frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Draft-tree decoding on Apple Silicon: M4 vs. M5/M6 and bit-exact Metal

https://blog.yuhuachen.com/blog/2ArXyvXoSTOQ9nFQtp3Fz
1•yuhuabill•2m ago•0 comments

MotherDuck, Meet Excel

https://motherduck.com/blog/motherduck-duckdb-meet-excel/
1•Tomte•3m ago•0 comments

Why we're blind to the color blue (2021)

https://calebkruse.com/10-projects/seeing-blue/
1•sammatilda•4m ago•0 comments

Inkling 2 Small on Nvidia B300: 5.1x to 7.5x faster expert GEMMs, same bits

https://www.rightnowai.co/blog/bit-identical-fp4-moe-kernels-on-blackwell
3•OsamaJaber•7m ago•0 comments

Three-way translation merge: keeping translated pages in sync with an LLM

https://maxmart.github.io/posts/three-way-translation-merge/
1•maxmart•8m ago•0 comments

Show HN: Cloud-drift-report – an open-source driftctl alternative for AWS

https://github.com/orchesterra/cloud-drift-report
1•akshaykilam•8m ago•0 comments

A Slinky on an Escalator Is a Weirdly Good Test of LLM Intelligence

https://morethanamachine.com/posts/slinky-escalator-benchmark/
1•reddmachine•9m ago•1 comments

Gemini: Can't prevent training or review even with no history retained

https://support.google.com/gemini/answer/13594961?hl=en
2•iotage•9m ago•1 comments

Hex Marks the Spot

https://www.museumofplay.org/blog/hex-marks-the-spot/
1•jonbaer•13m ago•0 comments

A History of Clojure (2020)

https://dl.acm.org/doi/pdf/10.1145/3386321
1•tosh•15m ago•0 comments

Qlaude – Claude Code against your own model server

https://github.com/trail-openers/qlaude
1•velmu•21m ago•0 comments

The Manufacturing of Novelty

https://www.ft.com/content/99711b1a-3e54-4349-9c81-769cf7c96b97
1•JumpCrisscross•24m ago•0 comments

Probabilistic data structures in Go, part 2

https://antonz.org/probabilistic-data-structures-2/
1•mfrw•24m ago•1 comments

California train burglars carefully executed heists for 'staggering' $22M haul

https://www.theguardian.com/us-news/2026/nov/21/california-arrests-train-heists
2•ljf•27m ago•0 comments

Ask HN: Who Owns API Drift

1•Nalin_2408•29m ago•1 comments

Can Psychedelics Improve Decision-Making?

https://nautil.us/can-psychedelics-improve-decision-making-1286863
2•the-mitr•36m ago•0 comments

The Engineering Manager's role is under attack (again)

https://www.manager.dev/newsletter/the-engineering-manager-s-role-is-under-attack-again
8•fagnerbrack•40m ago•0 comments

Ask HN: Is Anthropic being public better or worse for your org?

3•ares623•41m ago•1 comments

Web Search on Amazon Bedrock expands language support and country-aware search

https://aws.amazon.com/about-aws/whats-new/2026/11/web-search-amazon-bedrock/
1•yackrru•46m ago•0 comments

Notes on the discrete Fourier transform (DFT)

https://eli.thegreenplace.net/2026/notes-on-the-discrete-fourier-transform-dft/
10•kouosi•49m ago•1 comments

Show HN: App Pixel3D – Dream to 3D Reality

https://apps.apple.com/hu/app/pixel3d-dream-to-3d-reality/id6479245806
1•cflsdf•49m ago•0 comments

Postgres 19 is out: try it on Neon

https://neon.com/blog/postgres-19-is-out-try-it-on-neon
2•yackrru•50m ago•0 comments

Show HN: iOS App Dancing Particle is an interactive app

https://apps.apple.com/us/app/dancing-particle/id6444665083
1•cflsdf•50m ago•0 comments

Amazon is complaining about NYC Council on shipping labels, too

https://www.theverge.com/tech/1039564/amazon-nyc-council-packages-shipping-label
2•burnt-resistor•54m ago•0 comments

Show HN: Prompt Upgrader – a no-signup tool that scores and rewrites AI prompts

https://madgpt1.github.io/prompt-upgrader/
1•MadGPT•1h ago•0 comments

AI resolved all null pointer dereferences in C

https://kapteko.com/ai-resolved-all-null-pointer-dereferences/
1•fessguid•1h ago•1 comments

The Indignity of Makework

https://poisonandoak.wordpress.com/2026/11/21/the-indignity-of-makework/
2•wburglett•1h ago•0 comments

IronWAL: High performance, high durability, deterministic WAL

https://github.com/excsn/ironwal
1•excsn•1h ago•1 comments

How to read a CVSS score and decide what to patch

https://p99log.com/reading-cvss-scores/
1•collinstonui•1h ago•0 comments

Why agent swarms get better at optimizing code as they grow

https://frontier.replit.com/throwing-headcount-at-the-problem/
2•DonAtimator•1h ago•0 comments
Open in hackernews

A Slinky on an Escalator Is a Weirdly Good Test of LLM Intelligence

https://morethanamachine.com/posts/slinky-escalator-benchmark/
1•reddmachine•9m ago

Comments

reddmachine•9m ago
I want to preface that I've done LLM evals in a proper setting before. This is far sillier. It's derivative of the "pelican riding a bicycle" test, something simple that shows how capable an LLM is at perceiving intent, dealing with physics and general aesthetics. I run this for every new model now and it gives me a decent idea of what to expect.