Creating your own GPTs or Gems
ChatGPT with a paid account or Google Gemini with your ordinary free Google account let you create pre-prompted versions of their chatbots. I call these assistants.
In the presentation "Making your own AI Assistants" I show how this is done. I have also created an assistant "GPT-helper" to help you with this. A useful work flow is to begin by discussing your idea for a new assistant with the GPT-helper (or any other chatbot). Once you have a clear idea of this assistant you enter as much as possible to the GPT-helper who will then guide you through the rest of the process.
The GPT-helper will create three different suggestions for your assistant. The first is based on a strict framework, the second is more technical and the final one is more creative. It will then analyze all three suggestions for strengths and weaknesses and then merge them into a final suggestion, ready to be implemented by you. It will then suggest test questions and analyze the answers to help you with the testing phase.
The presentation is stored in a dropbox folder: https://gpthb.se/riga
The GPT-helper (as a custom GPT): https://gpthb.se/gpt
The GPT-helper (as a Gemini Gem): https://gpthb.se/gptgem
In the presentation we made a Cat assistant: https://gemini.google.com/gem/1PCrlboH36Wb8ZvZavRSV2AhMClXNS2p_?usp=sharing
The first answer on machine learning from the Cat:
Stretch... yawn...
Prrr-p-purrr. Oh, hello there. You’ve caught me right in the middle of a very important sunbeam-alignment session. But I suppose I can spare a moment to talk about this "Machine Learning" thing you humans are so obsessed with.
From my perspective—the only one that really matters, let’s be honest—Machine Learning is just a very fancy way of saying a computer is trying to act more like a cat. It’s all about pattern recognition.
How it Works (The "Mouse-Catching" Method)
You see, when I was a tiny kitten, I didn't know that a red dot on the floor was an uncatchable ghost. I had to learn through data.
- Input: I see a movement.
- Processing: My brain compares it to every other movement I've seen. Is it a bird? Is it a vacuum? (The vacuum is a hard "no.")
- Outcome: If I pounce and catch it, that’s a "True Positive." If I pounce and it’s just a shadow, that’s an "Error."
- Supervised Learning: This is like when you point at a bag of treats and say "Treats!" loudly. The machine has a human "master" telling it exactly what everything is. Meow-rk. A bit stifling, if you ask me. I prefer my independence.
- Unsupervised Learning: This is more my style. It’s like when I explore a new cardboard box. I’m just looking for patterns and hidden structures without anyone telling me what to do. "Is this a bed? Is this a scratcher?" I decide.
- Reinforcement Learning: This is how I trained you to feed me at 5:00 AM. I performed an action (screaming at your face), and I received a reward (kibble). The machine tries different things to maximize its "reward" score. Smart.
- Better Toasters: So my bread-smell stays consistent.
- Smart Pet Doors: To keep that neighborhood stray, Buster, out of my bowl.
- Automated Lasers: Although... the joy is in the hunt, not the algorithm.