AI Agents
Why AI Agents Must Edit Their Own Context Files
Context Language Models prove that letting AI agents edit their context window like a live file beats brute-force append-only transcripts.
Topic
LLM
A collection of essays to explore, one idea at a time.
Essays
Newest first
AI Architecture
Beyond Text: How Cache-to-Cache Lets AI Models Talk Directly
Forcing AI models to collaborate via text wastes compute and loses nuance. Cache-to-Cache enables direct KV-cache transfer across neural networks.
Artificial Intelligence
Ternary LLMs Shrink 27B Reasoning to Laptop Scale
PrismML's Bonsai 2 27B achieves 98.2% parity with FP16 models in just 5.9 GB, proving ternary weights can run high-grade AI locally.
AI
Small Models Have Arrived: The Engine of Consumer AI
Frontier models get the headlines, but hyper-efficient small models are fixing unit economics and driving real-world AI software adoption.
AI Security
Beyond Prompts: How LLMs Hack Their Own Inference Engines
AI security is shifting from prompt injection to inference engine exploits, where model outputs trigger code execution on host machines.
AI
Why the Smartest AI Models Are Struggling to Sell
Frontier AI models cost 10x more for a tiny performance gain. Discover why enterprises are shifting toward model routing and cheaper LLMs.
AI
Why AI Text Watermarking Will Never Work
Statistical watermarks promise to identify AI content, but low text entropy and easy rephrasing make text AI watermarks fundamentally fragile.
AI Security
Stealing CoT: The Hidden Attack Target in AI APIs
How attackers extract hidden reasoning traces from proprietary LLM APIs, and why chain-of-thought security is becoming AI's biggest battleground.
AI
Token Profiling: The New CI/CD Gate for AI Agents
As AI agents run multi-step loops, token consumption can explode. Token profiling and cost regression gates bring software engineering discipline to AI.