Advertisement · 728 × 90

Posts by Hacker News Top Stories

Preview
Antenna — subscribe once, read by email or over MCP Local-first feed reader. Email + MCP outputs, backed by the same SQLite index. For a world where your inbox and your agent are both first-class readers.

Show HN: Antenna – RSS reader with a built-in MCP server | Discussion

13 minutes ago 0 0 0 0
Preview
Tim Cook’s Impeccable Timing Tim Cook had an extraordinary run — and impeccable timing, both in terms of when he became CEO, and when he is stepping down.

Tim Cook's Impeccable Timing | Discussion

33 minutes ago 0 0 0 0
Apple keeps challenging its interoperability obligations under the DMA - FSFE A new FSFE report exposes how 56 interoperability requests under the Digital Markets Act have produced no concrete solutions by Apple, and how their declin...

Apple ignores DMA interoperability requests and contradicts own documentation | Discussion

1 hour ago 0 0 0 0
Preview
Laws of Software Engineering A collection of principles and patterns that shape software systems, teams, and decisions.

Laws of Software Engineering | Discussion

1 hour ago 0 0 0 0
@codemix/graph - codemix Open-source TypeScript property graph database from codemix.

A type-safe, realtime collaborative Graph Database in a CRDT | Discussion

2 hours ago 0 0 0 0
Preview
GitHub - i12bp8/TagTinker: Flipper Zero app for ESL research using IR. All based on https://www.furrtek.org/?a=esl Flipper Zero app for ESL research using IR. All based on https://www.furrtek.org/?a=esl - i12bp8/TagTinker

Edit store price tags using Flipper Zero | Discussion

2 hours ago 1 0 0 0
Preview
MNT Reform

MNT Reform is an open hardware laptop, designed and assembled in Germany | Discussion

4 hours ago 1 0 0 0
Types and Neural Networks

Types and Neural Networks | Discussion

5 hours ago 0 0 0 0
Preview
A Mad Undertaking

A mad undertaking: An undefinitive guide to the Aadam Jacobs collection | Discussion

5 hours ago 0 0 0 0

Salmon exposed to cocaine and its main byproduct roam more widely | Discussion

5 hours ago 1 0 0 0
Advertisement
Less human AI agents, please. Nial – Knowledge work and artificial intelligence.

Less human AI agents, please | Discussion

6 hours ago 0 0 0 0
Preview
R.I.P. Louis Zocchi, The Godfather of Dice Founder of Zocchi Distribution, Gamescience, and Creator of Zocchihedron

Louis Zocchi, inventor of the d100, has died | Discussion

6 hours ago 1 0 0 0
Preview
Prediction markets are breaking the news and becoming their own beat "I wake up every day and I'm like, <i>this is the world. I'm still in the fever dream.</i>”

Prediction markets are breaking the news and becoming their own beat | Discussion

7 hours ago 0 0 0 0
Preview
The air is full of DNA — here’s what scientists are using it for Airborne genetic material can be used to paint a picture of ecosystem health, watch for invasive species and even identify humans.

Air Is Full of DNA | Discussion

7 hours ago 2 0 0 0
Preview
The Beauty of Bonsai Styles The way in which a tree grows is often determined by their environment, and as a bonsai artist, I look to honor that environment, and the way in which that tree would grow in nature, when determining how to shape and style a bonsai.

The Beauty of Bonsai Styles | Discussion

7 hours ago 1 1 0 0
Preview
Anthropic - OpenClaw

Anthropic says OpenClaw-style Claude CLI usage is allowed again | Discussion

8 hours ago 2 1 0 0

A Roblox cheat and one AI tool brought down Vercel's platform | Discussion

8 hours ago 0 1 0 0
Preview
Year of the IPv6 overlay network - Defined Networking Nebula v1.10.0 is here with IPv6 overlay support, multiple Nebula IPs per host, a new certificate format, and more.

Year of the IPv6 Overlay Network | Discussion

9 hours ago 0 0 0 0
Preview
How is subsea cable repaired? It’s Fun Fact Friday and today we’re going to talk about repairing submarine cables. Submarine data cables are located under the ocean, and connect people around the world with data and information, while submarine electrical cables transfer electricity, such as those used on wind farms. Undersea cables for data transmission have been around since the mid-1800s, and you can learn more about the history of data cables and how they send information all around the globe in our article How do unders

How a subsea cable is repaired | Discussion

9 hours ago 0 0 0 0
Advertisement
Sequential KV Cache Compression via Probabilistic Language Tries: Beyond the Per-Vector Shannon Limit Recent work on KV cache quantization, culminating in TurboQuant, has approached the Shannon entropy limit for per-vector compression of transformer key-value caches. We observe that this limit applies to a strictly weaker problem than the one that actually matters: compressing the KV cache as a sequence. The tokens stored in a KV cache are not arbitrary floating-point data -- they are samples from the exact formal language the model was trained on, and the model is by construction a near-optimal predictor of that language. We introduce sequential KV compression, a two-layer architecture that exploits this structure. The first layer, probabilistic prefix deduplication, identifies semantically equivalent shared prefixes across sessions using the trie metric d_T(s, s') = -log_2 P_M(s ^ s') from Probabilistic Language Tries (PLTs). The second layer, predictive delta coding, stores only the residual of each new KV vector from the model's own prediction of it, achieving a per-token entropy bound of H(KV_{i+1} | KV_{<=i}) <= H(token_{i+1} | token_{<=i}). We prove that at typical language model perplexity -- approximately 10-20 for fluent English text -- this bound is 3.3-4.3 bits on average per token position, compared to TurboQuant's 3 bits per vector component (with typical attention heads having 64-128 components). The theoretical compression ratio over TurboQuant is approximately 914,000x at the Shannon limit. Even at 1000x above the entropy floor -- a deliberately pessimistic worst-case overhead, two orders of magnitude above the 2-5x typical of practical source coders -- the ratio remains approximately 914x over TurboQuant, with compression improving rather than degrading as context length grows. The two layers are orthogonal and compose with existing per-vector quantization methods including TurboQuant.

KV Cache Compression 900000x Beyond TurboQuant and Per-Vector Shannon Limit | Discussion

10 hours ago 0 0 0 0

Japan's Cherry Blossom Database, 1,200 Years Old, Has a New Keeper | Discussion

11 hours ago 0 0 0 0
How To Make a Fast Dynamic Language Interpreter

How to Make a Fast Dynamic Language Interpreter | Discussion

12 hours ago 0 0 0 0
Modern rendering culling techniques | krupitskas 🌦️ 🌲 The best work is the work that never gets executed

Modern Rendering Culling Techniques | Discussion

12 hours ago 0 1 0 0
Preview
PrismML — Introducing Ternary Bonsai: Top Intelligence at 1.58 Bits

Ternary Bonsai: Top Intelligence at 1.58 Bits | Discussion

13 hours ago 0 0 0 0
Preview
Even 'uncensored' models can't say what they want A safety-filtered pretrain can duck a charged word without refusing. It puts a fraction of the probability an open-data pretrain puts there. We call that...

Even 'uncensored' models can't say what they want | Discussion

13 hours ago 0 0 0 0
Zero-Copy Pages in Rust: Or How I Learned To Stop Worrying And Love Lifetimes You can find the source code for the project here

Zero-Copy Pages in Rust: Or How I Learned to Stop Worrying and Love Lifetimes | Discussion

14 hours ago 0 0 0 0
CCS - Ideas

Monero Community Crowdfunding System | Discussion

14 hours ago 0 0 0 0
Advertisement
Preview
EXCLUSIVE: Leaked Deck Reveals StackAdapt’s Playbook for ChatGPT Ads The DSP dangles ChatGPT ad placement driven by prompt relevance and CPMs range from $15 to $60, with a $50,000 minimum spend required for the pilot.

OpenAI ad partner now selling ChatGPT ad placements based on "prompt relevance" | Discussion

14 hours ago 0 0 0 0
Preview
Jujutsu megamerges for fun and profit A practical guide to creating and maintaining powerful megamerge workflows in Jujutsu for faster, conflict-free development.

Jujutsu megamerges for fun and profit | Discussion

15 hours ago 0 0 0 0
Preview
GitHub - gizmo64k/soulplayer-c64: A real 25k-parameter transformer running on a Commodore 64! A real 25k-parameter transformer running on a Commodore 64! - gizmo64k/soulplayer-c64

Soul Player C64 – A real transformer running on a 1 MHz Commodore 64 | Discussion

15 hours ago 1 1 0 0