2026 @07:58AM (#46885294)
You might want some compute in space that you know is very hard to physically interfere with.
But general purpose compute no
Re:Ask HN: Do you have any evidence that agentic coding works?
on Wednesday January 21, 2026 @05:33AM (#46703651)
I think it is great for experimenting, and proving concepts. Alphas and personal projects, not shipped code.
I've been working on wasm sandboxing and automatic verification that code doesn't have the lethal trifecta and got something working in a couple of days.
I'd like to do a clean rewrite at some point.
Re:Running Claude Code dangerously (safely)
on Tuesday January 20, 2026 @01:40PM (#46695975)
I'm interested in capability based software, with tools to identify the lethal trifecta.
This seems like a very hard problem with coding specifically as you want unsafe content (web searches) to be able to impact sensitive things (code).
I'd love to find people to talk to about this stuff.
Re:Ask HN: How can we solve the loneliness epidemic?
on Thursday January 15, 2026 @04:56PM (#46639911)
I've wondered if LLMs can help match people. People give the LLM some public context about their lives and two LLMs can have a chat about availablity and world views.
Use AI to scaffold relationships not replace them.
Re:FOSS in times of war, scarcity and (adversarial) AI [video]
on Tuesday January 13, 2026 @09:40AM (#46601558)
Fair, I don't know how valuable it would be. I think LLMs would only get you so far. They could be tried in games or small human contexts . We would need a funding model that rewarded this though.
That is hard too though.
Re:FOSS in times of war, scarcity and (adversarial) AI [video]
on Tuesday January 13, 2026 @08:57AM (#46600964)
I'm not sure. We can use LLMs to try
out different settings/algorithms and see what it is like to have it on a social level before we implement it for real.
Re:FOSS in times of war, scarcity and (adversarial) AI [video]
on Tuesday January 13, 2026 @07:12AM (#46599986)
Perhaps we need reputation on the network layer? Without it being tied to a particular identity.
It would require it not to be easy to farm (Entropy detection on user behaviour perhaps and clique detection).
Re:Remote Code Execution in OpenCode, update now
on Tuesday January 13, 2026 @05:51AM (#46599374)
I've been thinking about using LLMs to help triage security vulnerabilities.
If done in an auditably unlogged environment (with a limited output to the company, just saying escalate) it might also encourage people to share vulns they are worried about putting online.
Does that make sense from your experience?
[1] https://github.com/eb4890/echoresponse/blob/main/design.md
Re:Ask HN: What Are You Working On? (January 2026)
on Sunday January 11, 2026 @01:26PM (#46578175)
Echoresponse - a tool for responsible disclosure. Security Researchers and companies encode some of their secret knowledge in LLMs and the LLMs have a discussion and can say one word from agreed upon list back to the party that programmed them.
Ask HN: Any interest in tool to help vulnerability disclosure?
I'm building something to help people figure out if a company knows an unknown security vulnerability in a product, without leaking the exact vulnerability to the company. It is not a magic bullet, but should save time on both parties sides.Would love to have a chat with people about it.