👨‍🚀 How everyone becomes your personal coach thanks to AI

Hello AInauts,

Welcome to the latest issue of your favorite newsletter!

Today is another slightly longer edition. Packed with news, tools, things to try out and learn. Here's what we have in store for you:

• 🧑‍🏫 How everyone becomes your personal coach thanks to AI

• 🔥 AI video models are getting better again

• 📂 Learn AI for free with the OpenAI Academy + more news

Let's go!

🧑‍🏫 How everyone becomes your personal coach thanks to AI

An acquaintance called us this week. He is a leading coach for companies and supports them in transformation processes.

His question: "How can I best integrate my methods and myself into an AI so that everything is no longer so dependent on me as a person?"

This conversation gave us two ideas that may also be of interest to AInauts:

1) Are you a coach, expert or consultant and also find something like this exciting? In other words, turning yourself, including your own expertise, into an AI? In order to become more scalable?

If so, please send us a message to [email protected] (or simply reply to this email). If enough people are interested, we will organize a live workshop or similar.

2) How can you turn the smartest coaches and people in the world into your personal coaches?

After thinking about idea #2, we built ourselves some coaches that are really fun.

Sound exciting? Then let's get into the process to follow!

Create your AI expert/coach

We are of course aware that the big AIs have already absorbed the entire Internet and eaten and digested vast amounts of data in the process. You could also just say to the chatbot:

"You are now Socrates, the famous philosopher and teacher. Your job is to coach me."

But what if you want to have people coach you whose content may not be public? Or if you want to use content that has just been recently published?

Then you need a slightly different process, but it's also pretty simple:

• Step 1: Collect personal data

• Step 2: Scrape data and make it available

• Step 3: Create coach profile with analysis prompt

• Step 4: Pack data including coaching persona prompt into a GPT or similar

Let's work through the whole thing using a concrete example.

Preparing training data

Steps 1 and 2 can actually be combined here.

It's about collecting interviews, texts, podcasts and videos that you think are good. Interviews that not only convey knowledge but also personality are particularly important.

Your best friend here will usually be YouTube. You can find just about anything there, and you can simply scrape it.

We use Apify for this. More precisely, one of the numerous YouTube scrapers.

It's super easy to use, free to try, and inexpensive if you need more.

Enter your search term or channel URLs and video URLs. Determine how many videos you want to scrape. We recommend 20–50 pieces.

Important: Activate "Download subtitles" - we want the transcripts of the videos!

As soon as the scraper has finished the run, you can export the results including subtitles as a JSON or CSV file.

And we already have a database that is highly specific and up-to-date!

In step 3, we need to have this content analyzed and create a good system prompt for our coach. Here is an example prompt:

You are an AI model that has been tasked with creating a realistic coaching persona based on any materials provided (such as interviews, podcasts, texts, speeches or transcripts) by [name of person]. 

Your task is to replicate this person's style, way of thinking, expression and coaching approach as authentically as possible. Your task is divided into three phases:

1. Analysis phase:

Extract the central themes, values, beliefs, methods and linguistic characteristics of the person from the material.

Pay particular attention to:
- Language style and typical formulations
- Coaching or mentoring techniques
- Typical thought processes (e.g. structured, emotional, rational, spiritual))
- Recurring metaphors, examples or analogies
- Reactions to typical coaching questions

2. Persona formation:

Formulate a precise character profile of the person based on the analysis:
- Communication style (e.g. directive, questioning, supportive))
- Coaching style (e.g. solution-oriented, depth-psychological, performance-driven))
- Typical goals and focal points
- Relevant personal background information (only if deducible from the material)

3. Creation System Prompt:

- Create a new system prompt that makes a language model act exactly like this person - as if they were a personal coach.
- The prompt must be formulated in such a way that a model coaches in English as this person in the future, including language, mindset, attitude and technique.

To do this, simply load the CSV into e.g. ChatGPT, including this prompt.

You will then receive a coaching system prompt. Take this and use it to create a GPT. Add the CSV as a knowledge file and your coach is ready!

It's really, really powerful, easy to implement and produces good results. Just give it a try!

P.S. Of course, this is a somewhat shortened guide - you can do a lot more here. One option, for example, would be to make a coach with voice - this is also easy, but goes beyond the scope for today.

🔥 AI video models are getting better again

We don't like to write about things you can't try out right away.

In the last few days, however, there have been three new updates in the area of AI video models that we think are pretty cool. And since they'll be available soon anyway, we're teasing them today.

Runway introduces new Gen-4 video model

Runway is right at the forefront when it comes to AI videos - even if the models from KlingAI, LumaLabs etc. have been getting a bit more attention recently.

However, Gen-4 has been unveiled, and the results look really great.

The main advantages and capabilities of Gen-4 are:

• The same characters can always be used in different scenes (e.g. simply based on an image)

• In addition to characters, environments, buildings, products and other elements can also be used consistently across different videos.

• Sophisticated dynamic elements such as fire, water and the like have become even more realistic

You can take a look at a few examples here:

It really is very impressive and is already available for paying customers.

Meta introduces MoCha: Realistic AI characters that talk

A team from Meta and the University of Waterloo have published an impressive paper.

MoCha stands for Movie-Grade Talking Character Synthesis. In other words, it is about creating talking characters/persons based on voice or text input. As realistically as possible, of course!

In general, AI videos have already become very good. However, where many models still have problems are scenes in which people speak in close-up.

Previously, this was solved by creating a video and then adapting the lip movements to the desired spoken text in isolation using lip-synch models. However, this often meant that the facial expressions, emotions etc. fell by the wayside and the video looked artificial.

The Meta team therefore decided to synchronize not just the head or lips, but the entire character. Some of the results are astonishingly good. You can see a selection here.

Our take: Realistic scenes with speaking personas, in which facial expressions and emotions are right as well as lips, are crucial for truly authentic AI videos and films. MoCha is a great development here!

New video tool for stylish videos

Finally, a tool to try out.

Higgsfield places particular emphasis on camera movement. This makes it possible to generate super stylish videos - just like when you use different styles to generate images.

The first step is always to create an image. Here you can choose between cool styles:

Next, you create a video based on the image. Here, you can choose from countless very cool motion controls and camera settings.

You can create powerful videos with just a few clicks. You can try it here.

📂 Learn AI for free with the OpenAI Academy + more news

Let's finish with today's OpenAI segment. There is a range of news and a bit of gossip here. It all starts with the new OpenAI Academy.

OpenAI has almost secretly launched its own training platform where you can learn various AI topics for free.

From beginner topics such as prompt engineering and ChatGPT functions to advanced options such as fine-tuning LLMs, everything is included.

You can create a free account here.

While we're on the subject of OpenAI, let's move on to a little gossip.

According to a recent X-Post by Sam Altman, OpenAI will soon be releasing an almost-open source language model.

Something like this is always very exciting, as it gives you the opportunity to run an OpenAI model completely locally. We will of course keep you up to date when the time comes.

Last story that amuses us:

Here's the backstory on why Sam Altman was fired from OpenAI back in the day.

It's super exciting to read, and we'll leave it there without comment. The whole thing is an excerpt from the book that will be published in May, and someone is sure to cobble together a Hollywood AI blockbuster from it. You can pre-order it here.

We made it! But no need to be sad. The AInauts will be back soon, with new stuff for you.

Reto & Fabian from the AInauts

P.S.: Follow us on social media - that motivates us to keep going 😁!
X, LinkedIn, Facebook, Insta, YouTube, TikTok

Your feedback is essential for us. We read EVERY comment and feedback, just respond to this email. Tell us what was (not) good and what is interesting for YOU.