🔥 Weekly AI news: Did you miss it?!

👨‍🚀 The most important AI updates at a glance

AI-HOI, AInauts!

Maybe you didn't catch all the news, tools, and hacks about AI last week, or maybe you've only recently joined us. Either way, here's our recap with all the headlines from the newsletter - just one click away!

Click the links to jump right to the article - or read our picks below.

• 😲 How to build an AI phone agent in 5 minutes, without coding

• 🥇 Why Nvidia is the most important company of our time

• 🙈 GPT-4 is the dumbest model …

• 🤖 Robots have their ChatGPT moment

• 💸 $190’000 dollars per month with a simple AI wrapper app

• 🤖 This is how AI presents your PowerPoint for you

• ⚙️ Our favorite AI tool for perfect audio recordings

→ Selection of the top posts of the last week ←

😲 How to build an AI phone agent in 5 minutes, without coding

Phone communication is likely to change soon … Thanks to text-to-speech (TTS) and fast, intelligent voice models in combination with the corresponding infrastructure (groq, anyone?), completely new possibilities are opening up:

• You can outsource time-consuming or annoying phone calls to your virtual assistants, and they can hold the line for you.

• Or you can use such assistants to handle incoming calls (e.g., in customer service, office, consulting, …) and let the AI try to solve them.

One provider that promises a solution for this is RetellAI.com. It is a platform for the creation of voice-based AI applications and designed to enable realistic and human-like conversations with a low latency (= response time).

You can see how this works in the diagram here:

There are three options for using the platform:

• Dashboard: Build an MVP in 5 minutes without coding, and have an AI phone agent with a Twilio number - for all those who can only click and don't code

• Retell LLM: A powerful API for seamless integration of voice AI - for anyone who is not afraid of APIs and speaks JavaScript

• Custom LLM: Implement your desired LLM with telephony services such as Vonage or SignalWire - for everyone who can also communicate with 0 and 1

We can't code, so the only option we can use is the Dashboard ... and it promises that you can set up such an agent in 5 minutes. Really?

The live test: Building a telephone agent in 5 minutes?

Naturally, we wanted to find out more, so we put it to the test. It's super easy:

1. Log in with your existing Google account (or create a new one)

2. Choose a voice from ElevenLabs

3. Customize your prompt, change settings if needed

4. Test with a web call (10-minute test is free)

We were quite pleased with the result, you should definitely give it a try!

🥇 Why Nvidia is the most important company of our time

So, from the practical test to the man who made it all possible …

Sometimes there are these videos where so many lightbulbs go off in our heads. Like this discussion with NVIDIA CEO Jensen Huang! He is super likeable and easy to follow as he talks about what we can expect in the future.

Let's summarize the highlights and go a bit deeper.

Fact: Without Nvidia, all AI cloud providers would have a real problem …

Nvidia is the biggest player in the AI industry and has achieved true technological leaps in the last 10 years. The share price has even exploded by over 1800%.

Sure, because these chips are currently driving practically the entire GenAI infrastructure of Google Cloud, Microsoft Azure (which also powers the Open AI models), Meta, Amazon AWS and so forth.

Of course, the big players are trying to build their own chips or look for alternatives. But that won't happen overnight.

Meta's Chief AI, Yann LeCun, has just confirmed that Meta has invested $30 billion (!) in Nvidia GPUs to train models like the brand-new Llama 3. For comparison: that's more than the Apollo moon mission cost, he says (… but beware, that's marketing-speak and not inflation-adjusted - still sounds good, doesn't it?).

Everyone wants chips for data centers, but also for autonomous machines

Nvidia has already reduced the cost of computing and deep learning by a factor of millions (yes, you read that right …). And in the next 10 years, computing power is set to skyrocket a million-fold 🤯 …

But that's not all. CEO Jensen Huang has the declared aim of reducing costs even further and pushing them as close to zero as possible! The reason? When costs fall, penetration increases and AI is omnipresent.

Behind these efforts are the latest chips. These are real heavyweights, with over 30 kilos of high-tech, consisting of 35,000 individual parts. This computing power practically replaces an entire data center!

Nvidia is already skimming off as much of the market as it can, accounting for over 70% of all AI chips sold. But this is just the beginning, as the company has big plans for the future and is also making progress in the field of autonomous machines. Yes, like Terminators humanoid androids (see post below).

That's why language models are not yet continuously improving - but this will change soon!

Technological advances mean that training and inference (i.e. chatting with the language model) will no longer have to be separated in the future.

At the moment, the model is trained first, and we then interact and chat with the frozen model, so to speak. In the future, however, AI will learn, question itself and improve around the clock.

The aim is for AI to become an exponential superhero in all areas. And that is something that we humans find quite difficult to imagine!

In the future, content will no longer be retrieved, but generated.

But the future belongs to generative computing. Static content (as we mostly use today) will be a thing of the past. Instead, much of what we consume will be generated directly “on the fly”.

Of course, this change requires a radical overhaul of the computer infrastructure - a digital spring-cleaning of epic proportions. But don't worry, Nvidia has already rolled up its sleeves and is ready for the big transformation!

What's more, on the network side technologies such as 6G are already in the making, so we'll never again be faced with a screen that says “Buffering …”.

So, that was the Nvidia story. Keep up the good work, Nvidia - and keep an eye on the marginal costs!

💸 $190’000 dollars per month with a simple AI wrapper app!

Let's start with a story that we think is absolutely great.

Sure, the headline just sounds like clickbait. But if you look at the concept behind it, it becomes evident that we are living in the most exciting and crazy times ever!

So, we came across the following post on X:

In a nutshell: A solopreneur has built a simple little app in which you can upload screenshots of your dating conversations from Tinder, Bumble, WhatsApp etc. - and then get suggestions for possible answers, making you a flirting genius.

We don't want to get into a discussion about whether something like this is good or bad. The fact is: the creator behind the app RizzGPT makes around $190’000 dollars in revenue per month!

Why is this so important?

The exciting thing behind the story is that thanks to AI, almost anyone has the opportunity to launch such apps. You just need a good idea!

OpenAIs GPT-model powers this app in a very simple way. You upload an image, GPT-Vision understands it and is tasked with generating a clever answer based on it and then feeding it back into the app.

Alternatively, everyone can build and use such an app in 2 minutes with ChatGPT Plus - just build a GPT! Or you could also create a simple OpenAI Assistant and control it with Zapier.

We initially thought that the foundation models and the companies behind them would make all other software wrappers useless.

But we can see now that the exact opposite is the case!

Just because everyone seems to be talking about AI and thanks to our flood of emails, you might think that everyone knows what GPTs etc. are. But that is not the reality!

Examples like this one are impressive proof.

The big disadvantage: Because it is so easy to implement, there are quickly many imitators, as a search in the App Store proves:

How can you build an app like this?

The brilliant thing is that even non-tech-savvy folks can implement such things, thanks to no-code apps and automation.

Sure, building your own app is always a bit scary.

But with tools such as Bubble.io and an OpenAI access, you can build something very solid - even if the learning curve with Bubble should not be underestimated. But you could also cover it super simply via a Zapier chatbot.

This is precisely why we consider the ability to combine AI, automation and no-code tools to be one of the most important for the coming years.

Thanks to the large AI models that are accessible to everyone, each of us has the opportunity to build impressive things. Even if you're not an absolute techie.

🤖 This is how AI presents your PowerPoint for you

AI is particularly useful when it does the things for us we don’t want to do.

For many people, especially in the field of coaching and sales, it is tiresome to give the same presentations and answer the same questions over and over again.

Wouldn't it be cool if AI could do this instead?

Spoiler alert: In principle, this is already possible! Even with you as the virtual presenter.

It is still a bit technical at the moment, but there are some apps that make it increasingly simple.

Step 1: Google Slides integration of Living AI

With Living AI, you can add simple avatars to your Google presentation.

Just install the plugin. Select an avatar. Enter your text. Choose voice and language. You’re done.

You can then export your presentation as a video and even create a talking avatar based on your image:

The advantage of Living AI is that it is quite inexpensive. Unfortunately, the avatars are still a bit mechanical and Google doesn't allow you to easily export the whole presentation as a coherent video.

Which brings us to the next stage ...

Step 3 - Automate PowerPoints with D-ID

One tool that we have written about several times is D-ID. It makes it very easy to create avatar videos.

D-ID also has its own PowerPoint plugin, you can simply download it via the extensions feature.

You can then integrate your D-ID avatars directly into PowerPoint. The process is the same as above.

You can also watch an instruction video here

The big advantage: with PowerPoint we have the option of exporting a continuous video! D-ID is slightly more expensive, but also significantly better.

This brings us one step closer to making ourselves completely redundant.

By the way, Canva.com also has a direct integration of D-ID, if you work with presentations there.

Step 3 - Pose live questions with streaming avatars

The last and final stage is that you create an avatar of yourself, with your voice and armed with your knowledge.

There is also something like this available in a first version of D-ID.

Here you can chat with Alice for free and also send her voice messages. She will then reply to you by voice.

Setting up a chat like this is simple with D-ID. What's more, everyone realizes that it's an avatar. But it's still pretty cool.

A better, more realistic job is currently being done here by HeyGen with the streaming avatar feature.

Here the avatar is more complex to generate, because it is not just an animated photo. On the other hand, you have real backgrounds and everything looks very natural.

Of course, this is also still in beta. But it's quite exciting, especially for chats with potential new customers or for support requests. You can tell that it's AI, but as long as the answers are good, nobody should be bothered by it. And perhaps the other person will even be impressed by how progressive the company is.

By the way, we are also currently building a streaming avatar chatbot with HeyGen - and will post an update here as soon as it is live and working.

See you soon with a fresh round of news, hacks and insights!

Your AInauts, Fabian & Reto

Follow us on Twitter & LinkedIn!

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.