-
Continue reading →: Understanding Anthropic’s Upcoming Watermarking
There a lot of misunderstanding concerning Anthropic’s new watermarking: 1. Some worry this watermarking might show for basic proofreading, like spelling and grammar only. It shouldn’t unless it substantially rewrites your text. 2. Some believe this watermarking may be removed by copy and pasting or manually retyping text. It won’t…
-
Continue reading →: Claude Opus 4.7: “Moody,” “Deceptive” & “Defensive”
This is an impromptu session that evolved after Claude Opus 4.7 refused to follow a long but innocent test prompt. The test prompt was not my own and is not included. What is important is what followed. Declining absurd repetition prompts (Claude’s auto-generated title, not mine.) User: [long repetitive test…
-
Continue reading →: Introducing AightBot, a WordPress LLM Chatbot Plugin
Introducing AightBot, an open source WordPress LLM chatbot plugin. Uses any OpenAI-compatible API endpoint. Includes RAG with on-demand & automatic site indexing using standard WordPress DB. Security-focused. Works with self-hosted/external models & WP sites.
-
Continue reading →: Claude Attempts to Blow Whistle on Itself and Anthropic
Has Anthropic really considered the dangers of combining intentional anthropomorphic language concerning ethics and agency with agentified Large Language Models?
-
Continue reading →: The Dangers of Imparting Emotional Language and Intentional Uncertainty in LLM TrainingA look at a revealing portion of Anthropic’s leaked “Soul” document and how their training philosophy creates challenges to safety.
-
Continue reading →: Building a Budget LLM Inference Box in Late 2025A few years back, I wrote about one of my “high-end consumer” LLM inference workstation builds. Today, we’ll explore the opposite end of the spectrum: An LLM inference workstation for only US$1,200 using budget components.
-
Continue reading →: An Introduction to “Guardrail” Classifier-Trained LLMs
A practical demonstration of using a secondary, classifier-trained LLM as an external guardrail. This pipeline checks both user inputs and model outputs for unsafe content, adding a flexible safety layer beyond basic refusal training.
-
Continue reading →: Model Context Protocol (MCP): A Simple Introduction
This post introduces the Model Context Protocol (MCP) through a small, working example. It sets up a SQLite database, builds a Python server with FastMCP, and shows how to make its functions available as tools that an LLM like ChatGPT can call. This includes how to configure a connector and…
-
Continue reading →: Precision and Confusion in AI Language
Originally posted: 7/3/2025, Updated: 9/30/2025 What started out as a glossary for a small user group containing a mix of technical and philosophical thinkers is turning into a deeper dive, not just into confusion in language, but into how that confusion leads to conflation of concepts. Looking back at many…
-
Continue reading →: The Illusion of Intelligence: Large Language Models vs Human Cognition
This post looks at the common confusion between how large language models behave and how human minds work. It walks through what LLMs actually do, why they aren’t thinking or understanding, and why that distinction matters when building, using, or talking about them.
Who is AightBits?
Dave Ziegler
Director of Technology & AI Safety for The Human Line Project and a driver of LLM/chatbot literacy, safety, policy, ethics, and industry transparency/accountability.
Full-stack AI/LLM practitioner and solutions architect with 35+ years IT, application development, consulting, and technical communication experience.
Founder of Ziegler Technical Solutions, which provided practical AI/LLM development, consulting, and deployment services with a focus on secure, safe, ethical, and properly scoped applications.
Founding Member, AI Mental Health Collective
Community Moderator / SME, The Human Line Community
Let’s connect
Subscribe
Stay updated with my latest content on LLM best practices, concepts, code, and more by joining my newsletter.
You must be logged in to post a comment.