OpenAI 01-Preview — SECRETS 99% of People Don’t Know

How to get the most out of 01-preview

I’ve been playing around with 01-preview since it came out.

I’m loving it!

I’m even teaching it in my new AI Growth Hacking course.

I’m excited to share some key insights about how to get the most out of this.

Most people have no clue how 01-preview works.

First of all, it is not just a “thinking” model.

You need to understand a little bit about how it works before you can take full advantage of it.

If you don’t have a paid Medium account you can read for free here.

👉 Sign up to our free 5-Day email course to grow 🚀 and earn💲in the AI age

How does OpenAI 01-preview work?

01-preview is not really a new model.

It combines other models and a “system prompt” that tells it to iterate several times before a response comes out.

All other models work by providing the first response the model comes up with.

01-preview is designed to plan and experiment before a final answer comes out.

An example will help.

Imagine you tell GPT-4o to write a coherent paragraph that is exactly 80 words long, and has the word “tomato” as the 4th word, the 19th word, and the 72nd word.

GPT-4o (and all other models) will fail at this task because it is too difficult to just spit out the first answer that comes to mind.

This type of question needs experimentation.

Think if you were given the same task.

You need to “play around” with this task to try to fit the word “tomato” in those spots in a way that makes sense.

You can’t just start writing and see what happens.

You will realize that you need to change some sentences around, and words around, in order to fit the word “tomato” in there.

Also, when you get near 80 words, you need to plan how to stop at exactly that number. You might wish to go back and delete an unnecessary word from the first sentence for example.

The reason 01-preview can do this type of thing is the way it “thinks”.

It will first break the problem down and say something like, “Come up with a plan to solve this problem”.

Then it will write an approximate first guess (probably using GPT-4o)

Then it will say to itself, “Re-read the question and see if you can make any tweaks or adjustments”.

Then it will say, “do a double-check to see if your response is perfect. If it is, display it, if it is not, keep tweaking.”

Then it will say, “Repeat this process until your answer is 100% perfect. Always remember to double-check your final answer before displaying it”.

For example, the first sentence of the first response might be this.

“Sandy picked a red tomato from her garden.”

Then 01-preview would change it to, “Sandy picked a tomato from her garden”.

This way, it would successfully move the word tomato from the 5th word to the 4th word.

It would keep making tweaks by having an internal conversation with itself.

How to get the most out of 01-preview?

Now that you have an idea about how 01-preview “thinks”, you can start to understand how to get the most out of it.

You need to divide your own questions into ones that require “thinking” and ones that don’t.

Many questions don’t require “thinking” from the models.

For example, if you tell it to write you a funny story about a girl named Sandy who has a tomato garden, then you don’t need to use 01-preview.

Why not?

Because there are few constraints.

There are many ways to do this. It is essentially open-ended.

The story doesn’t need to have a certain length.

The model can just start writing, throw in a joke or 2, and be done with it.

It won’t need to go back to the first sentence and count the number of words or anything.

The point is this:

If you are asking the model for something specific that would be difficult to do in one try without experimenting, then you should use 01-preview.

If you are asking for something open-ended, then use the other models.

You need to use 01-preview sparingly because you are only given a limited number of queries.

It will always be more limited than other models because it uses far more resources than other models.

The good news is that 01-preview will make far fewer mistakes than other models.

Also, it will be able to answer questions that previous models failed at.

Now is a better time than ever to learn how to leverage AI to grow your business and earn more money.

I am teaching how to use this in my AI Growth Hacking course.

I’ll also be incorporating this model to make the custom AI Agents and chatbots I build even more accurate.

This will make AI agent builders like Stammer even more powerful.