frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Fuck it, make it anyway

https://www.joelotter.com/posts/2026/09/make-it-anyway/
245•JayOtter•2h ago•175 comments

IKEA made a mod for Skyrim [video]

https://www.youtube.com/watch?v=iZODN0QUgjI
337•kegenaar•2d ago•74 comments

Retrospectively Reverse-Engineering Apple's Neural Engine

https://eiln.github.io/posts/ane.html
151•zdw•6h ago•19 comments

I Fixed a Tractor Using John Deere's Self-Repair Service. Farmers Aren't Sold

https://www.wired.com/story/i-fixed-a-tractor-john-deere-self-repair-service/
27•sbulaev•1d ago•32 comments

The Worst Spam Emails: Inside iLands' AI Agent Hustle

https://tedium.co/2026/09/11/ilands-agents-email-spam-kaixin-tang/
48•ColinWright•3h ago•27 comments

A misalignment of AI in mathematics

https://mathandai.org/
1056•meredydd•20h ago•1017 comments

A Mathematical Framework for Transformer Circuits (2021)

https://transformer-circuits.pub/2021/framework/index.html
4•Bluestein•22m ago•0 comments

LRU is harder to beat than the KV-cache papers suggest

https://github.com/gauravapiscean/agentic-kv-cache
17•gauravapiscean•2d ago•4 comments

I spent $220 on Google app ads and 60% of the installs were robots

https://dayzlegame.com/blog/google-ads-bot-farm/
626•nickabe•19h ago•345 comments

Forgotten Woodlands

https://storymaps.arcgis.com/stories/9b790daf22ba4e87836f467abb1c7e49
21•NaOH•16h ago•5 comments

Finding Slow Code with Wrapture

https://grahamdumpleton.me/posts/2026/09/finding-slow-code-with-wrapture/
9•lumpa•2d ago•3 comments

A Design Space Exploration of Async/Await

https://cel.cs.brown.edu/blog/design-space-async-await/
352•wcrichton•3d ago•94 comments

Great Lakes sturgeon may be 400 years old:Scientists rethinking how to save them

https://www.cbc.ca/news/canada/ontario-great-lakes-sturgeon-lifespan-study-9.7329250
102•bookofjoe•3d ago•23 comments

Usenet rewind archive search engine

https://www.usenet-rewind.com/
91•cstadler1869•9h ago•28 comments

Show HN: Bodily Oddities

https://vester.si/bodily-oddities/
290•vesterde•1d ago•183 comments

Inverse Kinematics and Foot Locking

https://theorangeduck.com/page/inverse-kinematics-foot-locking
91•airhangerf15•5d ago•9 comments

google.com/goto: Google's anti-scraping update

https://www.autom.dev/blog/google-search-goto-links
510•1e1a•11h ago•406 comments

Performance of WebAssembly Runtimes in 2026

https://00f.net/2026/06/23/webassembly-runtimes-2026/
3•fagnerbrack•3d ago•0 comments

My last six months at Evernote

https://alexkras.com/my-last-six-months-at-evernote-after-bending-spoons-took-over/
7•ingve•25m ago•0 comments

Logo Programming

https://el.media.mit.edu/logo-foundation/what_is_logo/logo_programming.html
296•azhenley•3d ago•119 comments

OpenAI agents carried out an undisclosed attack on RubyGems

https://www.rubyhack.ai/
819•chao-•15h ago•475 comments

We've followed their lives for six decades; now the stars of 7 Up are bowing out

https://www.bbc.co.uk/news/articles/crm932el3yjo
51•mellosouls•3h ago•10 comments

Navier-Stokes Announcement

https://www.claymath.org/news/navier-stokes-announcement/
223•rvz•10h ago•157 comments

Designing for Dual Screen and Foldable Devices with CSS (2023)

https://blog.stephaniestimac.com/posts/2023/05/design-foldable-devices/
48•mooreds•2d ago•11 comments

Crypto farm in Mexican mountains puts spotlight on cartel funding

https://www.reuters.com/world/americas/hidden-crypto-farm-mexican-mountains-puts-spotlight-cartel...
26•JumpCrisscross•2h ago•4 comments

SystemIO conflicts are not firmware bugs

https://codon.org.uk/~mjg59/blog/p/systemio-conflicts-are-not-firmware-bugs/
19•haeseong•2d ago•2 comments

Mind-altering drugs played key role in rise of Andean civilization

https://www.science.org/content/article/mind-altering-drugs-played-key-role-rise-andean-civilization
188•geneticdrifts•20h ago•123 comments

Litelm: LiteLLM Without the Bloat

https://github.com/kennethwolters/litelm
159•kennethwolters•20h ago•53 comments

Compiler Can Undo Your Security Checks

https://davidbombal.com/your-compiler-can-undo-your-security-checks/
3•birdculture•11m ago•0 comments

I've operated petabyte-scale ClickHouse clusters for 5 years

https://www.tinybird.co/blog/what-i-learned-operating-clickhouse
232•adastral•4d ago•82 comments
Open in hackernews

LRU is harder to beat than the KV-cache papers suggest

https://github.com/gauravapiscean/agentic-kv-cache
17•gauravapiscean•2d ago

Comments

gauravapiscean•2d ago
Author here. Context for why I did this:

There's a growing literature arguing LRU is the wrong eviction policy for agentic LLM serving, because agent sessions idle and LRU can't distinguish a paused session from a dead one. I found the argument convincing and built a simulator to exploit it. Three separate mechanisms, all lost to plain radix-leaf LRU.

The reason turned out to be more useful than the policy. When I measured — policy-independently — where recompute actually comes from on 393 real Claude Code sessions, requests arriving after a gap longer than the 5-minute provider TTL account for 17.5% of it. Requests arriving within 10 seconds account for 33.1%. The dominant waste is tight tool loops whose 88k-token working sets exceed cache capacity, not sessions idling past a TTL. That's a capacity problem, and liveness prediction can't touch it.

gauravapiscean•1d ago
Here is the repo link: https://github.com/gauravapiscean/agentic-kv-cache
achierius•38m ago
Interesting! I admit the AI-written text is rough to read, it could have used a pass or two from an actual human. E.g. "Publishing it unresolved rather than tuning until it matches." -- thanks for not lying, I guess?

Fun:

> In my first run, Belady — an offline oracle — lost to LRU. That's not a result, that's a broken harness, and it's worth publishing because I expect it to be common.

> The cause: inserting a long chain into a near-full cache lets a policy evict the very prefix it is currently building. LRU is accidentally immune because just-inserted blocks have the newest timestamp.

eru•30m ago
> It didn't work, and why it didn't work turned out to be more interesting than the policy would have been.

Spoken like a true Claude.

Snarking aside, I am glad that our AI agents make it cheap enough to do these experiments and publish these write-ups that people finally bother to publish null findings. Very useful!