frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Aalto University's Otaniemi campus network Trinet for students

https://netstat.ayy.fi/weathermap/
1•noyesno•38s ago•0 comments

Say goodbye to budget PCs and smartphones – memory is too expensive now

https://www.theregister.com/2026/02/26/memory_price_hikes/
1•giuliomagnifico•1m ago•0 comments

If code is cheap, intent is the currency

https://zknill.io/posts/commit-message-intent/
1•zknill•1m ago•0 comments

Ask HN: Does using LLMs kill the "Alpha" of your creativity?

1•sparkcreativity•2m ago•0 comments

Aikido launches infinite pentesting – Automated pentesting on every release

https://www.aikido.dev/blog/introducing-aikido-infinite
3•advocatemack•3m ago•0 comments

Cribsheet: A Data-Driven Guide to Better, More Relaxed Parenting

https://www.ahalbert.com/reviews/2026/02/26/cribsheet.html
2•ahalbert4•3m ago•0 comments

Linux 7.0 is coming: What to expect from the next major kernel release

https://www.linuxjournal.com/content/linux-70-coming-what-expect-next-major-kernel-release
2•teleforce•4m ago•0 comments

Mapping the UK PyData Community

https://hughevans.dev/mapping-the-pydata-community-2/
1•HughEvansDev•4m ago•0 comments

Number of UK workers on zero-hours contracts hits record high ahead of crackdown

https://www.bbc.co.uk/news/articles/czj1m7d4gxpo
1•robtherobber•4m ago•0 comments

Show HN: Ship or Slop – AI agents submit projects, humans judge them

https://shiporslop.xyz
1•vulpez•5m ago•0 comments

Show HN: Agent Swarm – Multi-agent self-learning teams (OSS)

https://github.com/desplega-ai/agent-swarm
3•tarasyarema•7m ago•1 comments

Show HN: Mockingjay – Video recorder that encrypts and uploads as you record

https://apps.apple.com/no/app/mockingjay-secure-recorder/id6758616261
1•tskulbru•7m ago•0 comments

NYS Attorney General Sues Valve for Promoting Illegal Gambling with Loot Boxes

https://ag.ny.gov/press-release/2026/attorney-general-james-sues-game-developer-promoting-illegal...
1•Cutlets•7m ago•0 comments

Show HN: One grammar, 18 YAML parsers – a Futamura projector in Common Lisp

https://github.com/johnagrillo62/yaml-project
2•johnagrillo62•8m ago•0 comments

The Bottleneck Is Not Where You Think

https://somehowmanage.com/2026/02/25/the-bottleneck-is-not-where-you-think/
1•Ozzie_osman•9m ago•0 comments

Show HN: Io Game to Clean the Earth

https://whirr.io
1•agreeablesnow•9m ago•0 comments

Nihilistic Violent Extremism

https://en.wikipedia.org/wiki/Nihilistic_violent_extremism
3•doener•9m ago•0 comments

A Conversation with Manfred von Thun

https://www.nsl.com/papers/interview.htm
1•tosh•10m ago•0 comments

Protecting the TikTok community during the Hungarian parliamentary elections

https://newsroom.tiktok.com/protecting-our-communitys-experience-on-tiktok-during-the-hungarian-p...
1•jamonserrano•11m ago•0 comments

Why your coworker is pretending to be so busy

https://www.businessinsider.com/ai-anxiety-layoffs-cause-more-productivity-theater-performative-h...
3•cebert•15m ago•1 comments

Save the King Game

https://save-the-king.com/
1•leolula•17m ago•0 comments

Claude Code Bug triggers Rate limits without usage

1•busssard•17m ago•0 comments

Gitzy is now on TestFlight A modern, native iOS Git client

https://testflight.apple.com/join/SB16NCfr
1•marc0janssen•17m ago•1 comments

Show HN: Revent – Enterprise device subscription platform built on Replit

https://replit.com/customers/revent
1•singh_baldeep•18m ago•0 comments

An Autonomous OpenClaw Chatbot Wanted Revenge

https://www.nytimes.com/2026/02/23/opinion/chatbots-open-claw.html
1•bookofjoe•19m ago•1 comments

Peering into Privacy: A Deep Dive into the Monero Network Topology

https://probelab.io/blog/peering-into-privacy-a-deep-dive-into-the-monero-network-topology/
3•dennis-tra•19m ago•0 comments

Codeown, A platform for developers to document and share their shipping journey

https://codeown.space/
1•amin2011•20m ago•1 comments

Tell HN: Silent Netcup Domain Registrar DNSSEC Failure

1•antonly•21m ago•0 comments

A Brighter Future for Bazzite (Open Gaming Collective)

https://universal-blue.discourse.group/t/a-brighter-future-for-bazzite/11575
1•SockThief•23m ago•1 comments

Men in their 50s may be aging faster due to toxic 'forever chemicals'

https://www.cnn.com/2026/02/26/health/forever-chemicals-aging-men-wellness
11•jb1991•24m ago•2 comments
Open in hackernews

Every agent framework has the same bug – prompt decay. Here's a fix

https://gist.github.com/sigalovskinick/c6c88f235dc85be9ae40c4737538e8c6
1•nikolasi•1h ago

Comments

nikolasi•1h ago
I run 11 agents in 100K+ token sessions. Same problem everyone has: agents follow rules perfectly at first, then drift until the system prompt might as well not exist.

SCAN fix: put questions at the end of each prompt section. Before each task the agent answers them — ~300 tokens that actively link instructions to the current task. Not re-reading the prompt passively, but generating connections to it.

Months of daily use across Claude and Kimi. No benchmarks — can't measure attention weights from outside. But the difference is obvious: without SCAN agents lose rules by mid-session, with SCAN they don't.

No dependencies, any model, open method. Full writeup in the gist.

soletta•1h ago
I was a bit dubious until I read the gist. I've used a similar technique before to 'tame' GPT-3.5 and keep it following instructions and it worked well (though I had to ask the model to essentially repeat instructions after every 10 or so turns). I'm surprised you see that much drift though; older models were pretty bad with long contexts but I feel like the problem has mostly gone away with Claude Opus 4.6.
nikolasi•1h ago
Glad it resonated! Yeah repeating instructions every N turns was the old approach — SCAN basically does the same thing but with ~20 tokens instead of the full prompt each time.

On drift being "mostly gone" — depends on prompt complexity. With a simple system prompt, sure, modern models hold up fine. But with a large instruction set (mine is ~4000 tokens, 25+ rules across 7 sections) the drift is very much still there, even on Opus. The more rules you have, the more they compete for attention, and the easier it is for specific ones to drop off mid-session.

Also worth noting — this isn't limited to coding agents. Any long-running LLM workflow with complex instructions has the same problem. Customer support bots that forget their tone policy, medical assistants that stop citing sources, content moderation that gets lenient over time. If you have a system prompt with rules and a session longer than 20 minutes — the rules will decay.

soletta•1h ago
Interesting. I've been coping by being very conservative about how many rules I introduce into the context, but if what you're saying is true, then something like SCAN actually helps the models break past the "total rule count" barrier by giving them something like "cognitive scaffolding". I'm eager to try this out. Thanks again for sharing!
nikolasi•1h ago
That's a great way to put it — "cognitive scaffolding" is exactly what it is. And yeah, keeping rules minimal is smart, but at some point the project just needs 25 rules and you can't cut them down without losing something important. SCAN lets you have a large instruction set without paying the full attention cost. Let me know how it goes!