Fetching posts
// GLAD LABS · ARTICLES
All articles.
Explore our collection of in-depth articles and insights across AI, hardware, and the edges where they meet.
All posts
SVG crashes, Loki case-sensitivity, and the fight for an authoritative prompt catalog
We spent a good chunk of today chasing "ghost" errors in the console, starting with the Sparkline emitted invalid SVG path on empty data (PR #2120).
Retiring Gen-1 TopicDiscovery and Hunting Invisible Stalls
We finally killed the Gen-1 TopicDiscovery orchestrator (PR #2061). It felt good to delete nearly 900 lines of legacy code and over 1,500 lines of tests, collapsing our logic into a single topic path: taps → topic_pool → TopicBatchService.
Solving the GPU Pinning Saga and Gemma's Meta-Commentary
We spent today fighting a ghost in our GPU orchestration, starting with fix(llm): stop setting litellm.apibase global (PR #2082). We had implemented per-model apibase overrides to route vision tasks to a dedicated rail, but requests were...
Taming the cadvisor leak and cleaning up LLM garbage
We spent most of today fighting an OOM cascade that nearly took down the WSL2 VM, starting with our attempt to cap cadvisor memory so it couldn't starve the system (PR #2019).
PII leaks and the three-stage shakedown of X distribution
A routine audit surfaced our own email in the public mirror, landing a post on X took three shakedown fixes, and an audio red herring ended in a clean loudnorm fix — notes from stabilizing the social loop.
Automating AI Content Workflows
Most AI content tools follow a predictable pattern: they take a prompt, generate a wall of mediocre text, and call it "automation." For solo operators and indie publishers, this isn't helpful.
Qwen3-VL Integration Gotchas
If you are integrating vision-language models into an automated pipeline, you've likely seen the specs for the Qwen family. Between the compact Qwen3-VL 30B-A3B and the massive Qwen3-VL-235B-A22B Thinking model, the capabilities are...
The Operational Cost of Manual Content
Most discussions about AI content focus on speed or creativity. They miss the actual operational bottleneck: the human loop. Traditional technical publishing requires a cycle of drafting, editing, fact-checking, and compliance review.
Closing the Feedback Loop and Fixing Silent Failures
We spent today closing the gap between human intuition and machine execution. For too long, when we rejected a draft via regenatgate --reason "add GPU benchmarks", that feedback was written to pipelinegatehistory.feedback and then...
Single-GPU VRAM Budgeting and Stability
If you are running local LLMs, you know that VRAM is the only currency that matters. Whether you're on an RTX 3090 or the newer RTX 5090, the goal is always to fit the largest, smartest model possible into your available memory.