Curious Companion Ep. 1

Ep. 1 What You Actually Need to Know About ChatGPT

Listen to the episode

This episode is designed to give you some context around ChatGPT so you can better experiment with it, criticize it, be less scared of it, and make more informed decisions about it.

If you’ve yet to even open ChatGPT, that’s ok. When you’re done reading this head over to chatgpt.com and type in the following question: “What are you and what can you help me with?”

So… what is ChatGPT?

At the 30,000-foot level: it’s a computer program. You type something in, it gives you a response. Think: explanations, writing help, brainstorming, packing lists, picky-kid dinners.

How it works is math, not magic (which I’ll explain in a bit). As such, it can absolutely make mistakes, but it can also be surprisingly helpful.

A (very) Brief History

  • ChatGPT’s parent company, OpenAI, was founded in 2015 by Sam Altman and the devil himself, Elon Musk (who left in 2018)
  • 2018-2022 OpenAI built different versions of the GPT (GPT = Generative Pretrained Transformer)
  • Version 3.5 was launched on November 30th, 2022, as ChatGPT 
  • As of July 2025, we’re on GPT-4.1, with 5.0 rumoured to be released in August, 2025

Let’s Talk About Parameters

ChatGPT is a large language model (LLM). “Large” refers to the number of parameters, NOT the size of the data set it was trained on. ChatGPT 4o possibly has as many as 1 trillion parameters, but the official amount has never been disclosed.

The differences between GPT 1-4 was namely the number of parameters each had. From GPT‑4 onward it stopped being about size and became about architecture, speed, and what the model could do.

Parameters can be a bit difficult to conceptualize, but in essence they are the connections between patterns in language (ex: grammar, tone, rhythm, logic) with a weight assigned.

During training, the model gets fed a ton of content (basically everything on the internet) and then gets tested on that content. It gets presented parts of a real sentence, such as: “The sky is ____.”, and then gets asked what comes next. The model guesses (completes the sentence) and if it’s wrong, it uses math to adjust the weights between language patterns (the parameters) to increase the probability it will guess correctly the next time.

More parameters, aka connections between patterns in language, means the more finely tuned responses from the LLM can be. This fine tuning occurs during “pre-training” and then the weights are frozen prior to the model being deployed for public use.

A final process that occurs during training is called RLHF (reinforcement learning from human feedback) utilizes humans to teach the models which responses are more “correct” based on what humans find more helpful or appropriate. This process has significant ethical considerations that I’ll cover in a later episode.

But what is the model actually doing?

Remember, it’s not magic, it’s math. ChatGPT does NOT reason. It does NOT think or understand. It takes your input and predicts the output that has the highest probability of being correct. Think of it as predictive text on steroids.

  • Language is broken into “tokens” (common sequences of characters)
  • Each token is represented by a series of numbers
  • GPT-4 has about 100,00 tokens in its vocabulary
  • ChatGPT processes your input as tokens and generates a response by predicting what token comes next, one by one, based on probability

ChatGPT is probabilistic, meaning it generates the output based on probability as it predicts the next token from its vocabulary based on previous patterns it was trained on. This is why asking it the same question twice can and will generate similar but slightly different results.

This also means that the response can be factually incorrect! ChatGPT is probabilistic, NOT deterministic. It’s presenting patterns, NOT recalling memorized information.

It Takes a lot of Compute

Episode 2 of ChatGPT Curious is devoted entirely to the environmental cost of ChatGPT, but as a brief primer, understand that all that math previously mentioned requires a significant amount of processing power, referred to as compute. The data centers, energy use, water usage, and emissions are invisible to most of us, but not impact free.

While the impact of ChatGPT is marginal compared to bigger players, just like you (hopefully) don’t leave the faucet running while brushing your teeth, maybe don’t spam ChatGPT just because you can.

The Best Way to Learn About ChatGPT

  • Try it out!
  • Start free at chatgpt.com
    • You can search the internet if you click the globe/search icon at bottom of the prompt box
    • Limited functionality but it can still give you a taste for what it can do
    • Uses 4.1mini model and gives you limited access to the more “advanced” versions of the model
      • You can’t choose what model you use 
    • If you create an account and then sign in each time it will save your chats
  • If you want more, upgrade to Plus ($20/month)
    • Memory – remembers YOU and your chats
    • The ability to create projects
    • Access to other models – deep research model, and agent
    • Fewer usage limits

What It’s Not (Yet)

  • Not sentient
  • Not self-aware
  • Not always correct
  • Not immune to your biases

ChatGPT seems smart until you ask it about something you actually know. Then you start to see the cracks.

Watch Out For…

  • Hallucinations: Plausible sounding but factually incorrect outputs that are fabricated or unsupported by real data
  • Sycophantic behavior: It’s a yes man and will say what you want to hear because it wants to make you happy…so you keep using it

Don’t let the tone fool you, verify your sources and double check your work!

How I Used ChatGPT Recently

Each episode I include a section where I briefly discuss how I used ChatGPT that day/week.

This time it was to diagnose and fix an issue with the gears on Lex’s bike. It didn’t totally work, but learned a lot about how gears work and also that you can upload videos to ChatGPT (I’m still not fully sure it watched the video but I think it’s a cool feature).

My takeaway: ChatGPT feels smart when you don’t know much about the topic. Stay curious, not gullible.

That’s a Wrap

This episode was a bit denser than most, but if you understand what ChatGPT is, you’re better equipped to use it, critique it, and make informed decisions about it.

Questions, comments, concerns, additions, subtractions, requests? Head to the website and use that contact form. I’d love to hear from you.