Hey all, I grabbed the captions from the video below, threw it into my ChatGpt ‘Cody Sharp 1.0 Model’ I made that tries to write like me (only with more clarity but unfortunately less asides). For the readers out there, this is pretty good summary with some clarifying statements – I’ll find the reference links and throw them below as well.  Warning – there are no AI filters used on the visuals within the video below unfortunately – I’m stuck looking like that.  

Hello everyone, this is Cody Sharp from Sharp Guys Web Design. Today, we’re going to delve into an interesting topic: AI, specifically, the things AI doesn’t do very well. As marketers, we always strive to find niches where we can shine and set ourselves apart. One such area where humans still outshine AI? The ability to solve logical word problems.

Let me illustrate this with an example. I coach a baseball team and I ask the boys to participate in a ranked-choice voting system for the All-Star game. They need to choose their top three peers for the team, with the first choice worth 3 points, the second 2 points, and the third 1 point. This introduces the boys to the concept of ranked-choice voting and also makes the selection process fairer. I always encourage them to vote for themselves first, just as any politician or any person aiming for something would do.

This system works well in practice. I end up with a list of boys and their respective points – a simple list that any 3rd or 4th grader could easily understand and manage. However, when I tried to get AI to process this list, things didn’t go as smoothly as I had anticipated. Even though the task was straightforward, the AI failed to accurately calculate and assign the points.

I was certain that AI, with all its progress and capabilities, would be able to manage this simple task. However, when I uploaded this list to Bing, the results were disappointing. The AI failed to calculate the points correctly, giving some boys too many points and others not enough. Even after pointing out the error and asking for a recalculation, the AI still couldn’t get it right.

This got me thinking. We often assume that AI knows what it is doing, especially when it comes to tasks involving math. However, this example shows that even basic math can stump AI. It’s not about the complexity of the task, but rather about the logic involved and the AI’s understanding of it.

Now, you might argue that AI will improve over time, and that’s certainly true. However, the rate of improvement is not always exponential. Sam Altman, the CEO of OpenAI, once said that the step from OpenAI’s ChatGPT 3 to 3.5 saw a massive improvement, but the subsequent step from 3.5 to 4 was only a marginal one. This is how technology generally progresses – incremental gains based on the same foundational technology.

To make a significant leap forward, we need an entirely new technology or approach, just as Google’s open-source research paper on transformers enabled a significant leap in the development of ChatGPT. However, despite these advancements, there are still tasks that AI cannot handle well, like the simple math and logic problem I presented earlier.

This doesn’t mean that AI isn’t useful or powerful. On the contrary, AI can perform complex math and provide correct answers most of the time. However, we can’t rely on it blindly. For example, lawyers who relied on ChatGPT to draft a lawsuit against an airline ended up referencing nonexistent cases and airlines, leading to embarrassment and potential legal complications.

As we continue to use AI, it’s essential to be aware of its limitations and use it judiciously. For example, the new “Browse the web” add-on for ChatGPT Plus allows me to find reference links for a lot of the AI’s statements. This is an excellent way to verify information and ensure accuracy.

In conclusion, as you embark on your AI journey, remember that while AI can be a powerful tool, it’s not infallible. Always double-check the information, especially when it comes to All Star game tallies.