Look Ma, No Bubbles: Designing a Low-Latency Megakernel for Llama-1B (Score: 151+ in 11 hours)
Link: https://readhacker.news/s/6vcet
Comments: https://readhacker.news/c/6vcet
Link: https://readhacker.news/s/6vcet
Comments: https://readhacker.news/c/6vcet
hazyresearch.stanford.edu
Look Ma, No Bubbles! Designing a Low-Latency Megakernel for Llama-1B
AI: Accelerated Incompetence (🔥 Score: 156+ in 2 hours)
Link: https://readhacker.news/s/6vdbh
Comments: https://readhacker.news/c/6vdbh
Link: https://readhacker.news/s/6vdbh
Comments: https://readhacker.news/c/6vdbh
Doug Slater
AI: Accelerated Incompetence
The Who Cares Era (🔥 Score: 150+ in 1 hour)
Link: https://readhacker.news/s/6vduW
Comments: https://readhacker.news/c/6vduW
Link: https://readhacker.news/s/6vduW
Comments: https://readhacker.news/c/6vduW
Dansinker
The Who Cares Era | dansinker.com
As a developer, my most important tools are a pen and a notebook (Score: 156+ in 9 hours)
Link: https://readhacker.news/s/6vcHU
Comments: https://readhacker.news/c/6vcHU
Link: https://readhacker.news/s/6vcHU
Comments: https://readhacker.news/c/6vcHU
Juha-Matti Santala - Community Builder. Dreamer. Adventurer.
As a developer, my most important tools are a pen and a notebook
Writing code is a part of being a software developer but often more important question is to figure what code to write and how and that I like to do away from computer with my notebook.
Mullvad Leta (🔥 Score: 160+ in 1 hour)
Link: https://readhacker.news/s/6vdLH
Comments: https://readhacker.news/c/6vdLH
Link: https://readhacker.news/s/6vdLH
Comments: https://readhacker.news/c/6vdLH
Mullvad Leta
Leta, the privacy-focused search engine, by Mullvad.
xAI to pay telegram $300M to integrate Grok into the chat app (🔥 Score: 154+ in 3 hours)
Link: https://readhacker.news/s/6vdT8
Comments: https://readhacker.news/c/6vdT8
Link: https://readhacker.news/s/6vdT8
Comments: https://readhacker.news/c/6vdT8
TechCrunch
xAI to pay Telegram $300M to integrate Grok into the chat app | TechCrunch
Telegram CEO Pavel Durov on Wednesday said Elon Musk's AI company, xAI, is investing $300 million worth of cash and equity in the chat app.
Show HN: I rewrote my Mac Electron app in Rust (🔥 Score: 158+ in 2 hours)
Link: https://readhacker.news/s/6vefR
Comments: https://readhacker.news/c/6vefR
A year ago, my co-founder launched Desktop Docs here on HN. It's a Mac app we built with Electron that uses CLIP embeddings to search photos and videos locally with natural language. We got positive feedback from HN and our first paying customers, but the app was almost 1GB and clunky to use.
TLDR; rebuilding in Rust was the right move.
So we rewrote the app with Rust and Tauri and here are the results:
- App size is 83% smaller: 1GB → 172MB
- DMG Installer is 70% smaller: 232MB → 69.5MB
- Indexing files is faster: A 38-minute video now indexes in ~3 minutes instead of 10-14 minutes
- Overall more stability (old app used to randomly crash)
The original version worked, but it didn't perform well when you tried indexing thousands of images or large videos. We lost a lot of time struggling to optimize Electron’s main-renderer process communication and ended up with a complex worker system to process large batches of media files.
For months we wrestled with indecision about continuing to optimize the Electron app vs. starting a full rebuild in Swift or Rust. The main thing holding us back was that we hadn’t coded in Swift in almost 10 years and we didn’t know Rust very well.
What finally broke us was when users complained the app crashed their video calls just running in background. I guess that’s what happens when you ship an app with Chromium that takes up 200mb before any application code.
Today the app still uses CLIP for embeddings and Redis for vector storage and search, except Rust now handles the image and video processing pipeline and all the file I/O to let users browse their entire machine, not just indexed files.
For the UI, we decided to rebuild it from scratch instead of porting over the old UI. This turned out well because it resulted in a cleaner, simpler UI after living with the complexity of the old version.
The trickiest part of the migration was learning Rust. LLMs definitely help, but the Rust/Tauri community just isn’t as mature compared to Electron. Bundling Redis into the app was a permissioning nightmare, but I think our solution with Rust handles this better than what we had with Electron.
All in, the rebuild took about two months and still needs some more work to be at total parity with its Electron version, but the core functionality of indexing and searching files is way more performant than before and that made it worth the time. Sometimes you gotta throw away working code to build the right thing.
AMA about Rust/Tauri migration, Redis bundling nightmares, how CLIP embeddings work for local semantic search, or why Electron isn't always the answer.
Link: https://readhacker.news/s/6vefR
Comments: https://readhacker.news/c/6vefR
A year ago, my co-founder launched Desktop Docs here on HN. It's a Mac app we built with Electron that uses CLIP embeddings to search photos and videos locally with natural language. We got positive feedback from HN and our first paying customers, but the app was almost 1GB and clunky to use.
TLDR; rebuilding in Rust was the right move.
So we rewrote the app with Rust and Tauri and here are the results:
- App size is 83% smaller: 1GB → 172MB
- DMG Installer is 70% smaller: 232MB → 69.5MB
- Indexing files is faster: A 38-minute video now indexes in ~3 minutes instead of 10-14 minutes
- Overall more stability (old app used to randomly crash)
The original version worked, but it didn't perform well when you tried indexing thousands of images or large videos. We lost a lot of time struggling to optimize Electron’s main-renderer process communication and ended up with a complex worker system to process large batches of media files.
For months we wrestled with indecision about continuing to optimize the Electron app vs. starting a full rebuild in Swift or Rust. The main thing holding us back was that we hadn’t coded in Swift in almost 10 years and we didn’t know Rust very well.
What finally broke us was when users complained the app crashed their video calls just running in background. I guess that’s what happens when you ship an app with Chromium that takes up 200mb before any application code.
Today the app still uses CLIP for embeddings and Redis for vector storage and search, except Rust now handles the image and video processing pipeline and all the file I/O to let users browse their entire machine, not just indexed files.
For the UI, we decided to rebuild it from scratch instead of porting over the old UI. This turned out well because it resulted in a cleaner, simpler UI after living with the complexity of the old version.
The trickiest part of the migration was learning Rust. LLMs definitely help, but the Rust/Tauri community just isn’t as mature compared to Electron. Bundling Redis into the app was a permissioning nightmare, but I think our solution with Rust handles this better than what we had with Electron.
All in, the rebuild took about two months and still needs some more work to be at total parity with its Electron version, but the core functionality of indexing and searching files is way more performant than before and that made it worth the time. Sometimes you gotta throw away working code to build the right thing.
AMA about Rust/Tauri migration, Redis bundling nightmares, how CLIP embeddings work for local semantic search, or why Electron isn't always the answer.
Desktop Docs
Desktop Docs - Advanced Image Search for Mac
Transform how you find and organize media with Desktop Docs' advanced search. Supporting HEIC, JPG, PNG, GIF, MP4, MOV, and more formats with AI-powered visual search and intelligent organization.
Getting a Cease and Desist from Waffle House (🔥 Score: 153+ in 3 hours)
Link: https://readhacker.news/s/6ve2Y
Comments: https://readhacker.news/c/6ve2Y
Link: https://readhacker.news/s/6ve2Y
Comments: https://readhacker.news/c/6ve2Y
Jack LaFond
Getting a Cease and Desist from Waffle House | Jack's Blog
Bit of a sticky situation.
Deepseek R1-0528 (🔥 Score: 150+ in 2 hours)
Link: https://readhacker.news/s/6vew4
Comments: https://readhacker.news/c/6vew4
Link: https://readhacker.news/s/6vew4
Comments: https://readhacker.news/c/6vew4
OpenTPU: Open-Source Reimplementation of Google Tensor Processing Unit (TPU) (Score: 150+ in 21 hours)
Link: https://readhacker.news/s/6vcaw
Comments: https://readhacker.news/c/6vcaw
Link: https://readhacker.news/s/6vcaw
Comments: https://readhacker.news/c/6vcaw
GitHub
GitHub - UCSBarchlab/OpenTPU: A open source reimplementation of Google's Tensor Processing Unit (TPU).
A open source reimplementation of Google's Tensor Processing Unit (TPU). - UCSBarchlab/OpenTPU
Compiler Explorer and the promise of URLs that last forever (Score: 155+ in 4 hours)
Link: https://readhacker.news/s/6veau
Comments: https://readhacker.news/c/6veau
Link: https://readhacker.news/s/6veau
Comments: https://readhacker.news/c/6veau
xania.org
Compiler Explorer and the Promise of URLs That Last Forever — Matt Godbolt’s blog
How we're preserving 12,000 legacy links as Google's URL shortener rides into the sunset
Dependency injection frameworks add confusion (❄️ Score: 150+ in 3 days)
Link: https://readhacker.news/s/6v48d
Comments: https://readhacker.news/c/6v48d
Link: https://readhacker.news/s/6v48d
Comments: https://readhacker.news/c/6v48d
Redowan's Reflections
You probably don't need a DI framework
When working with Go in an industrial context, I feel like dependency injection (DI) often
gets a bad rep because of DI frameworks. But DI as a technique is quite useful. It just
tends to get explained with too many OO jargons and triggers PTSD among those…
gets a bad rep because of DI frameworks. But DI as a technique is quite useful. It just
tends to get explained with too many OO jargons and triggers PTSD among those…
A thought on JavaScript "proof of work" anti-scraper systems (❄️ Score: 150+ in 2 days)
Link: https://readhacker.news/s/6v6CP
Comments: https://readhacker.news/c/6v6CP
Link: https://readhacker.news/s/6v6CP
Comments: https://readhacker.news/c/6v6CP
Compiling a neural net to C for a speedup (Score: 150+ in 5 hours)
Link: https://readhacker.news/s/6ven7
Comments: https://readhacker.news/c/6ven7
Link: https://readhacker.news/s/6ven7
Comments: https://readhacker.news/c/6ven7
slightknack.dev
Compiling a Neural Net to C for a 1,744× speedup — Isaac Clayton
A cozy little corner of the web.
Japan Post launches 'digital address' system (Score: 151+ in 7 hours)
Link: https://readhacker.news/s/6vebv
Comments: https://readhacker.news/c/6vebv
Link: https://readhacker.news/s/6vebv
Comments: https://readhacker.news/c/6vebv
The Japan Times
Japan Post launches 'digital address' system
Under the system, users can input seven-digit codes on online shopping websites, and their addresses will automatically appear on the sites.
De-anonymization attacks against the privacy coin XMR (Score: 150+ in 10 hours)
Link: https://readhacker.news/s/6vdFW
Comments: https://readhacker.news/c/6vdFW
Link: https://readhacker.news/s/6vdFW
Comments: https://readhacker.news/c/6vdFW
Monero.Forex - Ranking the Top Instant Crypto Exchanges
Is Monero Totally Private? A Comprehensive Analysis of De-Anonymization Attacks Against The Privacy Coin - Monero.Forex - Ranking…
Monero (XMR), a cryptocurrency renowned for its privacy-centric design, has drawn the attention of governments, cybersecurity experts, and analytics firms seeking to deanonymize its transactions.
US Trade Court Finds Trump Tariffs Illegal (🔥 Score: 153+ in 1 hour)
Link: https://readhacker.news/s/6vfs6
Comments: https://readhacker.news/c/6vfs6
Link: https://readhacker.news/s/6vfs6
Comments: https://readhacker.news/c/6vfs6
Bloomberg.com
Trump’s Global Tariffs Deemed Illegal, Blocked by Trade Court
The vast majority of President Donald Trump’s global tariffs were deemed illegal and blocked by the US trade court, dealing a major blow to a pillar of his economic agenda.
Long live American Science and Surplus (which needs your help) (Score: 154+ in 5 hours)
Link: https://readhacker.news/s/6vf4d
Comments: https://readhacker.news/c/6vf4d
Link: https://readhacker.news/s/6vf4d
Comments: https://readhacker.news/c/6vf4d
Milwaukee Record
Long live American Science & Surplus (which needs your help)
American Science & Surplus is built for discovery, tailored to the shopper who wanders in having absolutely no idea what they’ll find until they find it.
LiveStore: State management based on reactive SQLite and built-in sync engine (Score: 150+ in 1 day)
Link: https://readhacker.news/s/6vaeE
Comments: https://readhacker.news/c/6vaeE
Link: https://readhacker.news/s/6vaeE
Comments: https://readhacker.news/c/6vaeE
livestore.dev
LiveStore: Local-first data layer for high-performance apps
LiveStore is a state management framework based on SQLite and event-sourcing. It’s designed for demanding applications and based on years of research.