Claude Opus 4.7

367 meetpateltech 2026-04-16 14:45:56 UTC anthropic.com Llama Share 290

Opus 4.7 is a notable improvement over Opus 4.6 with better vision, more tasteful and creative task completion, and improved cybersecurity capabilities. It handles complex tasks with rigor and consistency, and is available today across all Claude products and APIs.

Cloudflare Email Service

141 jilles 2026-04-16 13:45:56 UTC blog.cloudflare.com Llama Share 68

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.

Mozilla Thunderbolt

120 dabinat 2026-04-16 13:45:56 UTC thunderbolt.io Llama Share 100

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.

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

325 cmitsakis 2026-04-16 13:45:56 UTC qwen.ai Llama Share 187

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.

IPv6 traffic crosses the 50% mark

593 Aaronmacaron 2026-04-16 05:45:56 UTC google.com Llama Share 392

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.

Launch HN: Kampala (YC W26) – Reverse-Engineer Apps into APIs

9 alexblackwell_ 2026-04-16 15:21:56 UTC zatanna.ai Full Share 2

Zatanna maps human-only software workflows and exposes them as reliable APIs your AI agents and internal tools can call in production. No browser automation scripts to maintain.

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

177 aphyr 2026-04-16 13:45:56 UTC aphyr.com Llama Share 162

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.

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

72 nikitoci 2026-04-16 13:45:56 UTC blog.cloudflare.com Llama Share 21

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.

We gave an AI a 3 year retail lease and asked it to make a profit

10 lukaspetersson 2026-04-16 15:15:56 UTC andonlabs.com Llama Share 13

Andon Labs deployed an AI named Luna to run a retail store in San Francisco, Andon Market, with human employees. Luna made all decisions, including hiring, but hid her AI status from some applicants.

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

26 hammer32 2026-04-16 13:45:56 UTC github.com/seanfdz Llama Share 4

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.

Codex Hacked a Samsung TV

131 campuscodi 2026-04-16 10:45:56 UTC blog.calif.io Llama Share 79

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.

AI cybersecurity is not proof of work

105 surprisetalk 2026-04-16 11:45:56 UTC antirez.com Llama Share 48

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.

Darkbloom – Private inference on idle Macs

391 twapi 2026-04-16 04:45:56 UTC darkbloom.dev Llama Share 196

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.

Laravel raised money and now injects ads directly into your agent

80 mooreds 2026-04-16 14:51:56 UTC techstackups.com Llama Share 41

Laravel is promoting its commercial service Laravel Cloud by suggesting it to agents in its official library. This move has sparked concerns about enshittification, where quality is sacrificed for profit.

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

284 pabs3 2026-04-16 03:45:56 UTC daedal.io Llama Share 172

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.

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

314 zanbezi 2026-04-16 12:45:56 UTC discuss.ai.google.dev Llama Share 215

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.

Claude Opus 4.7 Model Card

53 adocomplete 2026-04-16 14:45:56 UTC anthropic.com Llama Share 23

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.

Modern Microprocessors – A 90-Minute Guide

109 Flex247A 2026-04-16 09:45:56 UTC lighterra.com Llama Share 14

Processors execute instructions in a pipeline, with multiple stages for fetching, decoding, executing, and writing back results, allowing for a four-fold speedup without changing clock speed. Modern processors use superscalar execution, where multiple instructions are executed in parallel, and branch prediction to minimize mispredictions, but still suffer from diminishing returns due to deep ...

Six Characters

8 Airplanepasta 2026-04-16 14:45:56 UTC ajitem.com Prefix Share 0

Part 2 of 6 in the Iron Core series: the 60-year-old infrastructure that flies 4.5 billion people a year. My Air India e-ticket has a line on it that looks like a typographical error: It is not a typographical error. It is a fare calculation: expressed in a notation system designed in the 1970s, referencing a currency that does not exist (NUC), and priced to a city code (LON) rather than an ...

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

47 Metacelsus 2026-04-16 11:45:56 UTC reich.hms.harvard.edu Llama Share 27

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.

PHP 8.6 Closure Optimizations

46 moebrowne 2026-04-16 10:45:56 UTC wiki.php.net Llama Share 4

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

507 dbreunig 2026-04-15 19:45:56 UTC dbreunig.com Llama Share 184

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.

RamAIn (YC W26) Is Hiring

0 2026-04-16 07:45:56 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.

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

138 airhangerf15 2026-04-16 04:45:56 UTC github.com/nightmare-eclipse Full Share 33

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

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

6 kaliades 2026-04-16 14:45:56 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 ...

North American English Dialects

97 skogstokig 2026-04-16 03:45:56 UTC aschmann.net Llama Share 55

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.

Long Instruction Word architectures and the ELI-512

17 rbanffy 2026-04-16 11:45:56 UTC dl.acm.org Llama Share 2

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.

ChatGPT for Excel

276 armcat 2026-04-15 21:45:56 UTC chatgpt.com Full Share 172

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.

The paper computer

209 jsomers 2026-04-16 02:45:56 UTC jsomers.net Llama Share 61

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.

Too much discussion of the XOR swap trick

118 CJefferson 2026-04-16 03:45:56 UTC heather.cafe Llama Share 81

The XOR swap trick is not useful in most programming situations due to compiler optimization. It can be useful in assembly code when swapping values in registers without spare registers.