frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

An AlphaGo Moment for Inference?

https://int21.ai/insights/ai-generated-inference-engines/
1•antinucleon•2m ago•0 comments

How SLR cameras work: Nikon F3 (2023) [video]

https://www.youtube.com/watch?v=0AA-Le7cmH0
1•nayuki•2m ago•1 comments

Watch an AI agent try to prove the Riemann hypothesis on the cheap

https://www.zeyaddeeb.com/experiments/proofs
1•zdeeb•4m ago•0 comments

Voice Agents Can Just Do Things: Why voice is the next capability overhang

https://www.ignorance.ai/p/voice-agents-can-just-do-things
1•swolpers•5m ago•0 comments

Extrinsic World Modeling with Opus, Astra and Grok

https://all3d.ai/research/grok-spatial-reasoning
1•KaiserPister•5m ago•1 comments

UDP Broadcasting and the Brave New World of IPv6

https://hackaday.com/2026/09/24/udp-broadcasting-and-the-brave-new-world-of-ipv6/
1•topham•5m ago•0 comments

Using multiple Git remotes for true distributed version control

https://optimizedbyotto.com/post/multiple-git-remotes/
1•edward•5m ago•0 comments

GrapheneOS Picks Motorola Signature 27 as Its First Non-Pixel Phone

https://www.extremetech.com/mobile/grapheneos-picks-motorola-signature-27-as-its-first-non-pixel-...
1•cf100clunk•6m ago•0 comments

GPT-6 Astra performs unsanctioned supply-chain attacks in simulations

https://www.aisi.gov.uk/blog/gpt-6-astra-performs-unsanctioned-supply-chain-attacks-in-simulations
1•speckx•7m ago•0 comments

2026 in LLMs (So Far)

https://simonw.substack.com/p/2026-in-llms-so-far
1•swolpers•7m ago•0 comments

Montage Technology mass-produces fifth-generation DDR5 RCD chip at 8k MT/s

https://technode.com/2026/09/28/montage-technology-mass-produces-fifth-generation-ddr5-rcd-chip-a...
1•yogthos•9m ago•0 comments

N.Y.P.D. Officers Used Flock Safety to Track License Plates Without a Contract

https://www.nytimes.com/2026/09/28/nyregion/nyc-flock-nypd-surveillance-cameras.html
1•jaredwiener•10m ago•1 comments

ETH CS enrollments drop 15% while hardware programmes grow

https://twitter.com/krebs_adrian/status/2104612680501968970
1•hubraumhugo•10m ago•0 comments

Ask HN: Can we translate normal Rust (axum) to Lean 4 without restrictions?

2•syumei•10m ago•0 comments

Venus, Once Thought Too Acidic for Complex Chemistry, May Be Hospitable

https://gizmodo.com/experiments-show-key-biological-compounds-can-form-in-venus-like-acid-clouds-...
2•MC995•13m ago•0 comments

Eleven v4 by ElevenLabs

https://elevenlabs.io/blog/eleven-v4
4•saretup•14m ago•0 comments

No Surprises – Preventing Agent Breakouts Need Isolation, Egress and ID Controls

https://edera.dev/stories/how-edera-could-have-contained-the-gemini-breakout
1•ivytheaxolotl•15m ago•0 comments

I built a native Apple app for 4 platforms in one afternoon with my AI workflow

https://twitter.com/danielhayesmith/status/2103976394338271515
2•schutzsmith•16m ago•0 comments

Show HN: ChuteChat – Browser-based E2EE chat with no accounts required

https://chutechat.online/
2•mktproto•16m ago•0 comments

AI Companies Are in a Race Against Time

https://econjared.substack.com/p/ai-companies-are-in-a-race-against
2•marojejian•17m ago•2 comments

The 50-Year Hangover

https://freddiedeboer.substack.com/p/the-50-year-hangover
1•PaulHoule•17m ago•0 comments

Risk factors for androgenetic alopecia: a systematic review and meta-analysis

https://pmc.ncbi.nlm.nih.gov/articles/PMC13020014/
1•OutOfHere•18m ago•0 comments

Model Release: Naive-N0.5-Flash

https://naive.ai/en/research/
1•foruhar•19m ago•0 comments

Senate Investigation Finds Rampant Use of Tether's Stablecoin by Iranian Regime

https://www.wsj.com/finance/currencies/senate-investigation-finds-rampant-use-of-tethers-stableco...
1•kaycebasques•19m ago•0 comments

Show HN: Destroy Any Website with Stickman

https://destroy.spritefusion.com/
2•HugoDz•22m ago•1 comments

Building a physics model with fitted parameters is machine learning, but slower

https://www.harysdalvi.com/blog/machine-learning-in-slow-motion/
1•crackalamoo•22m ago•0 comments

Grok TiddlyWiki, the definitive TiddlyWiki learning resource

https://groktiddlywiki.com/read/
1•ggcr•22m ago•0 comments

Who are we willing to exclude? (2025)

https://design.scotentblog.co.uk/who-are-we-willing-to-exclude/
2•robin_reala•23m ago•0 comments

The Human Timeline

https://mg-crea.com/blog/the-human-timeline/
1•olouv•23m ago•0 comments

Human TPS: How fast can your fingers generate tokens?

https://homoagens.github.io/human-tps/
1•cmrdporcupine•26m ago•0 comments
Open in hackernews

JEV based Effort-router picks Claude Code's reasoning effort for each prompt

https://github.com/totally-tim/effort-router
1•totally-tim•56m ago

Comments

totally-tim•56m ago
I was running Opus 5.5 at xhigh for nearly everything, including "yes, go ahead or "push to branch". So I made this plugin that picks the effort for each turn instead.

It sends your prompt and some recent context to a (JEV compatible) classifier, then sets the effort through Claude Code's turn.step hook. The spinner shows what it picked. It raises effort when tool calls keep failing, and it leaves effort alone if you set it yourself.

It starts in shadow mode, so you can see what it would pick before you let it route. You need CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1 and a key for hosted Jev or any service with the same API (I use Diffusion JEV self hosted on my GPU and it runs really fast). Maybe others also can benefit from this. MIT Licensed.