-
Best Graphics Cards for Running Local AI Models in 2026
A practical, budget-tiered guide to choosing a GPU for running local LLMs in 2026, plus a quick script to estimate VRAM requirements before you buy.
-
From ELIZA to GPT: A History of Chatbots
From ELIZA’s 1966 pattern matching to modern LLMs: six decades of chatbot history, and why the “ELIZA effect” still applies to today’s systems.
-
The Real Cost of Running AI Models at Scale
The real cost of running AI at scale: the frequently overlooked expenses beyond visible per-token pricing or raw GPU cost.
-
Small Language Models vs Large: Trade-offs Explained
Small language models vs large: the real trade-offs, and why a tiered small-model-first, escalate-when-needed approach is increasingly common in production.
-
Serving LLMs at Scale with vLLM
Serving LLMs at scale with vLLM: the memory-efficiency problem naive serving has, and how PagedAttention solves it.
-
Self-Supervised Learning Explained
Self-supervised learning explained: the key insight that let language models train on massive unlabeled data without costly human labeling.
-
Building a Simple Chatbot Using an LLM API
Building a working chatbot with an LLM API in a few lines of code, and the real engineering work — context management and guardrails — that…
-
BIOS vs UEFI: What Actually Changed
What BIOS’s real limitations were, and what UEFI’s GPT partitioning, Secure Boot, and modern execution environment actually add over it.
-
Rate Limiting APIs: Strategies and Trade-offs
Rate limiting strategies compared: fixed window’s burst problem vs token bucket’s smoother behavior, and how to communicate limits to API clients.
-
Small vs Large Context Windows: Cost and Latency Trade-offs
Small vs large context window trade-offs: why bigger isn’t automatically better, given the real cost scaling and “lost in the middle” attention problem.
-
Why Do AI Models Hallucinate? A Practical Explanation
Why AI models hallucinate: it’s an inherent consequence of next-token prediction, not a simple bug, and why that means mitigations reduce but don’t eliminate it.
-
The History of AI-Powered Voice Assistants
The history of AI-powered voice assistants: from Dragon NaturallySpeaking’s narrow commands to today’s genuinely open-ended LLM-powered conversation.
-
OAuth 2.0 Explained Without the Jargon
OAuth 2.0 explained in plain terms: the authorization code flow, and why access and refresh tokens serve genuinely different purposes.
-
Writing a Simple Compiler: Lexers and Parsers Explained
Building a tiny arithmetic expression compiler from scratch: a lexer that tokenizes, a parser that respects operator precedence, and an evaluator for the resulting tree.
-
Understanding Context Windows and Why Bigger Isn’t Always Better
What a context window actually measures, the “lost in the middle” problem with long contexts, and why sending only relevant content usually beats sending everything.
-
Running Local LLMs on Apple Silicon Macs
Running local LLMs on Apple Silicon Macs: why unified memory lets them handle larger models than comparably priced PCs, and the real speed trade-off.
-
Observability for LLM Applications: What to Monitor
Observability for LLM applications: why traditional latency and error-rate monitoring misses the AI-specific failure modes that actually matter.
-
The Transformer Architecture Explained, Layer by Layer
The Transformer architecture explained layer by layer: self-attention, multi-head attention, and why decoder-only models became the dominant LLM design.
-
Supervised vs Unsupervised Learning Explained
Supervised vs unsupervised learning explained with concrete code examples, and where modern LLM training actually fits between the two.
-
Understanding Tokens: How LLMs Read and Write Text
How byte-pair encoding tokenization actually works, why it explains models struggling with letter-counting, and how to estimate token counts and cost before sending a request.
-
Mixture of Experts: How Sparse AI Models Scale Efficiently
Mixture of Experts explained: how routing tokens to a subset of specialized “expert” networks scales capability without proportionally scaling inference cost.
-
AI Model Quantization Formats Compared: GGUF, AWQ, and GPTQ
AI model quantization formats compared: GGUF’s CPU-friendly flexibility versus AWQ and GPTQ’s more sophisticated GPU-optimized calibration.
-
Running AI Inference on a Raspberry Pi
What’s realistically achievable running AI inference on a Raspberry Pi, from small quantized LLMs via llama.cpp to real-time object detection with the Hailo AI HAT+.
-
Fine-Tuning a Small Language Model with LoRA
How LoRA makes LLM fine-tuning practical on a single consumer GPU by training small adapter matrices instead of the full model, with a working PEFT example.