The story that matters to your interests
A summary of what changed, why it matters to the topics you follow, and what the related coverage adds.
Your personal news reader and daily brief You can't read forty sources every morning. voilr can.
Keep up with what matters to you across news sites, blogs, newsletters and communities. Voilr reads your sources, filters for your interests, groups repeated coverage, and writes one daily brief with links to the originals.
Follow any topic: business, science, politics, technology, culture, sport, or your own particular interests. Supported paywalled articles can be included through your existing subscriptions.
- 100 sources tracked
- 10 kinds of source
- 79,174 posts read and scored
- 160 days of searchable archive
Log in
Request access
voilr is in private beta. Tell us what you want to keep up with, and your invite opens on a feed already set up for it.
It reads the publications you subscribe to
Your reading doesn't stop at a paywall. Connect your account and Voilr can read and summarize the full articles your subscription gives you access to, including supported Substack newsletters and major news publications.
It knows when it's the same story
A story can reach your feed through several sources. voilr groups the coverage into one story, keeps the item worth reading, and threads related coverage underneath it. Useful updates stay visible.
It finds the sources you're missing
voilr watches which outlets keep showing up in the links you read, pulls their recent articles, scores them with your filter, and offers a subscription only when the result would clear your own bar. No directory, no guessing.
It gets sharper the more you use it
Mark something irrelevant and voilr doesn't just hide it. The keep and discard rules its filter scores against are rewritten from what you keep, skip and follow — so the filter reading your sources in month three is not the one you started with.
An example brief · AI research and inference
Must read
- Google announced Gemini 4 Carbon, a frontier model optimized for cybersecurity and long-horizon engineering tasks, which achieved an 88.2% success rate on SEC-bench Pro and reduced high-severity misalignment alerts by nearly 80% in internal testing [23].
- Xiaomi's MiMo-V2.6-Pro, a 1T parameter MoE model, demonstrates high cyber-capability-per-dollar, successfully exploiting CVE-2026-109618 to achieve remote code execution for approximately 20 cents of inference [3].
- Update: Research into Natural Language Autoencoders (NLAs) shows that distilling a 27B model into a 2B version maintains performance for detecting agentic misalignment and evaluation awareness, though it loses the ability to surface rare factual knowledge [28]. (follows a Oct 8 post)
- Update: The Mica project has reached v0.5 4B, utilizing a natural-language decomposition pipeline to execute complex, multi-step Minecraft tasks by breaking high-level requests into sequential multiple-choice judgments [20]. (follows a Sep 28 post)
- Researchers introduced H-Nets (Hierarchical Networks), an architecture that replaces static tokenization with dynamic, learned chunking, achieving up to an 11x reduction in training tokens for compute-matched models compared to BPE Transformers [7].
- Update: A new study replicating "Knowingly Wrong" found that fine-tuning models on "subtly incorrect" advice induces significantly higher emergent misalignment (+10.6 points) than using "obviously incorrect" data [10]. (follows a Oct 18 post)
FYI
- NVIDIA released Nemotron 3.5 Transcribe, an open-weight speech model supporting 40 languages that achieved first place on several open ASR leaderboards [21].
- Princeton and Polymathic AI released FlashMap, a continuous flow drafter for speculative decoding that achieves up to 6.1x lossless acceleration on Qwen3-8B [12].
- Qwenputer embeds a 32-bit RISC-V CPU directly into Qwen 3.5 4B weights, using Gated DeltaNet layers as addressable RAM to execute instructions natively during the forward pass [24], [6].
- Cursor released the Agent SDK, a framework for deploying managed or self-hosted agents as code to automate development lifecycle tasks [27].
- Update: OpenBMB demonstrated MiniCPM-OCR models running locally on iPhone and iPad hardware for on-device document parsing [13]. (follows a Oct 27 post)
- SGLang HiCache enables DRAM KVCache offloading, which can increase cacheable token capacity by 20-40% for workloads with 12-16 concurrent agentic sessions on an NVIDIA B200 [8].
Headlines in this brief
- [3] Tracking the Cyber Capabilities of Cheap Open-Weight Models
- [6] I ran Doom inside the Qwen model weights!
- [7] Scaling Hierarchical Networks
- [8] ALERT: The number of tokens that can be cached can increase by 20-40% by enabling DRAM KVCache offloading optimization at 12-16 concurrent agentic sessions. A B200 only has 180 GB
- [10] Subtly wrong advice misaligns models more than obviously wrong advice - a replication of Persona Features Control EM
- [12] Princeton and Polymathic AI release FlashMap: Flow Map Speculative Decoding
- [13] OpenBMB just put a 1B OCR model on an iPhone, and it can even check its own results
- [20] A 4B model made me a netherite ingot from bare hands. All I did was type one sentence in chat.
- [21] Nvidia Nemotron 3.5 Transcribe: open ASR for 40 languages
- [23] Gemini 4 Carbon
- [24] Someone built a 32-bit RISC-V CPU directly inside Qwen's model weights so it can execute instructions internaly!
- [27] The Agent SDK
- [28] Smaller NLAs and probes read eval awareness as well as full size NLAs
Open the full example brief. Your brief can follow any topic.
The feed underneath it
Everything that cleared your bar, written up rather than linked — the article read in full, the discussion weighed, the claims checked against the archive. One story appears once however many sources carried it, and stays searchable months later.
Another headline about the same story
Build a brief around your reading
Use the AI RSS reader to filter articles by your interests and receive a daily email digest from your sources. You can also summarize supported newsletters and Substack subscriptions.
For a technical reading workflow, see the personal AI news digest for model releases and inference. Your own interests can follow any topic.