ChatGPT4o: The good and unexpectedly the bad?

Hey hey AI Soldiers,  

AI Dispatch Guy here!

It's been too long since we've been waiting for another model from OpenAI. 

Good news: the model is awesome and is the first AI assistant that actually feels like a human. Bad news: it's not GPT-5; this time it’s GPT-4 Omni.

In today's edition, we talk about GPT-4o, how to access it, the bad, the good, and the wild. 

What the h*ck is ChatGPT-4o?  

Well, it's the model that can video chat with you, talk as fast as humans (near-zero latency), handle audio, and generate images all in one place.

It’s an improvement on GPT-4 Turbo, twice as fast, 50% cheaper, and has 5X the rate limit (10 million tokens per minute). 

But in short, it's the AI companion from the movie “Her.” 

See this tweet to see how flirty it sounds haha 👇️ 

How to Access GPT-4o?

Now, how do you get access to ChatGPT-4o? 

• Just go to chat.openai.com and sign up. 

• From the dropdowns, just select GPT-4o. 

If you are a paid subscriber, this should already be available.

For free users, if you find it, you are in luck. If not, they will be rolling GPT-4o out to free users slowly. 

Wild Examples of GPT-4o ✨ 

I took GPT-4o for a spin and tried out use cases I found on Twitter to see what all the hype is about. 

Only including the ones not involving video or audio (Cause newsletter, duh! 🤷‍♂️) 

1. Draw to Code: 

I drew with a mouse (Forgive the bad drawing 😬)  

Input:  

Prompt: Show me the code to generate this and also make one yourself here. Keep it really short. 

Output:  

2. Emotion Analysis:  

Input:  

Prompt: Tell me what the person in the picture is feeling right now.  

Output:  

Based on the expression in the picture, the person appears to be feeling happy and relaxed. Their broad smile and the comfortable posture with hands behind the head suggest a state of contentment and ease. 

3. Correcting False Information:  

4. Making Games:  

I had to give this a try. I generated the code via ChatGPT and hosted it on GitHub (with instructions from ChatGPT). Took me 10 mins everything included.  

Check it out here 👉️ : Pong-Game Created with GPT-4o Instructions

5. Understand Old Texts:  

The text is from the 1800s.  

Input: 

Prompt: Tell me what is written here. 

Output:  

6. Create Visuals from Files:  

If you give it any file containing a dataset and tell it to create visuals (graphs, charts, plots etc.), it will do it. 

Input:  

Create some visuals from this dataset. Only create the visuals and nothing else. And additionally, create a visual on a world map on the location where which item was sold the most. Also, compile all 4 visuals into one frame in the end. 

Output:  

The craziest thing is that it even generates visuals on a world map. The dataset had sales figures in different countries. This world map shows which product sold the most.

Things GPT-4o Sucks At 👎️ 

Now, GPT-4o seems clearly a better model than GPT-4 Turbo, but is it? 

Not really, list of things GPT-4o sucks at (from what I saw on tweets and Reddit posts):

1. Misspelling while generating logos 

2. It can get a bit confused when it comes to understanding and displaying clock logic correctly.

3. Some said GPT-4 Turbo has better puzzle-solving capabilities than GPT-4o. If you are using GPT4o, tell it to use Chain of Thought. 

4. Some people noticed it doesn't follow system instructions as well as GPT-4 does.

5. Makes mistakes while editing stuff, adds extra misinformation that wasn’t needed. Yikes! This limits the use cases a lot. 

From what I understood, it sacrificed precision for creativity. The hallucination rate also went up.

Tools of the Week: 🧰 

That’s it for today! Let me know in the reply how you liked today’s edition!

AI Dispatch Guy, signing out! 🫡