frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Unitree G1 robot accidentally kicks its trainer in groin

https://scienceclock.com/robot-kicks-trainer-groin/
1•ashishgupta2209•3m ago•0 comments

Go: Testing Frameworks and Mini-Languages (2025)

https://matttproud.com/blog/posts/testing-frameworks-and-mini-languages.html
1•homebrewer•3m ago•0 comments

Show HN: Turn Any Android App Into An API

https://revrse.ai
1•arsrev•4m ago•0 comments

Lockheed Martin to more than triple Patriot missile production capacity

https://www.reuters.com/business/aerospace-defense/lockheed-martin-more-than-triple-patriot-missi...
1•toomuchtodo•5m ago•1 comments

I Played with the Lego Smart Brick

https://www.theverge.com/tech/855520/i-played-with-the-lego-smart-brick
1•ravenical•8m ago•0 comments

LLM Poetry and the "Greatness" Question

https://hollisrobbinsanecdotal.substack.com/p/llm-poetry-and-the-greatness-question
1•jger15•8m ago•0 comments

An Honest Review of Go

https://benraz.dev/blog/golang_review.html
1•benrazdev•11m ago•0 comments

Everything you never wanted to know about file locking (2010)

https://apenwarr.ca/log/20101213
3•SmartHypercube•11m ago•0 comments

Does Your Email Provider Know What a "Joejob" Is? (2016)

https://nxdomain.no/~peter/does_your_email_provider_know_what_a_joejob_is.html
1•speckx•12m ago•0 comments

Epiplexity: Rethinking Information for Computationally Bounded Intelligence

https://arxiv.org/abs/2601.03220
1•dcre•13m ago•0 comments

The CEO Who Couldn't Pick Two

https://federiconagy.substack.com/p/the-ceo-who-couldnt-pick-two
1•randoments•13m ago•1 comments

Supreme Court Increasingly Favors the Rich, Economists Say

https://www.nytimes.com/2026/01/05/us/politics/supreme-court-study-rich-poor.html
6•rainingmonkey•15m ago•1 comments

Policynuts – Democratizing Democracy

https://policynuts.com
1•dznodes•16m ago•1 comments

Show HN: GetIntel – Sales intelligence focused on signals, not just emails

https://getintel.ai/
1•tarang8811•16m ago•1 comments

The Cerdá Plan

https://en.wikipedia.org/wiki/The_Cerd%C3%A1_Plan
2•altilunium•17m ago•0 comments

Confluence Cloud Outage

https://confluence.status.atlassian.com
2•joeyhage•18m ago•0 comments

Music Assistant 2.7 – Taking over the airwaves

https://www.home-assistant.io/blog/2025/12/17/music-assistant-2-7/
1•lode•18m ago•0 comments

Show HN: Scribble – streaming-friendly transcription engine in Rust

https://github.com/itsmontoya/scribble
1•hatchjosh•19m ago•0 comments

Show HN: The Coasean Nightmare – Why Seamless AI is a Cognitive/Legal Liability

1•daikikadowaki•19m ago•0 comments

Ask HN: What data modeling approaches work for convs. AI systems?

1•tanmaydesh5189•21m ago•0 comments

Venturing into GitHub First Time

https://github.com/SeanB88/Cypher_project
1•seanb88•22m ago•1 comments

Octopus-inspired 'synthetic skin' changes colour and texture on demand

https://www.nature.com/articles/d41586-026-00052-7
1•Brajeshwar•22m ago•0 comments

Products now powered by Raspberry Pi

https://www.raspberrypi.com/news/hundreds-of-products-now-powered-by-raspberry-pi/
1•Brajeshwar•22m ago•0 comments

Rv Clean-Install

https://spinel.coop/blog/rv-clean-install/
1•amalinovic•23m ago•0 comments

DTAP – super simple testing protocol for infrastructure testing

1•melezhik•23m ago•1 comments

Generate Apple Music Playlists with ChatGPT

https://www.macrumors.com/how-to/generate-apple-music-playlists-with-chatgpt/
1•7777777phil•24m ago•0 comments

Everything you might have missed in Java in 2025

https://www.jvm-weekly.com/p/everything-you-might-have-missed-886
5•mihau•24m ago•1 comments

C# is programming language of the year 2025

https://www.tiobe.com/tiobe-index/
5•vips7L•25m ago•2 comments

Neon Exchange: Gaming and Crypto and AI

https://neonexchange.co/en/map
1•kengeo•25m ago•0 comments

Our Changing Planet, as Seen from Space

https://e360.yale.edu/digest/nasa-satellite-images-2025
3•YaleE360•27m ago•0 comments
Open in hackernews

Alpie Core: a 32B reasoning model trained and served at 4-bit

https://huggingface.co/169Pi/Alpie-Core
2•ChiragArya•1d ago

Comments

ChiragArya•1d ago
We have been experimenting with whether strong multi-step reasoning actually requires a full-precision scale.

Alpie Core is a 32B parameter model trained and served entirely at 4-bit precision. Instead of training in FP16 and compressing later, we optimized the model end-to-end for low-precision reasoning.

Despite the constraint, it performs competitively on reasoning and coding benchmarks like GSM8K, BIG-Bench Hard, and SWE-Bench Verified, while using far less memory and infrastructure.

We are especially interested in feedback on long-context behaviour, reasoning stability across retries, and how it performs inside agent or tool-using workflows.

ChiragArya•1d ago
I’m one of the builders behind Alpie Core. Happy to answer questions, clarify benchmarks, or share more technical details about the model, training setup, or deployment choices.