Qwen3.6-35B-A3B: Agentic Coding Power, Now Open to All

237 cmitsakis 2026-04-16 13:54:43 UTC qwen.ai Llama Share 112

Qwen3.6-35B-A3B is a sparse MoE model with 3B active parameters, delivering outstanding agentic coding performance and rivaling larger dense models. It's available as open weights on Hugging Face and ModelScope, and through Alibaba Cloud Model Studio API.

The Future of Everything Is Lies, I Guess: Where Do We Go from Here?

124 aphyr 2026-04-16 13:54:43 UTC aphyr.com Llama Share 73

The author is concerned about the impact of large language models on society, citing their potential to displace human workers and exacerbate existing problems. They advocate for slowing down the development of these models to give time for regulation and adaptation.

Claude Opus 4.7

122 AlphaWeaver 2026-04-16 14:30:43 UTC anthropic.com Llama Share 59

Opus 4.7 is a premium model for complex tasks, available on Claude for Pro, Max, Team, and Enterprise users, and on the Claude Platform natively, with pricing starting at $5 per million input tokens and $25 per million output tokens. It offers adaptive thinking, strong performance, and reliability for tasks like coding, agentic workflows, and enterprise tasks.

€54k spike in 13h from unrestricted Firebase browser key accessing Gemini APIs

291 zanbezi 2026-04-16 12:54:43 UTC discuss.ai.google.dev Llama Share 197

A €54,000+ Gemini API charge occurred after enabling Firebase AI Logic on an existing project, with usage not correlated to actual users and appearing automated. Google Cloud support classified the charges as valid usage, denying a billing adjustment despite the anomalous activity.

Cloudflare Email Service: now in public beta. Ready for your agents

59 jilles 2026-04-16 13:54:43 UTC blog.cloudflare.com Llama Share 25

Cloudflare Email Service is now in public beta, allowing applications and agents to send emails globally with low latency. It unlocks a full email client and agent pipeline for developers.

IPv6 traffic crosses the 50% mark

552 Aaronmacaron 2026-04-16 05:54:43 UTC google.com Llama Share 354

We measure IPv6 connectivity among Google users, showing deployment and reliability issues in various regions. Regions with high IPv6 deployment have fewer issues, while those with low deployment experience significant problems.

Cloudflare's AI Platform: an inference layer designed for agents

51 nikitoci 2026-04-16 13:54:43 UTC blog.cloudflare.com Llama Share 16

Cloudflare is launching AI Gateway, a unified inference layer that allows developers to access any AI model from any provider through one API. This simplifies AI development and management, enabling developers to build fast and reliable AI-powered applications.

Show HN: MacMind – A transformer neural network in HyperCard on a 1989 Macintosh

11 hammer32 2026-04-16 13:54:43 UTC github.com/seanfdz Llama Share 1

MacMind is a 1,216-parameter transformer that learns the bit-reversal permutation from random examples using HyperTalk, a scripting language. It demonstrates that backpropagation and attention are math, not magic, and can be run on a 68000 processor from 1987.

Darkbloom – Private inference on idle Macs

377 twapi 2026-04-16 04:54:43 UTC darkbloom.dev Llama Share 184

Darkbloom is a decentralized AI inference network connecting idle Apple Silicon machines to demand, reducing costs by up to 70% and giving operators 95% of revenue. It ensures end-to-end encryption and operator anonymity, eliminating data extraction risks.

AI cybersecurity is not proof of work

79 surprisetalk 2026-04-16 10:54:43 UTC antirez.com Llama Share 37

Proof of work analogy fails for AI security as bugs are not found through brute force but rather through model intelligence and understanding. Better models and faster access to them will win in AI security, not more computational power.

FSF trying to contact Google about spammer sending 10k+ mails from Gmail account

266 pabs3 2026-04-16 03:54:43 UTC daedal.io Llama Share 168

You're looking for a direct contact at Google's Gmail team to report a spammer issue. You've tried the abuse form but want to reach a human employee for a solution.

Modern Microprocessors – A 90-Minute Guide

94 Flex247A 2026-04-16 09:54:43 UTC lighterra.com Llama Share 11

Dr Patterson's work includes articles on processor design, compiler optimization, and software development, used by universities and startups worldwide. His research covers topics from instruction scheduling to executable optimization and open-source software impact.

Claude Opus 4.7 Model Card

22 adocomplete 2026-04-16 14:33:43 UTC anthropic.com Llama Share 5

Claude Opus 4.7 is a new large language model from Anthropic with improved capabilities in areas such as software engineering and knowledge work. It is largely well-aligned and shows low rates of reward hacking.

Ancient DNA reveals pervasive directional selection across West Eurasia [pdf]

42 Metacelsus 2026-04-16 11:54:43 UTC reich.hms.harvard.edu Llama Share 23

Scientists used ancient DNA to detect hundreds of genes under directional selection in West Eurasians over the past 10,000 years. These genes are associated with traits like body fat, schizophrenia, and cognitive performance, and may have been adaptive in ancient societies.

Codex Hacked a Samsung TV

108 campuscodi 2026-04-16 10:54:43 UTC blog.calif.io Llama Share 75

Researchers used AI to hack a Samsung TV, partnering with OpenAI, and successfully escalated privileges from a browser shell to root. The AI, Codex, used a physmap primitive to access physical memory and overwrite kernel credentials.

Show HN: Agent-cache – Multi-tier LLM/tool/session caching for Valkey and Redis

5 kaliades 2026-04-16 13:54:43 UTC news.ycombinator.com Prefix Share 0

Multi-tier exact-match cache for AI agents backed by Valkey or Redis. LLM responses, tool results, and session state behind one connection. Framework adapters for LangChain, LangGraph, and Vercel AI SDK. OpenTelemetry and Prometheus built in. No modules required - works on vanilla Valkey 7+ and Redis 6.2+. Existing options locked you into one tier (LangChain = LLM only, LangGraph = state ...

PHP 8.6 Closure Optimizations

36 moebrowne 2026-04-16 10:54:43 UTC wiki.php.net Llama Share 5

Two new optimizations for closures are proposed to improve performance by turning non-static closures into static ones and caching stateless closures. This may cause some backward compatibility breaks.

Cybersecurity looks like proof of work now

501 dbreunig 2026-04-15 19:54:43 UTC dbreunig.com Llama Share 182

Anthropic's Mythos LLM is highly capable at computer security tasks, completing a simulated corporate network attack in 3 out of 10 attempts. Security may be reduced to a simple equation of spending more tokens than attackers to harden a system.

Fly Drones from the Browser

7 mosfets 2026-04-16 13:54:43 UTC fpvsim.com Full Share 9

FPVSIM - Online Drone Simulator | UAV Simulator Online | Professional Race Lap Timer | RC Event Timing System

RedSun: System user access on Win 11/10 and Server with the April 2026 Update

131 airhangerf15 2026-04-16 04:54:43 UTC github.com/nightmare-eclipse Full Share 32

The Red Sun vulnerability repository. Contribute to Nightmare-Eclipse/RedSun development by creating an account on GitHub.

RamAIn (YC W26) Is Hiring

0 2026-04-16 07:54:43 UTC ycombinator.com Llama Share

RamAIn is a YC W26 company that builds AI agents for enterprise work, eliminating repetitive workflows. They're hiring a generalist for go-to-market, sales, and operations to work directly with the founders.

There's yet another study about how bad AI is for our brains

15 speckx 2026-04-16 14:32:43 UTC engadget.com Llama Share 4

Researchers found that AI assistance improves immediate performance but leads to cognitive dependence and burnout. Prolonged AI use erodes motivation and persistence, potentially hindering human innovation and creativity.

Long Instruction Word architectures and the ELI-512

14 rbanffy 2026-04-16 11:54:43 UTC dl.acm.org Llama Share 1

Researchers use trace scheduling to compile code for a parallel machine, expecting 10-30 times faster execution. This technique generates code for Very Long Instruction Word architectures, allowing many operations to execute in parallel within a single instruction stream.

Mozilla Thunderbolt

60 dabinat 2026-04-16 13:54:43 UTC thunderbolt.io Llama Share 53

Deploy on your infrastructure or let us help with deployment, keeping data in your control. We offer native apps, MCP integration, and open source options for customization.

Apple accelerates eco progress with highest-ever recycled materials

70 salkahfi 2026-04-16 12:54:43 UTC apple.com Llama Share 57

Apple achieved 30 percent recycled content across all products shipped in 2025 and removed plastic from packaging. The company also made progress toward its Apple 2030 goal of being carbon neutral by 2030.

ChatGPT for Excel

267 armcat 2026-04-15 21:54:43 UTC chatgpt.com Full Share 168

Ask ChatGPT to build full spreadsheets, get insights across tabs and formulas, and update workbooks in real time so projects move forward faster. Available globally for ChatGPT Business, Enterprise, Edu, Teachers, and K-12 users, and for ChatGPT Pro and Plus users outside the EU.

North American English Dialects

81 skogstokig 2026-04-16 03:54:43 UTC aschmann.net Llama Share 45

Rick Aschmann created a map of North American English dialects with clickable states and provinces, allowing users to listen to audio or video samples of local dialects. The map is based on data from the Atlas of North American English and has been updated with contributions from users, with over 900 samples listed.

The paper computer

196 jsomers 2026-04-16 02:54:43 UTC jsomers.net Llama Share 57

The user wants to combine physical and virtual worlds to reduce screen time and distractions, allowing for flexibility and convenience. They envision a future where technology helps people work and interact without constant screen interaction.

Moving a large-scale metrics pipeline from StatsD to OpenTelemetry / Prometheus

56 jmarbach 2026-04-16 05:54:43 UTC medium.com/airbnb-engineering Llama Share 13

Airbnb migrated its metrics pipeline from StatsD to OpenTelemetry and Prometheus, collecting all metrics first to focus on technical challenges. The new pipeline uses a dual-write strategy, OTLP for internal services and Prometheus for OSS workloads, with a centralized aggregation tier for cost savings and scalability.

Cal.com is going closed source

360 Benjamin_Dobell 2026-04-15 15:54:43 UTC cal.com Llama Share 282

Cal.com is moving to closed source due to increasing security risks from AI-powered vulnerability scanning. They're releasing a modified open source version, Cal.diy, under the MIT license for developers and hobbyists.