In this episode of ChatGPT Curious, I lay the groundwork for understanding what ChatGPT actually is, without getting too lost in the weeds. I break down key terms like LLM and parameters, explain how the model was trained (hint: lots of math), why it sometimes spits out wrong info, and what all of this means for how you use it. I also touch on the environmental cost, what the free vs. paid versions can actually do, and how to think critically about its outputs. If you've ever felt a little confused, a little curious, or both, this one's for you.
Main Topics Covered
- Why it’s worth understanding the foundation of ChatGPT
- What is ChatGPT?
- Brief history of OpenAI and the development of GPT
- What GPT actually stands for and what changed with each version
- What “parameters” are and how they shape the model’s responses
- How language is turned into numbers via tokens
- What happens during training
- Human involvement in model training via RLHF (reinforcement learning with human feedback)
- Probabilistic vs deterministic systems and what that means for output accuracy
- The environmental cost of “compute” and an analogy for mindful use
- What the free version can do (and can’t), including search, uploads, and voice
- What the paid version offers
- What ChatGPT is not
- What to watch out for
- Real-life use case: Trying to fix a bike derailleur using ChatGPT
Links & Resources for This Episode
Chapters
- (00:00:00) - ChatGPT Curious
- (00:00:38) - What Is ChatGPT?
- (00:02:47) - ChatGPT
- (00:04:32) - parameters of OpenAI's LLMs
- (00:06:11) - Laplike 2.8
- (00:13:31) - ChatGPT and the Probability of Outputs
- (00:19:22) - ChatGPT: Algorithms That Predict the Next Word
- (00:21:39) - ChatGPT: Probabilistic Models (
- (00:28:28) - ChatGPT: What Can ChatGPT Do?
- (00:31:00) - What to Watch Out For With ChatGPT
- (00:36:13) - A Brief History of Chat GPT
