Study: Young users (9 to 18Y) ditch Google for AI, with unknown consequences

36 giuliomagnifico 2026-09-22 09:22:14 UTC norwegianscitechnews.com Llama Share 39

Researchers analyzed 173 studies on AI's impact on children and young people's learning, finding that little is known about its effects on basic skills in younger pupils. The study suggests that AI can either strengthen or weaken critical thinking and problem-solving abilities, depending on how it is used.

Can gzip be a language model?

150 networked 2026-09-22 06:22:14 UTC nathan.rs Llama Share 62

A compressor like gzip can be used for language modeling without neural networks by scoring continuations based on their compressed length. Gzip's DEFLATE algorithm uses a sliding window to find matches and compresses them, allowing it to predict the next bytes in a text prompt.

AMD's random number generator can't generate a 0?

35 BruceEel 2026-09-22 09:22:14 UTC board.flatassembler.net Llama Share 2

The author has discovered a bug in AMD's RDRAND and RDSEED instructions, which can't generate the number 0 at a given size.

MiMo v2.6

912 volf_ 2026-09-21 20:22:14 UTC mimo.xiaomi.com Full Share 405

Introducing the MiMo-V2.6 series: frontier intelligence, all the modalities, built in public.

Spymarks, Not Watermarks

451 possibilistic 2026-09-21 23:22:14 UTC brand.io Llama Share 113

Google's SynthID embeds secret signals into images, audio, and video to track users without consent. This technology is part of a growing trend of "spymarks" that can be used to monitor and control individuals' online activities.

I said no and Apple said yes

157 thatslast 2026-09-22 08:22:14 UTC dbushell.com Llama Share 94

The author discovered a feature in macOS 15.3 that phoned home with personal data and was unable to turn it off, highlighting the lack of consent in AI technology. The author expresses frustration with Apple's handling of their concerns and criticizes the AI industry for prioritizing profit over user privacy.

Transformers Explained Visually

419 aray07 2026-09-21 20:22:14 UTC poloclub.github.io Llama Share 65

Transformer is a neural network architecture that has revolutionized AI, used in text-generative models like GPT and Llama, and applied to various domains such as audio generation and image recognition. It operates on next-token prediction, using self-attention mechanism to process entire sequences and capture long-range dependencies.

Attention is all you have

818 zer0tonin 2026-09-21 15:22:14 UTC alicegg.tech Llama Share 245

The Tetris effect shows how focused attention shapes thoughts, but modern technology often takes control of what we see and hear. Intentional internet browsing is still available, requiring a slower pace to find quality content.

What Sun got wrong

593 chmaynard 2026-09-21 14:22:14 UTC bcantrill.dtrace.org Llama Share 343

The Oxide team is gathering in Emeryville for their annual meetup, wearing t-shirts that pay homage to past computer companies like Sun. The nostalgia for Sun has sparked a discussion about what the company got right and wrong, with lessons learned from both serving as inspiration and warning for Oxide's own approach.

MiMo-v2.6-Pro: Intelligence, Performance and Price Analysis

71 theanonymousone 2026-09-22 04:22:14 UTC artificialanalysis.ai Llama Share 19

MiMo-V2.6-Pro is a high-intelligence model with fast processing and multimodal capabilities, scoring 46 on the Artificial Analysis Intelligence Index, above average among open weight models of similar size. It costs $0.43 per 1M input tokens and $0.87 per 1M output tokens, making it moderately priced for its performance level.

I don't want to read what you didn't write

695 mooreds 2026-09-21 23:22:14 UTC blog.colinbreck.com Llama Share 282

The author expresses frustration with the increasing use of AI to generate writing, citing its inability to convey nuance and context. They argue that this approach can lead to unreadable documents and a loss of human connection in communication.

AI coding has made CI a bottleneck, so we reworked ours to keep up

246 julian_digital 2026-09-21 20:22:14 UTC linear.app Llama Share 270

The author optimized Linear's Continuous Integration (CI) process, reducing PR wait time and runner time by 34% and 52%, respectively. These changes improved the overall performance of CI, allowing for faster code shipping and reduced infrastructure costs.

Looking forward to Git 2.56 – and 3.0

131 chmaynard 2026-09-21 23:22:14 UTC lwn.net Llama Share 60

Git 2.56 has been released in release-candidate form, featuring over 700 non-merge commits and several improvements, including a new drop subcommand to remove offending commits from history. The next release is expected to be Git 3.0, which may include significant changes such as the switch to SHA-256 hash function by default and support for reftables.

Divide by depth for instant 3D

157 gabrieloc 2026-09-21 20:22:14 UTC gabrieloc.com Llama Share 29

The article explains that cameras in 3D graphics work by applying perspective projection, a simple math-based technique to project 3D coordinates onto a 2D screen. This involves dividing x and y coordinates by z, which represents depth, resulting in normalized device coordinates (NDC).

NASA’s Mars Sample Return mission is dead

389 Muhammad523 2026-09-21 19:22:14 UTC science.org Full Share 326

Engineering Memory: On learning to memorize first 100 digits of pi (2024)

17 fuzzythinker 2026-09-22 06:22:14 UTC gregorygundersen.com Llama Share 10

The author memorized 100 digits of pi using a person-action-object (PAO) system, which associates each digit with a unique image. This process converted memorization into engineering, allowing for efficient recall and minimal errors.

A font that reads what you wrote

7 RohanAdwankar 2026-09-22 08:22:14 UTC rohanadwankar.github.io LocalLLM Share 7

semfont automatically adjusts typography based on semantic scores like valence and salience. It uses contextual rules and a second pass to improve readability in streaming text.

Claude Status – Elevated errors for multiple models

114 corvad 2026-09-22 01:22:14 UTC status.claude.com Full Share 83

Claude's Status Page - Elevated errors for multiple models.

World Wide Words

17 Petiver 2026-09-22 05:22:14 UTC worldwidewords.org Llama Share 1

World Wide Words has documented English language changes over 30 years, featuring articles on word histories and idioms. The website relies on donations to cover costs and offers ad-free reading pleasure.

The Advisory Group on Mathematics and Artificial Intelligence

141 digital55 2026-09-21 19:22:14 UTC terrytao.wordpress.com Llama Share 66

The Advisory Group on Mathematics and Artificial Intelligence aims to advise AI companies on their interactions with mathematical research, promoting transparency and accountability. The group seeks input from the mathematical community to inform recommendations for OpenAI's release of significant results in mathematics produced by its internal model.

What It's Like to Work in One of America's Data Centers

11 JumpCrisscross 2026-09-22 06:22:14 UTC wsj.com Full Share 2

Please enable JS and disable any ad blocker

PDF Forgeries Are Surprisingly Rare (2022)

42 1317 2026-09-22 03:22:14 UTC gwern.net Llama Share 35

Some people create fake PDFs to spread misinformation or scam others, but editing existing PDFs is difficult and rare. This makes it easier for malicious documents to circulate online without being detected, as they can be easily created and shared without leaving behind digital evidence.

Used ThinkPad Buyer's Guide (2019)

23 Mr_Minderbinder 2026-09-22 04:22:14 UTC bobble.tech Llama Share 10

This is a comprehensive Buyers Guide for ThinkPads, covering various models and their features. The guide provides estimates of prices, screen quality, and compatibility with Windows 10 and Linux.

Socrates vs. the Written Word (2011)

46 spectraldrift 2026-09-22 00:22:14 UTC wondermark.com Llama Share 22

The author discusses how humans have a universal dislike for change and progress, often feeling disconnected from those who adopt new technologies or ways of life. This sentiment is reflected throughout history, with examples including ancient Greek philosopher Socrates' critique of written language in Plato's Phaedrus dialogue.

Python Workers are now generally available

229 torutofu 2026-09-21 14:22:14 UTC blog.cloudflare.com Llama Share 38

Cloudflare Workers now supports Python as a first-class language, allowing developers to run Python applications in the Cloudflare Workers runtime. This enables seamless integration with existing Python packages and frameworks.

HERMES radio enables voice and data communication over vast distances

139 SamuraiLion 2026-09-21 16:22:14 UTC spectrum.ieee.org Llama Share 60

Peter Bloom's nonprofit Rhizomatica builds shortwave radios with better data rates and encryption, connecting remote communities worldwide. Their open-sourced HERMES system uses the ionosphere for long-distance communication and can send various files over high-frequency radio waves.

How do traffic signals work? (2019)

86 at1as 2026-09-21 16:22:14 UTC practical.engineering Llama Share 63

Traffic management in dense urban areas is a complex problem with conflicting goals and challenges, requiring careful analysis of space, cost, approach speed, cycle time, sight distance, types and volumes of traffic. Traffic signals are a common solution to control the right-of-way at intersections.

Apple Copland D11E4 Booting in the Browser

141 luu 2026-09-21 18:22:14 UTC pagetable.com Llama Share 40

Apple's Copland OS was cancelled and replaced by Mac OS X, but its last build is now available in emulation. The 11 patches to unlock it are on a forked branch for wasm compatibility.

Frontier AI on Your Own Hardware

161 pretext 2026-09-21 19:22:14 UTC timdettmers.com Llama Share 81

The author argues that the future of research is not about having the most resources, but rather about using limited resources to create something valuable. Academia will experience a renaissance due to this shift in approach.

Grok 4.7

576 meetpateltech 2026-09-21 16:22:14 UTC x.ai Llama Share 488

Grok 4.7 is a highly capable model for coding and knowledge work, offering improved performance on difficult tasks and enhanced safeguards. It is competitively priced at $2 per million input tokens and available through various platforms.