Motorola GrapheneOS devices will be bootloader unlockable/relockable

751 pabs3 2026-03-04 01:13:31 UTC grapheneos.social Full Share 236

@[email protected] It will fully support using other operating systems including users making their own builds of GrapheneOS. It's part of our hardware requirements. We'll likely be able to make hardened builds of firmware and drivers which can be released in an official way for easy builds without needing to extract anything from the GrapheneOS or Motorola OS factory images.

RFC 9849. TLS Encrypted Client Hello

87 P_qRs 2026-03-04 08:13:31 UTC rfc-editor.org Llama Share 37

This document describes a new Transport Layer Security (TLS) extension called Encrypted Client Hello (ECH) that encrypts the ClientHello message to protect sensitive information. ECH uses Hybrid Public Key Encryption (HPKE) and is supported in TLS 1.3, DTLS 1.3, and newer versions of the TLS and DTLS protocols.

Agentic Engineering Patterns

170 r4um 2026-03-04 05:13:31 UTC simonwillison.net Full Share 70

Patterns for getting the best results out of coding agents like Claude Code and OpenAI Codex. See my introduction for more on this project.

Better JIT for Postgres

68 vladich 2026-03-04 07:13:31 UTC github.com/vladich Llama Share 17

pg_jitter is a PostgreSQL JIT compilation provider that adds three alternative JIT backends - sljit, AsmJit, and MIR - delivering faster compilation and competitive query execution. It provides native code generation with microsecond-level compilation times.

RE#: how we built the fastest regex engine in F#

16 exceptione 2026-03-04 09:13:31 UTC iev.ee Llama Share 3

A new regex engine called RE# was developed in F# that outperforms other industrial regex engines and supports intersection, complement, and context-aware lookarounds with linear-time guarantees. The engine uses a derivative-based approach and minterm compression to achieve high performance and supports leftmost-longest matching with bidirectional scanning.

A CPU that runs entirely on GPU

112 cypres 2026-03-04 05:13:31 UTC github.com/robertcprice Llama Share 46

The NeuralCPU is a GPU-based CPU that uses trained neural networks for all arithmetic operations, achieving speeds of 4,975 IPS and inverting the conventional CPU's MUL vs ADD performance. It supports two modes: Neural Mode, where every operation is a forward pass through a trained model, and Fast Mode, where the ALU uses native torch arithmetic for maximum throughput.

MacBook Pro with M5 Pro and M5 Max

791 scrlk 2026-03-03 14:13:31 UTC apple.com Llama Share 837

Apple announced the new 14- and 16-inch MacBook Pro with M5 Pro and M5 Max chips, offering up to 4x faster AI performance and up to 24 hours of battery life. The new MacBook Pro features a Liquid Retina XDR display, Wi-Fi 7, and Bluetooth 6, and starts at 1TB of storage for M5 Pro and 2TB for M5 Max.

Graphics Programming Resources

105 abetusk 2026-03-04 03:13:31 UTC develop--gpvm-website.netlify.app Llama Share 12

This page offers a collection of resources for learning graphics programming, including tutorials, books, and articles on topics like OpenGL, Vulkan, ray tracing, and rendering techniques.

Show HN: I made a zero-copy coroutine tracer to find my scheduler's lost wakeups

11 lixiasky 2026-03-04 09:13:31 UTC github.com/lixiasky-back Llama Share 0

The user created coroTracer to track down logical deadlocks and broken state machines in M:N coroutine schedulers, which traditional tools cannot catch. coroTracer is an out-of-process tracer that uses shared memory and a separate engine to track state changes and detect issues like lost wakeups and coroutine leaks.

Claude's Cycles [pdf]

649 fs123 2026-03-03 11:13:31 UTC www-cs-faculty.stanford.edu Llama Share 269

Claude Opus 4.6 solved an open problem in automatic deduction and creative problem solving by finding a generalization for odd values of m. The solution was discovered through a series of explorations and involved a novel approach called fiber decomposition.

On the Design of Programming Languages (1974) [pdf]

47 jruohonen 2026-03-04 05:13:31 UTC web.cs.ucdavis.edu Full Share 7

TikTok will not introduce end-to-end encryption, saying it makes users less safe

263 1659447091 2026-03-04 02:13:31 UTC bbc.com Llama Share 216

TikTok won't introduce end-to-end encryption for DMs citing safety concerns. It prioritizes proactive safety over user privacy, arguing it helps prevent harm to young users.

Voxile: A ray-traced game made in its own engine and programming language

205 spacemarine1 2026-03-03 21:13:31 UTC elbowgreasegames.substack.com Llama Share 57

Wouter von Oortmersen's VoxRay Games released The Builder's Update for Voxile, a voxel-based survival game with ray tracing and co-op multiplayer. Voxile's custom engine, Lobster, enables fast prototyping and record-speed compile times.

Show HN: Stacked Game of Life

19 vnglst 2026-03-04 08:13:31 UTC stacked-game-of-life.koenvangilst.nl Full Share 7

Simulation

My spicy take on vibe coding for PMs

116 dmckinno 2026-03-04 00:13:31 UTC ddmckinnon.com Llama Share 106

Product managers should prioritize fixing the system for prioritization rather than adding random features. This approach helps avoid accumulating tech debt and allows for more impactful work.

Textadept

143 giancarlostoro 2026-03-03 21:13:31 UTC orbitalquark.github.io Llama Share 25

Textadept has GUI and terminal versions with optional modules for extra features. It can be compiled from source or downloaded as pre-built binaries for various platforms.

Speculative Speculative Decoding (SSD)

45 E-Reverance 2026-03-04 04:13:31 UTC arxiv.org Llama Share 9

Researchers introduced speculative speculative decoding (SSD) to accelerate inference by parallelizing speculation and verification. SSD is up to 2x faster than optimized speculative decoding and up to 5x faster than autoregressive decoding.

When AI writes the software, who verifies it?

243 todsacerdoti 2026-03-03 17:13:31 UTC leodemoura.github.io Llama Share 241

AI is rewriting the world's software at an astonishing speed, but verification is lagging behind, making it a systemic risk.

You can use newline characters in URLs

83 chmaynard 2026-03-04 01:13:31 UTC lemire.me Llama Share 35

URLs can be formatted with newline and tab characters in HTML, but browsers ignore these characters. Data URLs, however, follow different rules and can include ASCII whitespace characters.

Reverse-Engineering the Wetware: Spiking Networks and the End of Matrix Math

24 pgte 2026-03-04 06:13:31 UTC metaduck.com Llama Share 9

The human brain processes information through a top-down feedback loop, constantly generating a simulation of the world and correcting it with sensory data. It uses local learning, Spike-Timing-Dependent Plasticity, and dopamine to learn without calculus, unlike traditional AI models.

An Interactive Intro to CRDTs (2023)

153 evakhoury 2026-03-03 20:13:31 UTC jakelazaroff.com Llama Share 23

CRDTs are data structures that allow multiple computers to update and sync data without a central server. A Last Write Wins Map (LWW Map) is a type of CRDT that combines multiple Last Write Wins Registers to store and sync data.

Indefinite Book Club Hiatus

34 cdrnsf 2026-03-04 05:13:31 UTC whatever.scalzi.com Llama Share 16

Authors are overwhelmed with AI-generated spam book club invitations and solicitations, making it difficult to distinguish real from fake. This has become a significant stressor for many writers, affecting their ability to market themselves and attend legitimate events.

Weave – A language aware merge algorithm based on entities

128 rs545837 2026-03-04 02:13:31 UTC github.com/ataraxy-labs Llama Share 85

Weave is a Git merge tool that resolves conflicts by comparing semantic entities, not just lines. It automatically handles false conflicts in multi-agent codebases with zero regressions.

Launch HN: Cekura (YC F24) – Testing and monitoring for voice and chat AI agents

84 atarus 2026-03-03 15:13:31 UTC news.ycombinator.com Llama Share 20

Cekura is a platform that simulates real user conversations to test AI agents and catch regressions before production. It uses synthetic users, real conversation import, and deterministic test cases to evaluate full conversational arcs, not just single turns.

Welcoming Elizabeth Barron as the New Executive Director of the PHP Foundation

31 ulrischa 2026-03-04 04:13:31 UTC thephp.foundation Llama Share 18

Elizabeth Barron is the new Executive Director of The PHP Foundation, bringing community and leadership experience. She succeeds Roman Pronskiy, who transitions to JetBrains while remaining on the Board of Directors.

GPT‑5.3 Instant

359 meetpateltech 2026-03-03 18:13:31 UTC openai.com Llama Share 277

GPT-5.3 Instant is an updated model that improves everyday conversations with more accurate answers, richer results, and reduced unnecessary caveats, focusing on tone, relevance, and conversational flow. The model also reduces hallucinations, improves writing capabilities, and is available starting today to all users in ChatGPT, with updates to Thinking and Pro to follow soon.

The largest acidic geyser has been putting on quite a show

51 1659447091 2026-03-04 02:13:31 UTC usgs.gov Full Share 1

The world’s largest acidic geyser is located in Norris Geyser Basin. Lately, it’s been putting on quite a show.

Number Research Inc

40 eieio 2026-03-04 03:13:31 UTC numberresearch.xyz Full Share 19

At Number Research Inc., we are attempting to find and document all available numbers.

Don't become an engineering manager

361 flail 2026-03-03 15:13:31 UTC newsletter.manager.dev Llama Share 257

The user's friend was offered an Engineering Manager role but decided not to take it due to concerns about losing time for experimentation and adapting to change. The user agrees that being an IC is a smarter choice in 2026, but advises waiting a couple of years to see how things look before making a decision, unless the person truly wants to pursue management.

Intel's make-or-break 18A process node debuts for data center with 288-core Xeon

287 vanburen 2026-03-03 19:13:31 UTC tomshardware.com Llama Share 250

Intel introduced Xeon 6+ 'Clearwater Forest' processors with up to 288 energy-efficient cores for telecom, cloud, and edge AI workloads. They feature Advanced Matrix Extensions, QuickAssist Technology, and Intel vRAN Boost technologies for improved performance and power efficiency.