AI Makes a Comic 5 Minutes After the Conversation Right in the Room Where it Happened!​

Ok, this is absolutely insane. I overheard this loud interview in a popular Palo Alto Coffeeshop known for caffeinating heavy hitters, tech bros and sisters.

The interviewer breaks the ice with a nice causal "What do you do for fun?."

The interviewee replies with "I play tennis."

In peak, competitive, tech-bro, prove-it attitude, the interviewer fires off "What's your UTR?"

WHAT .... IS YOUR FAVORITE COLOR?

I'm knocked back. The interviewee is stunned. A comic comes to mind in my head and I can't draw for shit.

Game time!

This time we use the Cursor IDE and Claude 3.5-sonnet to build a game from a simple description.

The game idea comes from my 6.111 Introductory Digital Project Laboratory course at MIT. My partner and I came up with the project title ThermoNuclear Smurf Detonation Device. Our TA called it an automatic A+ project. So we had to make it work with digital chips and no microprocessor. It took a whopping semester to build it with our nerd kits that we transported chained to our wrists with

Alright, let’s level up. Advanced prompt engineering is about precision, leveraging capabilities of AI, and squeezing out nuanced or sophisticated responses. These strategies go beyond the basics, tapping into structure, context manipulation, and iterative refinement to get results that are razor-sharp or creatively explosive. Here’s how to push the envelope with AI:

Prompt engineering is like tuning an instrument — you tweak and adjust until the output sings. Since AIs are built to be helpful and flexible, I’ll share some techniques that can work well with them. These aren’t rigid rules, just practical ways to get sharper, more useful responses.

Start with a Role or Perspective

Crafting effective prompts for Grok, ChatGPT, or any similar LLM-based AI is all about clarity, context, and curiosity. Think of it like giving a friend instructions: you want to be specific enough to get what you need, but open enough to let the AI bring something interesting to the table. Here’s how to get the most out of the AI:

Claude in action​

So I used the Cursor AI-assisted integrated development environment (IDE) pointing to the claude-3.5-sonnet model and asked it to build a dependency diagram example. We have a big project at work and am looking to map out all the dependencies like a king. Well, claude wrote the code like a king and it worked first try.

Let's play a game.​

I decided to see how well DeepSeek writes a basic snake game in JavaScript. While it understood the concept well with a minimum of prompting, it didn't do as well with the code even though it provided the following instructions once it finished coding.

Neural Networks

Boolean logic hello