The future of ChatGPT

1. The future of ChatGPT

Welcome to the final lesson.

2. Recap

Wow, we’ve been on quite the adventure.

3. Recap

We started this course by looking at the building blocks of LLMs.

4. Recap

We then explored advanced prompting techniques to get the most out of ChatGPT.

5. Recap

In this final chapter, we’ve explored the wonders of custom instructions and building our very own custom GPT.

6. Recap

...which, if you haven’t already, you must experiment with on the GPT store.

7. The big picture

For this final lesson, let’s consider the big picture of ChatGPT. Generative AI caught mainstream attention with ChatGPT’s launch in November 2022.

8. Three key ideas

1. Why ChatGPT makes mistakes 2. How ChatGPT thinks 3. Where ChatGPT is heading

9. Idea 1: Why ChatGPT makes mistakes

As we explored in “Understanding Prompt Engineering”

10. Idea 1: Why ChatGPT makes mistakes - Reversal curse

ChatGPT’s knowledge base is imperfect and often quite strange. This is best highlighted by the reversal curse.

11. Idea 1: Why ChatGPT makes mistakes - Reversal curse

Ask GPT-3.5, “Who is Tom Cruise’s mother?” It says Mary Lee Pfeiffer.

12. Idea 1: Why ChatGPT makes mistakes - Reversal curse

But if you start a new conversation and ask, “Who is Mary Lee Pfeiffer’s son?” It says Jeff Bezos! How strange.

13. Idea 1: Why ChatGPT makes mistakes - Reversal curse

ChatGPT’s knowledge is often quite one-dimensional. You need to ask questions from certain angles for accurate answers.

14. Idea 1: Why ChatGPT makes mistakes

What about OpenAI’s GPT-4o model? Let’s ask, “How many words are in your response to this prompt?” It states, “There are 41 words in my response to this prompt”. This is a prime example of hallucination where ChatGPT confidently states a wrong answer. ChatGPT wasn’t trained to understand ‘words’ like humans. Instead, it focuses on language at the token level.

15. Idea 1: Why ChatGPT makes mistakes

What’s the takeaway? To uncover the truth, we’ve got to understand how ChatGPT thinks.

16. Idea 2: How ChatGPT thinks

We must remember ChatGPT’s function. ChatGPT predicts the next word based on the previous sequence. It’s a next-word prediction tool, trained to be conversational.

17. Idea 2: How ChatGPT thinks

System 1 versus System 2 thinking was popularised by the book 'Thinking, Fast and Slow' by Daniel Kahneman. System 1 thinking is fast, instinctive, and automatic, whereas system 2 thinking is slow, rational and conscious.

18. Idea 2: How ChatGPT thinks

Why is this relevant to how ChatGPT thinks? Well, LLMs today are only capable of System 1 thinking, imitating human responses. It's hard to go above human response accuracy if LLMs are only trained on humans.

19. Idea 2: How ChatGPT thinks

Models generate words like laying train tracks in front of them. What could it mean to give language models System 2 thinking? This remains a future development I'm excited about and sparks thoughts about where ChatGPT might be heading.

20. Idea 3: Where ChatGPT is heading

In May 2024, OpenAI partnered with Be My Eyes, aiding people with blindness through GPT-4o’s multimodality. As the AI sees through the camera and speaks, Andy builds a visual image and can fully appreciate what's in front of him. This is the next level up from traditional LLMs, which just process text.

21. Idea 3: Where ChatGPT is heading

The example with Be My Eyes is a spectacular and very real step towards a much more natural human-computer interface. ChatGPT will soon be integrated into our phones, computers, and, at some point, with the recent advancements of brain-computer interfaces like Neuralink, our minds.

22. Idea 3: Where ChatGPT is heading

Extrapolating ChatGPT's progress forward means we must respect the outcome that hovers over our shoulders when we eventually reach artificial general intelligence—a system capable of producing the same output as a human knowledge worker. There are three key questions that are important for humanity to address.

23. Idea 3: Where ChatGPT is heading

How do we share wealth? How does access to these systems work? Who decides what we get to use these systems for? How we think about a world like this is not yet obvious. Answering these questions will define humanity in the years to come.

24. Congratulations!

Congratulations. You’ve nailed this course, and you should be proud of how far you’ve come. After all, there are only two things left to do.

25. Congratulations!

First, use your newfound knowledge to create something meaningful to you with ChatGPT.

26. Congratulations!

Second, wow your friends at the dinner table with your new ChatGPT knowledge.

27. Thanks!

Thank you for joining me on Intermediate ChatGPT, and I’ll see you very soon.

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.