The Revenge of the Stack
Nvidia showed that controlling the full stack wins. China’s AI chip race is less about silicon alone than escaping CUDA dependency.
Development & Infrastructure
28posts
Natural Language Is the JSON of AI Agents
Language strives for simplicity - until machines use it as plumbing. AI agents may work better when text moves to the boundary, not the core.
The New Plumbing of AI: MCP, OKF, and the Battle for Context
AI is entering its plumbing phase: MCP connects agents to tools; OKF packages human-readable knowledge for machines. Context becomes infrastructure.
GreenBoost: What NVIDIA Does Not Deliver
GreenBoost stretches consumer GPU memory beyond VRAM limits, exposing what CUDA infrastructure could have delivered but didn’t.
PTP/1.0 — Prompt Transport Protocol
A playful mock protocol imagines prompts as transport packets, turning generative reconstruction into a deadpan internet standard.
Google Stitch Is Not the End of Design
Google Stitch is powerful, but the post argues that faster UI generation changes design work rather than eliminating design judgment.
CALM and the Revolt Against the Token
Continuous Autoregressive Language Models challenge the token-by-token bottleneck and hint at a different future for language generation.
Your AI Writes Tests That Pass. That Is the Problem.
AI-generated tests can look reassuring while proving very little, exposing a dangerous gap between green checkmarks and real verification.
The ROM Listing Spirit Lives On in Apple Silicon
Apple Silicon's reverse-engineered Neural Engine revives the old personal-computing spirit of manuals, memory maps, and productive trespass.
COBOL Was Never the Product. The Toll Booth Was.
COBOL modernization is not just a technical story; it threatens the consulting toll booths built around legacy systems.
Prompting’s Midlife Crisis: Turning Incantations into Infrastructure
Prompting is outgrowing folklore and becoming infrastructure: specifications, patterns, evaluation, and operational discipline.
PageIndex.ai: A persuasive “vectorless RAG” idea—especially for real PDFs
PageIndex.ai makes the case for document-aware retrieval that respects pages, structure, and references instead of blindly chunking PDFs.
Meta-Prompting: How to Get More Signal Out of Your Prompts
Meta-prompting treats the prompt itself as a draft to debug, producing clearer goals and fewer disappointing model outputs.
Programming as the Art of Naming
As AI writes more code, naming becomes even more central: the human craft shifts toward concepts, boundaries, and meaning.
Bridging Context Engineering in AI with Requirements Engineering
Context engineering and requirements engineering converge, suggesting better ways to specify AI-assisted software before code is written.
From Prompt Packs to Purpose-Built Models: When a Generalist Becomes a Specialist—and When It Still Doesn’t
Prompt packs can make general models behave like specialists, but the post asks where scaffolding ends and real specialization begins.
AI Magnifies Your Team’s Strengths and Weaknesses – Insights from Google’s 2025 DORA Report
Google's DORA findings suggest AI amplifies team quality: strong practices get stronger, broken processes get louder.
Neural Texture Compression: Revolutionizing Game Graphics for Gamers and Developers
Neural texture compression promises richer game graphics with lower memory costs, changing the pipeline for artists and developers.
Choosing the Right OpenAI API Interface: A Developer’s Guide for 2025
A developer-focused guide to choosing between OpenAI's Chat Completions, Responses, and Assistants APIs in 2025.
From 80s Supercomputers to Mac Studio Clusters: The MLX-Powered DIY AI Revolution
From Cray supercomputers to Mac Studio clusters, the post traces the strange continuity of DIY AI horsepower.
Bridging the Gap: RAG, OpenAI API, Anthropic MCP, and Ollama LLMs
A bridge between RAG, OpenAI tools, Anthropic MCP, and local Ollama models for more grounded AI systems.
How DeepSeek's Mathematical Optimizations Complement NVIDIA's NCCL for Efficient AI Training
DeepSeek's mathematical optimizations show how model design and NVIDIA communication infrastructure meet inside efficient training.
The Weight of Knowledge: Connecting Project Strawberry to a Physicist’s Sweet Revelation
Project Strawberry and the physical weight of the internet meet in a playful reflection on knowledge, storage, and scale.
The Five Levels of AI: Has OpenAI's 'Strawberry' Reached Level 2?
OpenAI's Strawberry rumors are mapped onto staged AGI levels, asking what real reasoning progress would look like.
Taking the Temperature of AI: How THERMOMETER is Turning Up the Heat on Overconfident Language Models
THERMOMETER targets overconfident language models, offering a way to calibrate systems that bluff too easily.
A Short Trip Through Data Science Workflows With KNIME
A practical introduction to KNIME and the shift from fragile spreadsheet work toward reproducible data workflows.
Unlocking the Potential of Large Language Models: A Guide to Effective Prompt Engineering
A practical guide to prompt engineering techniques for getting more reliable, useful behavior from large language models.
Why Mojo is the Secret Sauce in the AI and ML Kitchen
Mojo is presented as a promising language for AI and machine learning, blending Python-like usability with systems-level speed.