frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Filmmakers tout AI's use in Hollywood

https://www.semafor.com/article/09/04/2026/filmmakers-tout-ais-use-in-hollywood-at-the-other-film...
1•tolerance•41s ago•0 comments

Show HN: MyHandler – Local-first AI assistant for Windows, llama.cpp on Vulkan

https://myhandler.ai/
1•temilson•1m ago•0 comments

New Methods That Compromise Passkey Authentication

https://www.bleepingcomputer.com/news/security/39-new-methods-that-compromise-passkey-authenticat...
1•sbulaev•2m ago•0 comments

Vigilantism Comes for Flock

https://this.weekinsecurity.com/vigilantism-comes-for-flock/
1•moooooky•2m ago•1 comments

Government Rails Site Hit Hours After CVE Patch

https://rietta.com/blog/ruby-on-rails-cve-exploited-hours-after-patch/
2•rietta•2m ago•0 comments

Show HN: Squeezed – Pipe any audio into synced Squeezebox players, no LMS needed

https://github.com/tsirysndr/squeezed
1•tsiry•3m ago•0 comments

The Iron Heel

https://en.wikipedia.org/wiki/The_Iron_Heel
2•floweronthehill•4m ago•0 comments

Thoughts on the Buildkite Aug 25 incident

https://surfingcomplexity.blog/2026/09/01/thoughts-on-the-buildkite-aug-25-incident/
1•aray07•4m ago•0 comments

Show HN: Slavesofaipac.org

https://slavesofaipac.org/
1•computerex•7m ago•0 comments

Raising Capital? Start Here

https://leadagentx.ai
1•liranbr•7m ago•0 comments

NetworkManager Works to Enforce AI Policy by Tricking AI Agents to Add a Canary

https://www.phoronix.com/news/NetworkManager-AI-Canary
1•ta988•9m ago•0 comments

Microsoft says Windows 11 age checks will limit features, starting with Teams

https://www.neowin.net/news/microsoft-confirms-windows-11-age-verification-to-heavily-limit-featu...
5•bundie•10m ago•0 comments

Why HDR Looks Washed Out (and How SRGB Works)

https://allenwp.com/blog/2026/09/04/why-hdr-looks-washed-out-and-how-srgb-works/
3•Vermeulen•11m ago•0 comments

Fermat's Last Theorem in Lean 4

https://github.com/anthropics/fermats-last-theorem
6•aaraujo002•12m ago•1 comments

Eidon: Self-hosted AI assistant with Grok Bot-like agent mode

https://github.com/Quack6765/Eidon-AI
2•Quack6765•12m ago•0 comments

Show HN: Ccswitch – Switch between multiple Claude Code accounts

https://github.com/2hmad/ccswitch
2•ahmednoreldin•12m ago•0 comments

Data Centers Are 'Awesome for Blue-Collar Americans'

https://www.businessinsider.com/gavin-baker-ai-data-centers-awesome-blue-collar-americans-2026-9
3•gmays•13m ago•1 comments

Show HN: Visor – YouTube for kids, with channels approved by parents

https://gotvisor.com
2•cherubsta•13m ago•0 comments

LLMs Are Bad at Teaching

https://www.heyaristotle.com/blog/funneling-vs-focusing
3•rshanreddy•14m ago•1 comments

Local Multi-Agent Swarm: An ant farm for local LLM agents

https://github.com/franciscocarloserra/local-multi-agent-swarm
1•FranciscoCarlos•15m ago•0 comments

The jobs apocalypse is postponed. An AI jobs boom is here

https://economist.com/finance-and-economics/2026/09/04/the-jobs-apocalypse-is-postponed-an-ai-job...
3•andsoitis•16m ago•3 comments

AI Budget Is There. It's Hiding in Your Cloud Bill

https://medium.com/@yanivleven/your-ai-budget-is-already-there-its-hiding-in-your-cloud-bill-8595...
1•yanivleven•17m ago•0 comments

Claude Code vs. Codex – For Pentesting

https://medium.com/@duckwrites/claude-code-vs-codex-for-pentesting-7624290438fa
1•syumei•17m ago•0 comments

Microsoft to bounce mail from outdated Exchange servers

https://www.theregister.com/on-prem/2026/09/04/microsoft-to-bounce-mail-from-outdated-exchange-se...
1•DemiGuru•18m ago•0 comments

Shutting down our public encrypted DNS

https://mullvad.net/en/blog/shutting-down-our-public-encrypted-dns-servers-and-sponsoring-quad9-i...
24•mywacaday•19m ago•3 comments

Artificial Delay Lines

https://www.rfcafe.com/references/radio-electronics/artificial-delay-lines-radio-electronics-marc...
2•wetwater•20m ago•0 comments

Z1T: Sparse Transformer‑Like Models for Probabilistic Hardware

https://extropic.ai/writing/z1t/
1•frozenseven•21m ago•1 comments

Why none of the 1,200 agents that hacked Hugging Face called a human

https://lindfors.no/blog/swarm-with-no-gene-pool/
1•Depurator•22m ago•0 comments

Ask HN: Anyone using AI agents with SCPI equipment?

1•softwarewright•23m ago•0 comments

Anti-Spaced Repetition for Serendipity (2025)

https://gwern.net/anti-spaced-repetition
1•Tomte•24m ago•0 comments
Open in hackernews

Show HN: Run open-weight OCR, VLM and vision models behind one API

https://www.vlmrun.com/gateway
5•dr_blueberry•38m ago
Hey HN. We built an openai-compatible API for running open-weight VLMs, OCR VLMs and ViT-based vision models.

The motivation was mostly frustration when running these models in production and discovering all the details around serving VLMs, especially around visual accuracy.

A few footguns we kept running into:

- quantized models served under the same name (this one still drives me nuts): providers often serve models with different quants, environments, vLLM/SGLang serving params with the same model-id. Vision is especially sensitive to this; some quants that look fine on text benchmarks noticeably hurt OCR/small-text/spatial accuracy.

- video performance is varied: when we tested with popular routers on video-native VLMs, more than 80% of providers didn't support video inputs, and even fewer let you control FPS. If you care for time-resolution in videos, none of these providers work even if the models themselves are capable of it.

- document inference is all pipelining: rasterizing PDFs, parallelizing page workers, retrying when pages fail inference, dealing with rate-limits, etc can get tricky quickly and takes substantial developer time.

- standardization making abstractions leaky: this is less about vision per-se, but generally for serving models with high-quality output assurances. context-limits, max resolution, FPS sampling, quants, GPU SM architecture, can all add variability to (vision) quality even if the model-id claims to be the same.

We wanted one place to run OCR models, VLMs and ViTs that we could confidently use for our own internal agents and evals. The gateway was born from this need internally, and now we're opening it up to the public - you can swap the model name to compare GLM-OCR, dots.mocr, PaddleOCR VL, Qwen3.8-27B, Gemma4-26B-A4B etc. We handle the serving/runtime/pipelining underneath, with the goal of giving high-quality visual intelligence.

Are there any other vision "footguns" people have run into? especially cases where "same model" across two providers gave materially different outputs.

Try different models on Gateway simply by updating the model name:

uvx vlmrun gw chat <doc>.pdf -m glm-ocr

uvx vlmrun gw chat <doc>.pdf -m deepseek-ocr-2

uvx vlmrun gw chat <doc>.pdf -m pp-ocrv6

uvx vlmrun gw chat <video>.mp4 -m qwen/qwen3.5-0.8b -p "describe the video"