AI Toolkit

A working list of resources, recommendations, and practical guides for navigating the AI age — curated from my own daily experience using Claude and other models, completing the Tony Robbins AI Advantage Bootcamp, and pursuing Anthropic certifications.

This is a working document. I add to it as I learn what’s actually useful versus what just sounds impressive.

Where to learn

  • Anthropic Learn — Anthropic’s official learning resources, including free courses on Claude, prompting techniques, and building with AI. Where I’m currently working on certifications.
  • Tony Robbins AI Advantage Bootcamp — I completed this in 2026.

How LLMs actually work

Worth understanding the basics — not because you need to become a researcher, but because using these tools well requires knowing what they actually are.

The short version: Large language models (Claude, ChatGPT, Gemini, and the rest) are pattern-matching engines trained on enormous amounts of text. They learned the statistical relationships between words by predicting the next word in billions of sentences. When you ask a question, the model is predicting what words should come next based on those learned patterns. That is the core of it.

That has practical consequences:

  • They can sound confident even when they are wrong. This is called hallucination, and it is why my user preferences tell Claude to push back and minimize fabrication.
  • They do not “know” things the way humans know things. They predict what knowledgeable text looks like.
  • The quality of your prompt directly shapes the quality of the output. Vague in, vague out.
  • They are remarkably good at language tasks — writing, summarizing, explaining, translating, brainstorming, drafting.
  • They are NOT good at: precise arithmetic without a calculator tool, knowing facts after their training cutoff without a web search tool, or anything requiring real-time information.

Why this matters for you: Once you understand they are pattern engines and not oracles, you stop expecting magic and start using them as the productivity tools they actually are. You verify important facts. You give context. You ask for sources. You push back when something feels off. The mental model changes everything.

Recommended explainers

  • 3Blue1Brown — “But what is a GPT? Visual intro to transformers” (~27 minutes, YouTube). The best visual explainer I have found. No math background required. Search the channel name on YouTube; it is the top result.
  • Andrej Karpathy — “Intro to Large Language Models” (~1 hour, YouTube). Slightly more technical but still accessible. Karpathy was a co-founder of OpenAI and explains the field as well as anyone alive. Worth the hour.
  • Anthropic’s own explainers at anthropic.com/learn — shorter, official, and updated as the technology evolves.

Voice-to-text — the skill that should be in everyone’s toolkit

If you’re using AI tools by typing every prompt, you’re leaving most of the value on the table. Speaking to your device is two to three times faster than typing for most people — and more importantly, it changes how you think. You speak in the language of conversation, not the language of forms. Your prompts get more natural, more complete, and more useful.

I built voice-to-text solutions at UPMC and wrote an internal training paper teaching colleagues how to actually use it well. Jensen Huang at NVIDIA now publicly says voice is the future of human-computer interaction. I’ve been quietly teaching the same thing for years before that became fashionable.

How to use it

  • On Windows: Press Windows + H to bring up dictation in any text field — email, Word, Claude in the browser, anywhere you can type. Awkward on day one. Natural after a week.
  • On your phone: Tap the microphone button on your keyboard. It’s already there on every modern iPhone and Android. Most people have been ignoring it for years. Open Messages, tap the mic, speak. Done in 5 seconds what used to take 30.
  • For longer dictation: Use the system-level voice tools — built-in Apple Dictation on Mac and iPhone, or Windows Voice Access — which handle paragraphs and punctuation better than keyboard mics.

Why it especially matters for AI

When you type a prompt to Claude or ChatGPT, you naturally write shorter and more abbreviated. You leave out context. You ask the question without explaining the situation. When you speak the same request, you give context naturally — you mention what you’re trying to accomplish, what you’ve already tried, what you’re worried about. That extra context is exactly what these models work with best.

The prompts I dictate to Claude are noticeably better than the ones I type. Not because dictation is magic, but because speaking unlocks the way you actually think.

The catch

It takes about a week of consistent daily use before it stops feeling clunky. Most people quit on day two because it feels slow and the misrecognitions are frustrating. Push through. After the adjustment period, you won’t want to go back. It’s a skill, not a feature — you have to practice it.

Practical Claude tips

Specific workflows I’ve found that get more value out of Claude. Updated as I publish new ones.

More tips coming as I find them.

Sample Claude preferences

Claude lets you save user preferences that apply to every conversation. Mine emphasize directness, pushback when I’m wrong, no softening of bad news, no unearned praise, and minimizing hallucinations. A full copyable version is coming as I refine the wording.

What I’m not pretending to be

I’m not an AI engineer. I’m not a researcher. I’m a thirty-year IT veteran who uses these tools daily and figures out what actually works in real life. This page is the messy practitioner’s view, not the academic one. If you’re looking for technical depth, the Research page is a better starting point.