Feed aggregator

The AI model you choose to vibe code with can dramatically affect your final output.

The poll finds that AI is viewed less positively than ICE and President Donald Trump, and only more positively than Iran and the Democratic Party.

The US has unveiled a six pillar national cyber security strategy, with developing technological areas such as post-quantum cryptography and artificial intelligence from and centre.

Article URL: https://www.nonzero.org/p/iran-and-the-immorality-of-openai

Comments URL: https://news.ycombinator.com/item?id=47313242

Points: 1

# Comments: 0

Article URL: https://up.com.au/tree/

Comments URL: https://news.ycombinator.com/item?id=47313228

Points: 1

# Comments: 0

Hi HN, I'm the solo developer behind Ratschn.

I type a lot and got extremely frustrated with the current state of Mac dictation tools. Most of them are either heavy Electron wrappers, rely on cloud APIs (a privacy nightmare), or force you into a SaaS subscription for a tool that essentially runs on your own hardware. I wanted something that feels native, respects system resources, and runs entirely offline without forced subscriptions.

The stack is Rust, Tauri, and whisper.cpp. Here are the design decisions I made:

Model Size vs. Accuracy: Instead of using the smallest possible model just to claim a tiny footprint, the app downloads a ~490MB multi-language Whisper model locally on the first run. I found this to be the sweet spot for high accuracy (accents, technical jargon) to drastically reduce text correction time.

Hardware Acceleration: The downloaded model is compiled via CoreML. This allows the transcription to run directly on the Apple Neural Engine (ANE) and Metal on M-series chips, keeping the main CPU largely idle.

Memory Footprint: By using Tauri instead of Electron, the UI footprint is negligible. While actively running, the app takes up around 500MB of RAM. This makes perfect technical sense, as it is almost entirely the ~490MB AI model being actively held in memory to ensure instant transcription the millisecond you hit the global shortcut.

Input Method: It uses macOS accessibility APIs to type directly into your active window.

Business Model & Pricing: I strongly dislike subscription fatigue for local tools. There is a fully functional 7-day free trial (no account required). If you want to keep it, my main focus is a fair one-time purchase (€125 for a lifetime license). However, since I highly value the technical feedback from this community, I generated an exclusive launch code (HN25) that takes 25% off at checkout (dropping it to roughly €93 / ~$100).

Bug Bounty: Since I'm a solo dev, I know I might have missed some edge cases (especially around CoreML compilation on specific M-chips or weird keyboard layouts). If you find a genuine, reproducible bug and take the time to report it here in the thread, I will happily manually upgrade you to a free Lifetime license as a massive thank you for the QA help.

I'd love to hear your technical feedback on the Rust/Tauri architecture or how the CoreML compilation performs on your specific Apple Silicon setup. Happy to answer any questions!

Comments URL: https://news.ycombinator.com/item?id=47313205

Points: 3

# Comments: 0

Deadbranch identifies and removes old, unused git branches safely. It protects important branches, backs up everything it deletes, and lets you restore with one command.

v0.2.0 adds an interactive TUI (built with ratatui) with Vim-style navigation, fuzzy search, and visual range selection.

GitHub: https://github.com/armgabrielyan/deadbranch

Comments URL: https://news.ycombinator.com/item?id=47313168

Points: 1

# Comments: 0

Article URL: https://www.haaretz.com/middle-east-news/2026-03-08/ty-article/.premium/idf-officials-admit-military-struggling-to-intercept-hezbollah-drones/0000019c-ced2-d8f8-a9de-ffd358680000

Comments URL: https://news.ycombinator.com/item?id=47313162

Points: 4

# Comments: 0

I feel a bit paranoid now with the services I use. It seems like everyone is trying to harvest everything I do for AI training. For example, if I start using Notion, are they going to use everything to train some AI of theirs? I'm assuming Figma is training on what people create. It seems like Microsoft is going to be training on everything done using Recall.

It really shouldn't be so easy as burying an agreement in the terms somewhere to gain ownership over your data, but that's the way it is now. Where do we go from here? How do we use computers connected to the web while still maintaining privacy?

Comments URL: https://news.ycombinator.com/item?id=47313152

Points: 1

# Comments: 0

I kept opening Instagram to check a DM and losing 40 minutes to Reels. Deleting the app meant losing DMs and Stories. So I built a browser that loads the real mobile web versions with the short-form content already gone.

It's an iOS app using WKWebView with three layers of filtering per platform:

1. WKContentRuleList blocks URLs (/reels/, /shorts/) at the network level before they load. Also applies static CSS display:none. Compiled once at launch, zero JS cost.

2. CSS injected at document start for complex selectors (:has() combinators and such) that the content rule API can't express.

3. MutationObserver in JS catches content that Instagram/YouTube inject after initial load (both are SPAs that render feed items dynamically). WeakSet tracks processed nodes, rAF-debounced.

Selectors target aria-labels, data attributes, and href patterns instead of class names. Instagram rotates classes constantly. Each platform has its own manifest.json that defines what to block, so updating selectors when platforms change their DOM is a ~5 minute edit.

The part I didn't expect: removing algorithmic content isn't enough. Even "clean" Instagram is still designed to pull you back. So I added time limits, scheduled blocks (no Instagram after 11pm), and a friction gate. A small challenge appears before opening a platform. The friction gate alone cut my impulsive opens roughly in half.

Tradeoffs I'm aware of:

1. This is a cat-and-mouse game. Platforms change DOM, filters break, I update. The three-layer approach means if one layer breaks the others still catch most content, but it's ongoing work. 2. iOS only. WKWebView is the only way to do this on iOS without a VPN/proxy. Android would need a different approach entirely. 3. No free tier. $3.99/mo or $14.99/yr. I considered freemium but didn't want to gate filtering behind a paywall. It's all-or-nothing. 4. "Why not a Safari extension?" I tried this first. Extensions can't intercept SPA navigations reliably, can't do MutationObserver-based filtering, and the UX of opening Safari and navigating to instagram.com is much worse than tapping a platform icon.

No accounts, no analytics, no data leaves the device.

Comments URL: https://news.ycombinator.com/item?id=47313150

Points: 1

# Comments: 0

Article URL: https://www.youtube.com/watch?v=THAupOYwJTs

Comments URL: https://news.ycombinator.com/item?id=47313100

Points: 2

# Comments: 0

Article URL: https://www.eurekalert.org/news-releases/1118481

Comments URL: https://news.ycombinator.com/item?id=47313095

Points: 3

# Comments: 0

I built a small open-source tool to explore a software confidence gap that seems to be getting worse with AI-assisted coding.

Passing tests and coverage are often weaker signals than teams assume. Code can be exercised, tests can pass, and yet the actual acceptance criteria may still be only weakly protected.

More about the problem: https://www.dmytrohuz.com/p/i-built-ac-trace-to-question-the

ac-trace maps acceptance criteria to code and tests, then mutates the mapped code to check whether the tests really catch the breakage.

Current scope is intentionally narrow: Python + pytest. It’s an early experiment, and I’d really value feedback from people working on testing, QA/dev productivity, and AI-assisted development workflows.

Comments URL: https://news.ycombinator.com/item?id=47313085

Points: 2

# Comments: 0

I built a free browser-based tool for generating 3D text with multi-colour outlines, designed for 3D printers that support multiple filaments.

You type text, pick a font, add coloured outline layers, adjust thickness, and preview everything in real-time 3D. Then export as 3MF and each colour shows up as a separate named part that drops straight into Bambu Studio, no need for manual painting. You can also download the STLs.

Tech details for the curious:

- Vanilla JS client code - Three.js for 3D preview, opentype.js for font parsing - Geometry pipeline: opentype paths → Clipper-lib (boolean ops for outline rings via cumulative offsets + differences) → Tess2 (triangulation) → extruded meshes - Server-side export via Cloudflare workers (same geometry pipeline, runs headlessly to produce 3MF/STL) - Preview images generated by @cf-wasm/resvg on the edge - Every design is a shareable URL, all state lives in query params - Licensed CC BY 4.0.

Would love feedback especially from folks doing multi-colour printing.

Comments URL: https://news.ycombinator.com/item?id=47313073

Points: 1

# Comments: 1

Article URL: https://www.giacomoran.com/blog/act-smooth/

Comments URL: https://news.ycombinator.com/item?id=47313069

Points: 1

# Comments: 0

Article URL: https://startups.rip

Comments URL: https://news.ycombinator.com/item?id=47313042

Points: 2

# Comments: 0

Article URL: https://lookingforcontributors.web.app/

Comments URL: https://news.ycombinator.com/item?id=47313025

Points: 1

# Comments: 1

Article URL: https://mixpeek.com/blog/query-preprocessing-large-file-search

Comments URL: https://news.ycombinator.com/item?id=47313019

Points: 3

# Comments: 1

Article URL: https://www.philipzucker.com/thin1/

Comments URL: https://news.ycombinator.com/item?id=47313008

Points: 2

# Comments: 0