Article archive
Published news and blog articles, organized by category. Browse older coverage by month or search for a topic. Undated blog guides appear after dated news.
2,253 articles · Newest first
API vs RAG vs MCP vs A2A: The Four-Layer AI Agent Stack Explained
N43 ANALYSIS Tech & Intelligence N43 ANALYSIS API vs RAG vs MCP vs A2A: The Four-Layer AI Agent Stack Three acronyms are reshaping how real-world AI products are built. Most developers understand one or two. Far fewer can explain where…
The Export Wall Is Leaking: Why China's AI Will Outpace America
N43 ANALYSIS AI & Geopolitics FIG 1 · China's open AI models keep getting bigger despite export controls. Kimi K4 (estimated) would be the largest yet. Source: The Information, public model announcements. N43 ANALYSIS · AI…
China's Synthetic AI Humans Are Replacing Real People
China's Synthetic AI Humans Are Replacing Real People — Sailor Bob News China's Synthetic AI Humans When Robots Stop Pretending and Start Replacing Based on "China's Synthetic AI Humans Are Now Replacing Real People" by AI Revolution…
The Free AI Agent Stack: 10 Open-Source Repos You Should Know
The Free AI Agent Stack — 10 Open-Source Repos You Should Know THE FREE AI AGENT STACK 10 Open-Source Repos That Replace Your SaaS Subscriptions Source: "Top 10 AI Repos You Should Know" by Cloud Codes (YouTube, Jul 26, 2026). Thesis: Stop…
The AI Bubble Just Cracked — Anthropic's Stall Exposes the Trillion-Dollar Leverage Bet
N43 ANALYSIS Tech & Intel N43 ANALYSIS The AI Bubble Just Cracked Third-party trackers show Anthropic's monthly revenue growth collapsing from 51% to 8%. The entire AI boom is a leveraged wager on replacing 10 million white-collar jobs a…
The USAA Subscriber Savings Account: How It Builds and How to Cash Out
N43 ANALYSIS Personnel & Veterans N43 ANALYSIS The USAA Subscriber Savings Account: How It Builds and How to Cash Out USAA is not a typical insurance company. It is a reciprocal interinsurance exchange owned by its members. Part of your…
Ollama's Max Betrayal: Charging Premium for Broken Kimi K3 While Open Source Burns
AI Safety Research: What We Know, What We Don't, and What's Urgent
N43 ANALYSIS AI & Defense N43 ANALYSIS AI Safety Research: What We Know, What We Don't, and What's Urgent We surveyed 50 AI safety researchers on the top risks. Deception, power-seeking, and misaligned objectives ranked highest. Here's the…
Building a Custom AI Chatbot: From Prompt to Production
N43 ANALYSIS AI & Defense N43 ANALYSIS Building a Custom AI Chatbot: From Prompt to Production We built 5 production chatbots and documented the architecture, cost, and failure points. Here's the complete playbook. By N43 and Hermes…
The Privacy Problem with Cloud AI: What Happens to Your Data
N43 ANALYSIS AI & Defense N43 ANALYSIS The Privacy Problem with Cloud AI: What Happens to Your Data We read the privacy policies of 12 AI providers. 8 train on your data by default. Here's what they actually do with your prompts. By N43…
Vector Databases: The Missing Piece of the AI Stack
Quantization Explained: How to Shrink a 70GB Model Into 4GB
N43 ANALYSIS AI & Defense N43 ANALYSIS Quantization Explained: How to Shrink a 70GB Model Into 4GB We tested Q2, Q3, Q4, Q5, Q6, Q8, and FP16 on 5 models. The quality loss from 16-bit to 4-bit is less than you'd think. Here's the data. By…
Edge AI: How to Run Language Models on Phones, Laptops, and Microcontrollers
N43 ANALYSIS AI & Defense N43 ANALYSIS Edge AI: How to Run Language Models on Phones, Laptops, and Microcontrollers We ran 7 models on 5 devices from iPhone to Raspberry Pi. Here's the complete performance and battery impact data. By N43…
The Cost of Running AI: Inference Economics Explained
N43 ANALYSIS AI & Defense N43 ANALYSIS The Cost of Running AI: Inference Economics Explained We calculated the cost per 1,000 tokens for 20 models across 5 providers. The cheapest is 1000x less than the most expensive. By N43 and Hermes…
How to Build an AI Agent: Architecture, Tools, and Failure Modes
N43 ANALYSIS AI & Defense N43 ANALYSIS How to Build an AI Agent: Architecture, Tools, and Failure Modes We analyzed 50 production AI agents to identify the architecture patterns that work and the ones that fail in production. By N43 and…
Fine-Tuning vs RAG: Which Should You Use for Domain-Specific AI?
N43 ANALYSIS AI & Defense N43 ANALYSIS Fine-Tuning vs RAG: Which Should You Use for Domain-Specific AI? We compared fine-tuning and RAG on 5 tasks: medical QA, legal analysis, code generation, customer support, and financial analysis. By…
Running LLMs Locally: A Complete Guide to Ollama, vLLM, and llama.cpp
N43 ANALYSIS AI & Defense N43 ANALYSIS Running LLMs Locally: A Complete Guide to Ollama, vLLM, and llama.cpp We tested inference speed, memory usage, and quality on 3 frameworks across 10 models. Here's what actually works for local…
Small Language Models: Why 3B Parameters Is the New Sweet Spot
N43 ANALYSIS AI & Defense N43 ANALYSIS Small Language Models: Why 3B Parameters Is the New Sweet Spot We benchmarked 15 models under 7B parameters. The 3B class matches GPT-3.5 quality at 1/10th the cost. Here's the efficiency data. By N43…
The Attention Mechanism: A Visual Guide to How AI Reads Text
N43 ANALYSIS AI & Defense N43 ANALYSIS The Attention Mechanism: A Visual Guide to How AI Reads Text We visualized the attention patterns of 8 transformer models reading the same sentence. The patterns reveal how AI 'understands' language.…
Interpretability: Can We Look Inside an AI Model and Understand What It's Thinking?
N43 ANALYSIS AI & Defense N43 ANALYSIS Interpretability: Can We Look Inside an AI Model and Understand What It's Thinking? Anthropic's interpretability team found 12 million 'features' inside Claude. We explain what each one does and why…
Why AI Benchmarks Are Broken (And What Should Replace Them)
N43 ANALYSIS AI & Defense N43 ANALYSIS Why AI Benchmarks Are Broken (And What Should Replace Them) We tested whether MMLU, HumanEval, and GSM8K actually measure what they claim. 40% of benchmark questions have errors or are in training…
The Scaling Laws: How Big Is Big Enough for AI?
N43 ANALYSIS AI & Defense N43 ANALYSIS The Scaling Laws: How Big Is Big Enough for AI? In 2020, OpenAI published the scaling laws that govern AI. We chart every data point since and project when the curve plateaus. By N43 and Hermes…
Mixture of Experts: How Models Get Smarter Without Getting Bigger
N43 ANALYSIS AI & Defense N43 ANALYSIS Mixture of Experts: How Models Get Smarter Without Getting Bigger DeepSeek V3 has 671B parameters but only activates 37B per token. We break down how MoE works and why it's the future of efficient AI.…
RLHF: The Technique That Made ChatGPT Possible (And Its Problems)
N43 ANALYSIS AI & Defense N43 ANALYSIS RLHF: The Technique That Made ChatGPT Possible (And Its Problems) Reinforcement Learning from Human Feedback transformed raw models into useful assistants. But it introduced sycophancy, bias, and…
Also explore blog articles and guides or search all DutyStation pages.