Daily Tech News - 2026-06-28

Tech's relentless pace continues, with this week highlighting major leaps in AI, foundational software developments, and evolving privacy debates. A standout development is the official unveiling of the GPT-5.6 family, complete with a new "Ultra Mode" and advanced AI image generation tools, as reported in the AI Weekly Recap. This follows significant advancements in LLM efficiency and application, including a guide for configuring AMD Strix Halo clusters for distributed vLLM inference and the Wayfinder Router enabling deterministic routing between local and hosted LLMs. The innovative spirit extends to foundational AI, with "NanoEuler" demonstrating a GPT-2 class model built from scratch in C/CUDA, bypassing traditional ML libraries.

The drive for smarter AI is also evident in the growing focus on agentic workflows, with discussions on "Crafting an agent team that still includes me" and strategies for "Tail Control" to manage the inherent unreliability of LLMs. Benchmarking reveals impressive performance, as Semgrep notes GLM 5.2 surpassing Claude in cyber benchmarks, further expanding the open AI ecosystem.

Beyond AI, the world of operating systems sees the debut of QSOE 0.1, a QNX-inspired OS featuring selectable kernels, signaling fresh approaches to system design. Developers are also refining their craft, with new insights on identifying regression risks in Quarkus applications and optimizing container images for production. Meanwhile, privacy concerns intensify as "Flock cameras" tracking beyond license plates proliferate, underscoring the urgent need for vigilance in an increasingly monitored world.

Featured Articles

AMD Strix Halo RDMA Cluster Setup Guide

This guide details how to configure a two-node AMD Strix Halo cluster linked via Intel E810 (RoCE v2) for distributed vLLM inference using Tensor Parallelism.

TL;DR (Quick Start)
Concepts & Archit...
Keywords: cluster configure, running cluster, cluster vllm, vllm cluster, run cluster, configuration fedora, node cluster, cluster run, halo cluster, gpu node
Source: github.com

Wayfinder Router: deterministic routing of queries between local and hosted LLM

Deterministic prompt-complexity routing — send each prompt to your local or cloud model, offline, with no model call to decide. Quickstart · Benchmark · How it compares · Explainer · Changelog Wayfind...

Keywords: wayfinder decide, models routing, cloud model, models cloud, cloud wayfinder, gateway models, model cloud, wayfinder models, hosted wayfinder, routing decisions
Source: github.com

Show HN: NanoEuler – GPT-2 scale model in pure C/CUDA from scratch

A GPT-2-class language model built entirely from scratch in C/CUDA — no PyTorch, no autograd, no ML libraries. The forward and backward passes are written and verified by hand, and the whole training...

Keywords: corpus gpu, cuda engine, gpu model, model gpu, trained gpu, gpu fluent, inference gpu, written gpt, cuda pytorch, gpu engine
Source: github.com

Crafting an agent team that still includes me

That was fast. We’ve moved from prompting an LLM, to providing instructions to an agent, to having agents prompting other agents in a loop. “Loop engineering” is all the rage among the AI elite who ar...

Keywords: prompting agents, agent loop, agents prompting, agents loop, agent looped, agent tasks, agent needs, instructions agent, agent want, agent works
Source: seroter.com

A QNX-inspired operating system with selectable kernels

QSOE A QNX-inspired operating system with selectable kernels. News: QSOE 0.1 is out — the first public release of the unified two-kernel system (June 2026). Read the announcement → QSOE ships in two v...

Keywords: kernels qsoe, kernel qsoe, qsoe os, qsoe built, qsoe uses, build qsoe, qsoe variants, qsoe systems, qsoe runs, uses qsoe
Source: qsoe.net

Show HN: Bash4LLM+ – A lightweight, dependency-free Bash wrapper for LLM APIs

Bash4LLM⁺ 🇮🇹 🇬🇧 Bash4LLM⁺ — wrapper CLI sicuro, Bash‑first e completamente auditabile per l’API Chat Completions compatibile OpenAI di Groq (ed estendibile ad altri provider). Bash4LLM⁺ è un singolo s...

Keywords: bash4llm modello, bash4llm distribuito, specifico bash4llm, subito bash4llm, mitigato bash4llm, esterni bash4llm, selettiva bash4llm, bash4llm provider, bash4llm, possibile bash4llm
Source: github.com

🤖 AI Weekly Recap (Week 27)

Happy Sunday! We just had another crazy week in AI. ChatGPT has officially unveiled the GPT-5.6 family alongside a new "Ultra Mode", while there is a new free tool that can generate pro-quality images...

Keywords: ai chatgpt, week ai, week openai, agentic benchmark, ai aws, openai officially, agents plan, agents, multi agent, openai
Source: simplifyingcomplexity.tech

GitHub – librepods-org/librepods: AirPods liberated from Apple's ecosystem

Warning librepods.org is not an official website of the LibrePods project. It inaccurately claims to be the official website of the project by claiming copyrights and using the LibrePods logo in the f...

Keywords: apple librepods, librepods airpods, airpods android, use librepods, librepods org, using librepods, official librepods, available librepods, librepods project, airpods features
Source: github.com

Tail Control: The Counterintuitive Engineering of Reliable Agentic Workflows

Delivering that result is harder than it looks, because LLMs are notoriously unreliable. They fail frequently, in four flavors: an invalid answer (empty, unparseable, or simply wrong), a hard error, n...

Keywords: workflow reliability, fail workflow, workflow fast, flexible fail, workflow misses, unforgiving workflow, apis slow, unreliable fail, latency promise, step timeout
Source: towardsdatascience.com

Working around dragons with the Lemote Yeeloong laptop and OpenBSD

Now, in my case, the fact that it is a MIPS-family system meant I certainly needed one in my unusual laptop collection. And since it can run OpenBSD ... ... it seemed like a good way to get nerdsniped...

Keywords: mips laptop, ibm mips, thinkpad mips, modern openbsd, mips thinkpad, openbsd introduced, openbsd built, device openbsd, instead openbsd, openbsd ish
Source: oldvcr.blogspot.com

Artifacts 22: Zyphra, Cohere, and Poolside are expanding the breadth of the ecosystem

Artifacts 22: Zyphra, Cohere, and Poolside are expanding the breadth of the ecosystem An assessment of the open ecosystem and the motivations behind releasing models A trend we continue to see in open...

Keywords: sovereign model, models vast, model ecosystem, sovereign ai, makers models, open models, models motivations, new models, sourcing model, model makers
Source: interconnects.ai

Examining circuit boards from the Space Shuttle's I/O Processor

The Space Shuttle's five1 general-purpose computers played a critical role in each flight: controlling the engines, monitoring thousands of sensors, displaying data to the astronauts, and navigating t...

Keywords: explain processor, shuttle processor, shuttle cpu, processor iop, processor architecture, microprocessors, iop processors, physical processor, processor chips, designed microprocessors
Source: righto.com

How to Find Regression Risks in a Quarkus App Before You Write the Feature

User unregistration sounds simple until you look at the rest of the system. In a booking application, “delete this user” is never just one DELETE endpoint. It touches active bookings, foreign keys, hi...

Keywords: unregisters user, unregistration user, user unregistration, unregister, unregister really, unregisters, implementation unregister, unregister decision, user unregistered, unregistration
Source: the-main-thread.com

Lesson 62: Image Optimization - Building Production-Grade Container Images

Lesson 62: Image Optimization - Building Production-Grade Container Images What We’re Building Today Multi-stage Docker builds reducing final image size by 70-85% while maintaining full functionality...

Keywords: container images, images container, container image, docker builds, images cost, costs deployment, container security, container attack, docker, infrastructure overhead
Source: handsonk8s.substack.com

Armadillo – A DNS Server in Gleam for Homelab Use

A self-hosted DNS server for homelab use, written in Gleam. armadillo.showcase.mp4 Configure it once on your router as the DNS resolver and every device on the network resolves your local domains auto...

Keywords: runs dns, hosted dns, automatically dns, upstream resolver, dns resolver, local dns, run dns, stored dns, final dns, dns local
Source: github.com

A way to exclude sensitive files issue still open for OpenAI Codex

Notifications You must be signed in to change notification settings - Fork 14k A way to exclude sensitive files #2847 Copy link Copy link Open Labels enhancementNew feature or requestNew feature or...

Keywords: permissions sandboxing, preventing sensitive, permissions sandboxingissues, sensitive files, exclude sensitive, ignore file, mark files, model repository, sandboxingissues related, sandboxingissues
Source: github.com

Historical memory prices 1960-2026

Historic and current memory and storage prices, collected in the spirit of John C. McCallum's classic memory-price dataset — interactive, with the raw data downloadable. Hover for details, click the l...

Keywords: memory price, dram price, price gigabyte, hbm dram, unit memory, generation dram, memory type, gb cheapest, current memory, memory storage
Source: dam.stanford.edu

Reflections on Software Engineering in the Age of AI

Software Engineering in the Age of AI For those of you who don’t know, when I’m not writing novels, I spend my days as a software engineer, writing code. The software industry these days relies heavil...

Keywords: developers ai, developer ai, developer workflow, ai create, code ai, software engineering, ai code, ai process, use ai, ai write
Source: adiamond.me

Semgrep: GLM 5.2 beats Claude in our Cyber Benchmarks

We ran a set of popular open-source models against our IDOR benchmark, the same dataset and the same prompt we've used to evaluate frontier coding agents. The result surprised us: GLM 5.2, an open-wei...

Keywords: idor benchmark, open weights, benchmarks glm, harness idor, glm open, open models, ai harness, glm scaffolding, models idor, open weight
Source: semgrep.dev

Do LLMs pass the mirror test?

The mirror test — Gallup's original, the one with the red dot on the chimp's forehead — has been adapted for LLMs several times in the past, but as far as I'm concerned, every adaptation gets it wrong...

Keywords: mirror tests, test dogs, test dog, mirror test, testing human, experiment dog, testing self, evidence dog, tests, argue mirror
Source: blog.pascalschuster.de

The curious case of the disappearing Polish S

Marcin Wichary 2 February 2015 / 1,800 words Originally published in Medium Engineering The curious case of the disappearing Polish S One keyboard bug three decades in the making A few weeks ago, some...

Keywords: disappearing polish, language polish, polish language, polish keyboard, polish diacritics, polish characters, polish words, polish, key polish, article polish
Source: aresluna.org

The MUMPS 76 Primer – anniversary edition

1. Introduction
1. The Language at a Glance
1. Variables and Data Types
1. Expressions and Operators
1. Commands
5.1. SET — Assign a Value
5.2. WRITE — Output Data
5.3. READ — Input Da...
Keywords: subroutine execution, programming delimited, exit interpreter, interpreter write, terminates, halt write, terminates fors, subroutine executed, terminated command, interpreter entries
Source: github.com

Flock cameras track more than your license plate, and they're spreading fast

Flock cameras track more than your license plate, and they're spreading fast "You can't get a breath of fresh air ... without us knowing." Thanks to the rise of AI, a new kind of surveillance camera h...

Keywords: flock cameras, cameras police, surveillance cameras, cameras track, truck cameras, flock camera, cameras secure, cameras lurking, cameras documented, surveillance
Source: engadget.com

POSIX Is Not a Shell

Alexandre Gomes Gaigalas – June 28, 2026 When someone says "write it in POSIX shell for portability," they mean well. POSIX is a specification. Not a program. The thing that actually runs your script...

Keywords: posix shell, posix echo, posix sh, sh echo, posix contradict, docker run, new posix, new shell, shell different, bash dash
Source: alganet.github.io

Show HN: Decomp Academy – Learn to decompile GameCube games into matching C

Learn to decompile GameCube assembly into byte-matching C. Go from never having read a register to matching real Star Fox Adventures functions — instruction for instruction. You write C, the real 2001...

Keywords: instruction chainmultiplying, divisionmodulo instructionmultiply, subtracteda instruction, instructionmultiply, functions instruction, instruction mixed, instruction write, adda instruction, 2001 compiler, instructionmultiply addprecedence
Source: decomp-academy.dev

Tokenmaxxing is dead, long live Tokenmaxxing

Agentics / Tech Things: Tokenmaxxing is dead, long live tokenmaxxing The birth and death and rebirth of checks notes spending tons of money for no reason Generally speaking, if you spend tens of tho...

Keywords: money tokens, tokenmaxxing roi, spending tokens, million tokens, incentives tokenmax, spend tokens, tokens spend, buy tokens, token subsidies, tokens million
Source: 12gramsofcarbon.com

Bashblog – a single bash script to create blogs

A single Bash script to create blogs. I created it because I wanted a very, very simple way to post entries to a blog by using a public folder on my server, without any special requirements and depend...

Keywords: create blogs, blog bashblog, running blog, blog post, writing blogpost, blog run, blog using, blogs created, initial blog, start blogging
Source: github.com

Collabnix AI Weekly - June 2026 Edition

Your weekly digest of Cloud-Native AI, Docker Sandboxes, and Model Context Protocol innovations.

Keywords: ai docker, cloud native, docker sandboxes, cloud, docker, digest cloud, native ai, sandboxes, sandboxes model, protocol innovations
Source: ajeetraina.com