PlayAgit / Notes for builders
Build. Play. Explore.
A technology blog covering game development, AI, automation, and industry news — notes from what we actually build and run.
-
NVIDIA Puts Dropless Mixture-of-Experts Training on FP8 in JAX — and Ships a Convergence Debugger Next to It
NVIDIA has published a developer blog post titled “Accelerating Dropless MoE Training in JAX with NVIDIA Transformer Engine” (NVIDIA developer blog).
Latest from the workbench
-
One Official Repository, Many Reader Copies: What the Public Record of Overview Rust Actually Contains
Hands-On Rust teaches Rust through 2D games like Flappy Dragon and a dungeon crawler. Where the author’s code lives, and what readers published around it.
-
Alibaba Cloud and Cambricon Take Platinum Seats at the PyTorch Foundation — and One of Them Ships a Proprietary Compiler
The PyTorch Foundation announced that Alibaba Cloud, Ant Group, Cambricon and Huawei came together at PyTorch Conference China, held in Shanghai, to advance the open source AI stack (primary source: the PyTorch Foundation’s own announcement).
-
Three Things Called StemDeck: An Organization, a Local DJ App, and a Browser Practice Tool
Searching for “StemDeck” surfaces a project that presents itself in one line: a free, open-source AI stem separator that runs locally (source: the project’s own GitHub listing; not independently verified).
-
Ten Men Named, a State Lawsuit, and an $18 Million Settlement: The Activision Blizzard Cases from 2021 to the Microsoft Deal
The short version: press reporting from mid-2021 onward describes a cluster of overlapping legal matters — a private harassment suit, a state regulator’s suit, a federal employment-agency settlement, a labor complaint, and later an antitrust challenge to an acquisition — none of which has been verified here beyond the reporting itself.
-
Unity Ships an Official Claude Code Plugin That Drives the Editor Live — and Our Own 735 Claude Runs Show Where It Will Stop
Unity announced an official plugin for Claude Code, described in its own blog post about the plugin as installable with a single command.
-
Every Agent Run Leaves a Trace — Except the One Built to Leave Nothing
The simplest thing an agent leaves behind is wreckage: a diff, a log, some files that were not there before.
-
The Signed Receipt for a Write That Never Happened: How Agent Execution Records Became Evidence
The premise behind this whole category is that an AI execution can be watched but not re-run.
-
When an Agent Reports Success and Nothing Happened: Separate Verifiers, Isolated Workers, and Playbooks Built from Traces
The failure mode that keeps surfacing in agent discussions is not a crash.
