MIT asks arXiv to take down preprint of paper on AI and scientific discovery (Score: 153+ in 6 hours)
Link: https://readhacker.news/s/6uAF4
Comments: https://readhacker.news/c/6uAF4
Link: https://readhacker.news/s/6uAF4
Comments: https://readhacker.news/c/6uAF4
Thoughts on thinking (🔥 Score: 158+ in 2 hours)
Link: https://readhacker.news/s/6uBsd
Comments: https://readhacker.news/c/6uBsd
Link: https://readhacker.news/s/6uBsd
Comments: https://readhacker.news/c/6uBsd
Dustin Curtis on Svbtle
Thoughts on thinking
I have been stuck. Every time I sit down to write a blog post, code a feature, or start a project, I come to the same realization: in the context of AI, what I’m doing is a waste of time. It’s horrifying. The fun has been sucked out of the process...
I'm Peter Roberts, immigration attorney, who does work for YC and startups. AMA (Score: 153+ in 7 hours)
Link: https://readhacker.news/c/6uAEf
I'll be here for the next 5-6 hours. As usual, there are countless topics given the rapidly changing immigration landscape and I'll be guided by whatever you're concerned with. Please remember that I can't provide legal advice on specific cases because I won't have access to all the facts. Please stick to a factual discussion in your questions and I'll try to do the same in my answers.
Edit: I am taking a break now and will return later this afternoon/evening to respond to any comments and answer any questions. Thank you everyone for a great and engaged AMA so far.
Link: https://readhacker.news/c/6uAEf
I'll be here for the next 5-6 hours. As usual, there are countless topics given the rapidly changing immigration landscape and I'll be guided by whatever you're concerned with. Please remember that I can't provide legal advice on specific cases because I won't have access to all the facts. Please stick to a factual discussion in your questions and I'll try to do the same in my answers.
Edit: I am taking a break now and will return later this afternoon/evening to respond to any comments and answer any questions. Thank you everyone for a great and engaged AMA so far.
Show HN: KVSplit – Run 2-3x longer contexts on Apple Silicon (🔥 Score: 150+ in 2 hours)
Link: https://readhacker.news/s/6uBAK
Comments: https://readhacker.news/c/6uBAK
I discovered that in LLM inference, keys and values in the KV cache have very different quantization sensitivities. Keys need higher precision than values to maintain quality.
I patched llama.cpp to enable different bit-widths for keys vs. values on Apple Silicon. The results are surprising:
- K8V4 (8-bit keys, 4-bit values): 59% memory reduction with only 0.86% perplexity loss
- K4V8 (4-bit keys, 8-bit values): 59% memory reduction but 6.06% perplexity loss
- The configurations use the same number of bits, but K8V4 is 7× better for quality
This means you can run LLMs with 2-3× longer context on the same Mac. Memory usage scales with sequence length, so savings compound as context grows.
Implementation was straightforward:
1. Added --kvq-key and --kvq-val flags to llama.cpp
2. Applied existing quantization logic separately to K and V tensors
3. Validated with perplexity metrics across context lengths
4. Used Metal for acceleration (with -mlong-calls flag to avoid vectorization issues)
Benchmarked on an M4 MacBook Pro running TinyLlama with 8K context windows. Compatible with Metal/MPS and optimized for Apple Silicon.
GitHub: https://github.com/dipampaul17/KVSplit
Link: https://readhacker.news/s/6uBAK
Comments: https://readhacker.news/c/6uBAK
I discovered that in LLM inference, keys and values in the KV cache have very different quantization sensitivities. Keys need higher precision than values to maintain quality.
I patched llama.cpp to enable different bit-widths for keys vs. values on Apple Silicon. The results are surprising:
- K8V4 (8-bit keys, 4-bit values): 59% memory reduction with only 0.86% perplexity loss
- K4V8 (4-bit keys, 8-bit values): 59% memory reduction but 6.06% perplexity loss
- The configurations use the same number of bits, but K8V4 is 7× better for quality
This means you can run LLMs with 2-3× longer context on the same Mac. Memory usage scales with sequence length, so savings compound as context grows.
Implementation was straightforward:
1. Added --kvq-key and --kvq-val flags to llama.cpp
2. Applied existing quantization logic separately to K and V tensors
3. Validated with perplexity metrics across context lengths
4. Used Metal for acceleration (with -mlong-calls flag to avoid vectorization issues)
Benchmarked on an M4 MacBook Pro running TinyLlama with 8K context windows. Compatible with Metal/MPS and optimized for Apple Silicon.
GitHub: https://github.com/dipampaul17/KVSplit
GitHub
GitHub - dipampaul17/KVSplit: Run larger LLMs with longer contexts on Apple Silicon by using differentiated precision for KV cache…
Run larger LLMs with longer contexts on Apple Silicon by using differentiated precision for KV cache quantization. KVSplit enables 8-bit keys & 4-bit values, reducing memory by 59% with &am...
Moody’s strips U.S. of triple-A credit rating (🔥 Score: 153+ in 2 hours)
Link: https://readhacker.news/s/6uBNR
Comments: https://readhacker.news/c/6uBNR
Link: https://readhacker.news/s/6uBNR
Comments: https://readhacker.news/c/6uBNR
Ft
Moody’s strips US of top-notch triple-A credit rating
Agency warns of strains caused by rising government debt and a widening budget deficit
X X^t can be faster (Score: 150+ in 8 hours)
Link: https://readhacker.news/s/6uANa
Comments: https://readhacker.news/c/6uANa
Link: https://readhacker.news/s/6uANa
Comments: https://readhacker.news/c/6uANa
arXiv.org
$XX^{t}$ Can Be Faster
We present a new algorithm RXTX that computes product of matrix by its transpose $XX^{t}$. RXTX uses $5\%$ less multiplications and additions than State-of-the-Art and achieves accelerations even...
Java at 30: Interview with James Gosling (Score: 150+ in 11 hours)
Link: https://readhacker.news/s/6uAdJ
Comments: https://readhacker.news/c/6uAdJ
Link: https://readhacker.news/s/6uAdJ
Comments: https://readhacker.news/c/6uAdJ
The New Stack
Java at 30: The Genius Behind the Code That Changed Tech
From trash-diving teen to tech pioneer, James Gosling's pragmatic genius shaped three decades of Java and modern computing.
Getting AI to write good SQL (Score: 154+ in 4 hours)
Link: https://readhacker.news/s/6uBLa
Comments: https://readhacker.news/c/6uBLa
Link: https://readhacker.news/s/6uBLa
Comments: https://readhacker.news/c/6uBLa
Google Cloud Blog
Techniques for improving text-to-SQL | Google Cloud Blog
Learn about text-to-SQL techniques like context building and table retrieval, LLM-as-a-judge, and LLM prompting and post-processing.
ClojureScript 1.12.42 (Score: 151+ in 11 hours)
Link: https://readhacker.news/s/6uBDi
Comments: https://readhacker.news/c/6uBDi
Link: https://readhacker.news/s/6uBDi
Comments: https://readhacker.news/c/6uBDi
Pathfinding (Score: 150+ in 1 day)
Link: https://readhacker.news/s/6uwP7
Comments: https://readhacker.news/c/6uwP7
Link: https://readhacker.news/s/6uwP7
Comments: https://readhacker.news/c/6uwP7
itch.io
#9 - Pathfinding
Hello! I've recently been working on the pathfinding for NPCs in my game, which is something I've been looking forward to for a while now since it's a nice chunky problem to solve. I thought I'd write...
Náhuatl and Mayan Language Renaissance Occurring in Mexico (❄️ Score: 150+ in 3 days)
Link: https://readhacker.news/s/6uquf
Comments: https://readhacker.news/c/6uquf
Link: https://readhacker.news/s/6uquf
Comments: https://readhacker.news/c/6uquf
Yucatán Magazine
Náhuatl and Mayan Language Renaissance Occurring in Mexico
México is home to 68 officially recognized Indigenous languages and is experiencing a remarkable Mayan language renaissance
JavaScript's New Superpower: Explicit Resource Management (Score: 150+ in 8 hours)
Link: https://readhacker.news/s/6uCwD
Comments: https://readhacker.news/c/6uCwD
Link: https://readhacker.news/s/6uCwD
Comments: https://readhacker.news/c/6uCwD
v8.dev
JavaScript's New Superpower: Explicit Resource Management · V8
The Explicit Resource Management proposal empowers developers to explicitly manage the lifecycle of resources.
Evolution of Rust Compiler Errors (Score: 150+ in 1 day)
Link: https://readhacker.news/s/6uAh5
Comments: https://readhacker.news/c/6uAh5
Link: https://readhacker.news/s/6uAh5
Comments: https://readhacker.news/c/6uAh5
Kobzol’s blog
Evolution of Rust compiler errors
Blog about programming stuff.
Wow@Home – Network of Amateur Radio Telescopes (Score: 150+ in 12 hours)
Link: https://readhacker.news/s/6uCht
Comments: https://readhacker.news/c/6uCht
Link: https://readhacker.news/s/6uCht
Comments: https://readhacker.news/c/6uCht
phl.upr.edu
PHL @ UPR Arecibo - outreach
Introduction
Japan's IC cards are weird and wonderful (❄️ Score: 154+ in 2 days)
Link: https://readhacker.news/s/6uwBZ
Comments: https://readhacker.news/c/6uwBZ
Link: https://readhacker.news/s/6uwBZ
Comments: https://readhacker.news/c/6uwBZ
aruarian.dance
Japan's IC cards are weird and wonderful
Exploring what makes Japan's transit cards so unique compared to the West.
XTool – Cross-platform Xcode replacement (Score: 150+ in 12 hours)
Link: https://readhacker.news/s/6uChV
Comments: https://readhacker.news/c/6uChV
Link: https://readhacker.news/s/6uChV
Comments: https://readhacker.news/c/6uChV
GitHub
GitHub - xtool-org/xtool: Cross-platform Xcode replacement. Build and deploy iOS apps with SwiftPM on Linux, Windows, macOS.
Cross-platform Xcode replacement. Build and deploy iOS apps with SwiftPM on Linux, Windows, macOS. - xtool-org/xtool
Push Ifs Up and Fors Down (Score: 152+ in 6 hours)
Link: https://readhacker.news/s/6uCPf
Comments: https://readhacker.news/c/6uCPf
Link: https://readhacker.news/s/6uCPf
Comments: https://readhacker.news/c/6uCPf
matklad.github.io
Push Ifs Up And Fors Down
A short note on two related rules of thumb.
Rustls Server-Side Performance (❄️ Score: 150+ in 4 days)
Link: https://readhacker.news/s/6upUf
Comments: https://readhacker.news/c/6upUf
Link: https://readhacker.news/s/6upUf
Comments: https://readhacker.news/c/6upUf
Prossimo
Rustls Server-Side Performance
In past years, the Rustls project has been happy to receive substantial investments from the ISRG. One of our goals has been to improve performance without compromising on safety. We last posted about our performance improvements in October of 2024, and we're…
MCP: An in-depth introduction (❄️ Score: 150+ in 4 days)
Link: https://readhacker.news/s/6upNg
Comments: https://readhacker.news/c/6upNg
Link: https://readhacker.news/s/6upNg
Comments: https://readhacker.news/c/6upNg
Speakeasy
MCP: an in-depth introduction | Speakeasy
The API Platform. Generate, test, monitor, and document idiomatic SDKs directly from your OpenAPI spec.
Palette lighting tricks on the Nintendo 64 (Score: 151+ in 5 hours)
Link: https://readhacker.news/s/6uDgM
Comments: https://readhacker.news/c/6uDgM
Link: https://readhacker.news/s/6uDgM
Comments: https://readhacker.news/c/6uDgM