frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

OpenPawz Conductor Protocol

https://github.com/OpenPawz/openpawz/blob/main/reference/conductor-protocol.mdx
1•gotham64•36s ago•1 comments

Show HN: Unread, turns your unread newsletters into a daily podcast

https://app.unread.live
1•benfosterdev•38s ago•0 comments

Show HN: See Your Money in 3D

https://moneyvisualiser.com
1•KhazP•48s ago•0 comments

Show HN: Verified n8n community node for Anytype

https://github.com/splch/n8n-nodes-anytype
1•splch•1m ago•0 comments

Claude Code vs. Codex (Nate B Jones) [video]

https://www.youtube.com/watch?v=09sFAO7pklo
1•JnBrymn•3m ago•0 comments

Show HN: Synclippy – Ephemeral rooms for sharing text or files

https://synclippy.ujjwalvivek.com
1•ujjwalvivek•3m ago•0 comments

PKGSmith

https://pkgsmith.app/
1•Fogh•6m ago•0 comments

Plan 9 from User Space

https://9fans.github.io/plan9port/
1•tosh•7m ago•0 comments

GPT-5.4 code-golfs GPT-2

https://twitter.com/hansonwng/status/2030000810894184808
1•tosh•7m ago•0 comments

Re-creating the complex cuisine of prehistoric Europeans

https://arstechnica.com/science/2026/03/recreating-the-complex-cuisine-of-prehistoric-europeans/
1•apollinaire•8m ago•0 comments

Oracle and OpenAI drop Texas data center expansion plan

https://www.reuters.com/business/oracle-openai-end-plans-expand-texas-data-center-site-bloomberg-...
2•speckx•9m ago•0 comments

Palera1n Jailbreak Compiled and Run on a Samsung Galaxy S3 (PostmarketOS, ARMv7)

https://github.com/noxbitx/s3ra1n/tree/main
2•noxbit•9m ago•0 comments

Eval awareness in Claude Opus 4.6's BrowseComp performance

https://www.anthropic.com/engineering/eval-awareness-browsecomp
1•gcampbell•10m ago•0 comments

Show HN: I built an international calling platform/Android App

https://voklit.app
1•ahmgeek•11m ago•0 comments

If flip-phones can make a comeback, can Flash do the same?

https://disassociated.com/flip-phones-comeback-can-flash/
1•speckx•11m ago•0 comments

An AI disaster is getting ever closer

https://www.economist.com/briefing/2026/03/05/an-ai-disaster-is-getting-ever-closer
3•bookofjoe•14m ago•1 comments

Ecological Imperialism

https://kschroeder.substack.com/p/my-library-ecological-imperialism
1•MaysonL•14m ago•0 comments

Python 'Chardet' Package Replaced with LLM-Generated Clone, Re-Licensed

https://developers.slashdot.org/story/26/03/06/1614252/python-chardet-package-replaced-with-llm-g...
1•jakobdabo•16m ago•0 comments

Asana: Scaling our invalidation pipeline (Part 1)

https://asana.com/inside-asana/scaling-invalidation-pipeline-part-1
1•Bringoff•16m ago•0 comments

Host Claude Artifacts on your own domain

https://artifact.ninja/
1•mooreds•16m ago•0 comments

Issue: The Consciousness Question Is Being Asked Wrong

https://medium.com/@sheldonksalmon/issue-the-consciousness-question-is-being-asked-wrong-0e6d2eae...
1•sheldonksalmon•18m ago•0 comments

Obstructive sleep apnoea costs UK and US economies £137B a year

https://www.theguardian.com/society/2026/feb/24/obstructive-sleep-apnoea-costs-uk-and-us-economie...
1•PaulHoule•18m ago•0 comments

Show HN: GPT-5.4 is interesting for one boring reason: fewer retries

https://clipnotebook.com/blog/gpt-5-4-fewer-retries-real-work
3•diddddy•22m ago•1 comments

Jank is off to a great start in 2026

https://jank-lang.org/blog/2026-03-06-great-start/
3•todsacerdoti•23m ago•0 comments

Swift at scale: building the TelemetryDeck analytics service

https://swift.org/blog/building-privacy-first-analytics-with-swift/
1•frizlab•24m ago•0 comments

GLP-1 drugs may fight addiction across every major substance

https://theconversation.com/glp-1-drugs-may-fight-addiction-across-every-major-substance-accordin...
3•gmays•25m ago•1 comments

Watch BYD's 5-min Flash Charging in action on the new Seal 07 EV

https://electrek.co/2026/03/06/byds-new-seal-07-ev-with-5-min-flash-charging-video/
1•breve•25m ago•0 comments

Reflections on Using Acme (2020)

https://blog.jacobvosmaer.nl/0006-acme/
1•tosh•26m ago•0 comments

Show HN: Graph-Oriented Generation – Beating RAG for Codebases by 89%

https://github.com/dchisholm125/graph-oriented-generation
1•dchisholm125•26m ago•0 comments

Most of My Coding Is Now Agentic

https://www.justinmath.com/most-of-my-coding-is-now-agentic/
2•speckx•28m ago•0 comments
Open in hackernews

100% tool routing across 3,146 apps–zero LLM at runtime, self-improving, 7ms

https://github.com/glyphh-ai/model-pipedream/blob/main/white-paper.md
2•timmetime•8h ago

Comments

timmetime•8h ago
Show HN:

The scale problem first: Pipedream has ~10,000 actions. Full catalog = 750K tokens. GPT-4o context = 128K. The LLM literally cannot load the tools.

We inverted the architecture.

LLM runs once, offline, at build time — generates every possible way a human might phrase an intent. 22,614 exemplars compiled into an 8.5MB HDC vector space. At runtime: pure math, no LLM, 7ms.

Results across 85,125 test queries:

  First-pass action accuracy:    89.6%
  Action accuracy (with ASK):    100%
  App accuracy:                  100%
  Silent errors:                 0%
  Latency p95:                   <13ms
  Tokens per query:              0
  Model size:                    8.5MB, no GPU
  Improves with use:             Yes
The 10.4% that trigger ASK aren't failures. The system asks rather than guesses, the correct action is always in the candidate set, and every resolved ASK strengthens the model via Hebbian reinforcement. No retraining. No labeling pipeline. The production model is the learning model.

GPT-4o hits 98.5% accuracy — when given a pre-filtered shortlist of 200 actions and human-readable action keys. It can't do app selection across 3,146 apps. HDC does the whole thing in 7ms and gets better with every use.

We benchmarked honestly — full methodology in the paper including where GPT-4o wins.

Patent pending: US 63/969,729 Covers: build-time LLM→HDC pipeline, confidence gating, Hebbian self-improvement without retraining.

White paper + benchmark + Docker quickstart: https://github.com/glyphh-ai/model-pipedream