Fine tune GPT-4o to work better for what you need

Hey there!

Let me take you on an AI ride.

So, there's this town called Cheyenne in Wyoming. Recently, something really weird happened there.

A guy named Victor made an AI version of himself using ChatGPT. But he didn’t stop there—he ran for mayor, with his AI clone as his running mate!

Yep, he wanted a computer program to help run the town!

Victor said if he won, the AI would make all the decisions.

Can you imagine a robot mayor? Seriously I can’t imagine.

But when people voted, they weren't into it. Victor and his AI only got 327 votes out of more than 11,000.

That's like getting one slice of pizza when everyone else gets 33!

Victor says he started a "revolution," but I think he just gave everyone a good laugh.

What’s Inside Today’s Newsletter

🪰 Buzz Around AI

😀 AI Recap

🪄 Prompt Of The Month

🔹 AI Learning

♨️ What’s Hot In AI

📰 AI News

Read time: 5 mins

🪰 Buzz Around AI

You can now teach GPT-4o to work better for what you need.

This helps it do a better job for your tasks.

You can now make OpenAI's top-notch AI, GPT-4o, work just for you.

Think of it like teaching a robot to cook your favourite meal exactly how you like it. Every. Single. Time.

Here's the best part - from now until September 23, OpenAI is giving away free samples.

You can teach the AI about 750,000 words worth of stuff each day. And get this - even if you're not a coding person, you can do this!

Wanna give it a shot?

Here is how you can use it:

1. Head over to OpenAI's Fine Tune page.

2. Pick "gpt-4o-2024-08-06" as your AI buddy.

3. Feed it your teaching info.

Now, about that teaching info - it needs to be in a special format called JSON.

You don't need to freak out! It's not as scary as it sounds.

Each example just needs three parts:

1. A "system" bit (telling the AI its job)

2. A "user" bit (what you'd ask)

3. An "assistant" bit (how you want it to answer)

Let me show you what I mean. It looks something like this:

{"messages": [{"role": "system", "content": "You're a friendly cook who loves sharing recipes."}, {"role": "user", "content": "How do I make scrambled eggs?"}, {"role": "assistant", "content": "Scrambled eggs coming right up! Here's how: 1. Crack 2-3 eggs in a bowl. 2. Add a splash of milk and whisk. 3. Heat a pan with butter. 4. Pour in eggs, stir gently as they cook. 5. Take 'em off when still a bit wet. 6. Salt and pepper to taste. Voila! Enjoy your eggs!"}]}

Just remember to put each example on its line. You'll need at least 10 examples, but 50-100 is even better. The more you teach it, the smarter it gets!

Once you've got all that sorted, give your new AI a cool name (like "Chef-Egg-spert"). Keep the other settings as they are, and hit that "create" button.

When it's all done and dusted, pop over to the OpenAI Playground and have a chat with your new AI friend.

See how well it's learned your style!

Just a heads up - this is different from making a CustomGPT where you can use text or PDFs. For this fine-tuning thing, you gotta use that special JSON format we talked about.

A tiny startup called Hotshot just swooped in and beat OpenAI Sora.

OpenAI was working on a tool called Sora that turns words into videos, but it's not ready for everyone to use yet.

Meanwhile, a small company called Hotshot, with just four people. Managed to make their word-to-video tool and turned completely the table. They did this really fast, in just four months.

We have seen many companies like Runway, Luma, and Pika have also made tools like this. But to my curiosity, I wanted to know what Hotshot has and how it’s better than others in Just 4 months.

So here is what Hotshot can do.

• Make 10-second videos with good movement

• Create videos that match what you ask for

• Work well with other AI tools to make very real-looking videos

• Make videos in different styles, like comics or video games

• They think AI will soon make YouTube videos

• You can try it out yourself online here.

Why this matters: It shows that small companies can make big things in AI. As it gets easier to make new AI tools, we'll see lots more options. This means people who make things can choose the best tools for what they want to do.

Let’s check some examples and see whether it is truly a Sora competitor.

😀AI Recap

1. Scientists from Disney Research and ETH Zurich have developed a system. That can generate realistic, physics-based movements for virtual characters and robots.

   

2. NES-VMC, a new AI-based method, accurately calculates the excited states of atoms and molecules, a challenge in physics and chemistry that previously delayed solar tech improvements.

3. Due to AI, water usage in Virginia, also known as "data center alley", has increased by 60%+ over the last four years.

   

4. A startup named Trunk Tools is developing an AI model. That can help construction workers find project-related information quickly.

🪄 Prompt Of The Month

Prompt For Email marketing

Copy and Paste this Prompt into your favourite chatbot

"You're an email marketing specialist. draft a compelling email to promote [product/service]. include a catchy subject line, engaging body, and clear call to action. here's the product info and target audience: [details]."

🔹AI Learning

 How to Turn Midjourney Images into AI Videos

You can create cool videos from Midjourney images using an AI video generator.

This is one of the best ways to make AI videos that are as reliable as possible right now.

Look how beautiful video he has created using different AI tools.

Here’s what to do:

1. Create Your Image:

   • Go to Midjourney (or Flux).

   • Enter a prompt, like: “A serene Japanese garden at twilight, with a red wooden bridge over a koi pond, surrounded by cherry blossom trees.”

   • Choose your favorite image and save it.

2. Turn the Image into a Video:

   • Open a text-to-video generator, like Kling, Dream Machine, or Runway Gen 3.

   • Select the AI video option.

   • Drag and drop your saved image.

   • Write a simple prompt to guide the video creation.

3. See Your Results:

   • You might get an amazing video or… something that needs a bit more work. Keep trying!

And that’s it! Enjoy creating your AI video.

♨️ What’s Hot In AI

Midjourney just did something super cool!

Now anyone can use Midjourney on their website.

And the best part? It's free to try! At least for now.

It's like they're giving out free samples of the most awesome ice cream ever.

I always used to feel that using Midjourney was like being in a secret club. You had to use Discord and make tons of images just to get the VIP pass to their website?

Well, forget all that!

I love that they've thrown open the doors and rolled out the red carpet for everyone.

Here's how you can play with it:

1. Just type what you want to see in the "image" bar. It's like telling a genie what to draw.

2. Click "Create". You get four magical pictures.

3. If you like what you see, you can ask for more or tweak them.

Once you pick an image you like, you can do some cool stuff:

• Make it change a little or a lot (they call it "Vary")

• Make it bigger and clearer (that's "Upscale")

• Or play around in the "Editor" - it's like Photoshop, but way easier

In the Editor, you can do some wild things:

1. Make your picture bigger in any direction, and Midjourney fills in the blank spots.

2. Erase part of it and ask for something new there. It's like magic erasers for art!

3. Change the shape of your picture easily.

There are a ton of ways to get better at telling Midjourney what to draw.

But here's what is saddening - you only get 25 free tries before you have to pay. It's like a free trial of the world's coolest art toy.

📰 AI News

1. In Q4 2024, Perplexity will begin offering advertising on its AI-powered search platform.

2. Meta and Spotify's CEOs argue that EU regulations on artificial intelligence hinder open-source innovation in Europe and call for simpler, unified rules.

3. In version 2.0, Ideogram released five new image styles, an iOS app, a beta API, and over 1 billion public Ideogram images.

4. The XAI team has begun rolling out Grok 2, an AI model based on real-time data from X and Flux.1, which generates AI images that are relatively unfiltered right now.

5. NVidia and Mistral have just released Mistral-NeMo-Minitron 8B, a highly accurate small language model that can run efficiently on laptops and desktop computers.