openGym is a self-hosted gym and body-weight tracker that keeps data on the user's server instead of a subscription service. It supports weekly routines, guided workouts with supersets and cardio, 1,324 exercises, progression rules, muscle recovery and detraining views, body-weight charts, imports from FitNotes, Strong and Hevy, and passkey login. A React/Vite frontend and Node API run in Docker, storing JSON and merging concurrent edits by revision. Offline use, mobile apps, and an optional server-configured AI coach distinguish it.

At Symisc Systems, more than 20 students struggled to write complete functions without ChatGPT or Z.ai’s GLM models guiding them step by step. A master's student building a Think-Act dashboard function had to retrieve JSON statistics from Redis and display them with vanilla JavaScript, repeatedly asking elementary questions. The company allows AI assistance but evaluates whether students gain independence, not merely finish. It expects an initial attempt, specific questions, and code review explanations. Its production dashboard uses the same approach.

Anthropic reported a Florida woman to police after Claude’s safety system flagged a diary-style entry allegedly threatening to “shoot up” the Lee County Sheriff’s Office. A human reviewer deemed it credible, leading to her arrest and a second-degree felony charge under Florida Statute 836.10 for a written threat of violence. Anthropic says it may disclose user information in limited emergencies to prevent death or serious injury, underscoring that chatbot conversations are not private diaries.

Reflection introduced Beam, its first open-weight model: a sparse Mixture-of-Experts system with 501B total and 23B active parameters for coding, reasoning, and agentic workloads. It was pretrained on 23.8T curated tokens and strengthened with over 100M RL rollouts across 10.5K NVIDIA GB300 GPUs over four weeks. Beam reports competitive coding and agentic performance with 3–4× less inference compute than GLM 5.2, plus controllable reasoning effort. Weights and artifacts are planned under Apache 2.0 later this month.

ChatGPT has generated New Yorker-style cartoons bearing the signatures of more than 15 real New Yorker cartoonists, including Brendan Loper, Emily Flake, and Joe Dator, without permission or compensation. The images mimic the magazine’s format and sometimes reproduce its logo, while the signatures falsely imply authorship. OpenAI added a warning about third-party similarity after being notified, but the behavior continues. Condé Nast’s licensing deal with OpenAI does not authorize training on its cartoons, raising attribution, copyright, publicity, and labor concerns.

Cloudflare launched Web Search API in beta, enabling AI agents and applications to ground responses in live Internet information. It initially supports Ceramic.ai, Exa, and Linkup, each offering Zero Data Retention through Cloudflare and following verified bot crawling standards. Running through AI Gateway, requests appear in gateway logs and use provider list pricing without markup, with support for REST API, Workers AI bindings, and customer-owned provider keys.

Node has become the author’s preferred runtime over Deno after a month using it with SvelteKit. Modern ECMAScript support, improved APIs, and no required `require()` made the transition easy; FNM, PNPM, and `tsdown` addressed versioning, dependency, and TypeScript packaging needs. Migrating a Deno static site generator to Node v26.10.0 required mainly `node:fs`, Hono’s node adapter, and `node:path`, while builds became 15% faster. The author cites Deno reliability, JSR, and integration problems as reasons for leaving.

Caddy is a cross-platform, extensible Go server platform for HTTP/1.1, HTTP/2, and HTTP/3 that enables automatic HTTPS by default. It supports ZeroSSL and Let's Encrypt, managed local CAs, clustering, dynamic JSON/API configuration, multiple config adapters, and modular plugins. Designed without external dependencies, it can scale to hundreds of thousands of sites and remain available during certificate-related failures. Its unified configuration and application-based architecture also support running broader Go programs.

AnyPS5 automatically ports PS5 executables to Linux and Windows without emulation or a separate runtime process. Its relinker converts executables to native target formats, while implementations of system prx libraries support dynamic linking. The project includes SDL-mapped controller input, configurable keyboard and mouse controls, and a shader recompiler that produces validated SPIR-V. Dreaming Sarah reaches stable 60 fps on a GTX 1050 Ti and i5-7500; unsupported states terminate with std::runtime_error.

mold 3.0.0 is the first Rust rewrite of the high-speed mold linker, replacing the C++ 2.42.1 release while preserving its command-line interface, target architectures, output, and linking performance. Tests across supported targets, real-world workloads, and all Gentoo packages found no regressions. Rust adds bounds checks for corrupted inputs. The release also improves GNU ld compatibility, fixes crashes, incorrect or nondeterministic output, relocation and architecture-specific bugs, and changes builds from CMake to Cargo.

llama.cpp v0.6.0 adds the llama_batch_ext API and llama_process() for mixed token/embedding batches and MTP/deepstack state embeddings. It supports GLM-5.3-Flash, Clef, Ling 3.0 VL, Nimble, and Qwen4Exp MTP speculative decoding, which delivers about 1.5× faster decoding on DGX Spark. The release introduces the /v1/systemone decision-model API, a Hugging Face Hub-powered Web UI and model downloads, new Metal and Vulkan attention kernels, and ggml v0.26.0.