🔥 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.

• 🎙️ NotebookLM - The AI buddy for your notes can now also record podcasts

• ⚡️ Coding for all of us - How we are using Replit Agents

• 📰 AI news quickie - The HAIights from the industry

• ❌ Content is NOT king (AI ruins everything)

• 🚀 HubSpot launches an AI smash hit with Breeze

• 🤳 Superb AI (avatar) pictures with just one selfie?

→ Selection of the top posts of the last week ←

🎙️ NotebookLM - The ultimate AI buddy for your notes can now also do podcasts

A thousand tabs open, notes scattered everywhere and your desktop cluttered with documents - sounds familiar?

Google brought us the ultimate AI assistant with NotebookLM - perfect for chaotic people like us to turn chaos into a mastermind. You simply throw in your PDFs, notes and websites and everything is structured, analyzed and networked. We ❤️ it!

You can then chat with the content as if it was your personal brainstorming group. Creating summaries, FAQs, tables of contents and learning aids is just as easy as finding sources and sharing your project with others.

In short, it helps you to keep track and explore topics.

The tool itself has been around for a while, but it has recently become more widely available - and has added a new killer feature that we'll talk about in a moment.

Particularly exciting is the massive context window, which ensures high relevance and accuracy: Bye-bye, AI hallucinations! And of course, you can continuously add notes to expand the understanding of your notebook.

Possible use cases for NotebookLM

• Collect notes and use them to brainstorm and generate ideas

• Use learning material as a basis to produce quizzes

• Upload travel documents and use them to create an itinerary

• Upload manuals and use them as an (internal) database for support

• Upload code and have it explained to you

• Pull in a book and get a summary

• ... or upload your own newsletter and have it analyzed, like we did!

So much for the basics - but now it's getting really exciting!

🎧 Podcast about your notebook at the touch of a button

NotebookLM turns dry facts into an ultra-realistic podcast! With one click, you can generate a discussion between two AI hosts who present your topics in a realistic, informative and entertaining way.

On a technical level, it's an impressive blend of many things we' ve seen before - like natural voices and human emotions. But the ideas are also summarized super well and entertainingly enriched with examples and comparisons.

• It's perfect for anyone who enjoys learning or multitasking with audio

• You can adjust the playback speed - turbo or slow-mo

• And of course you can also download the podcast and enjoy it offline

Here is the audio version of this edition:

Our take: NotebookLM is cool, and the podcasts are just 🤯

We are really excited about NotebookLM! It's perfect for anyone who juggles a lot of information. Complex topics become digestible and you see connections you might otherwise have missed.

And the ultra-realistic podcast generator makes learning and understanding information so much easier and more entertaining. Try it out, just click on the "Notebook Guide" menu at the bottom right.

P.S.: We also wanted to translate the podcast with ElevenLabs, but unfortunately this is not possible due to an AI watermark...

⚡️ Coding for all of us - How we are using Replit Agents

You may not have noticed it yet, but in the last two months the world of software development has radically simplified (see also this article).

You can create complex applications directly in the browser using natural language, without having to deal with complicated setups or endless installations.

Sounds magical? Welcome to the era of Replit Agent!

Replit Agent is an AI-powered helper that helps you to turn your ideas into working software faster than ever before. The real magic is not necessarily in the current outputs - they are usable, but still have potential.

But many people are experiencing the fascination of programming for the first time, and new tools like these are changing the way we see and interact with technology.

Advantages and downsides: Where the Replit Agent shines and where it gets stuck

The Replit Agent is ideal for MVPs or proof-of-concept prototypes that run directly in the browser. In fact, anything that is browser-based is a home run for Replit.

The ease of use and accessibility via browser drastically lowers the entry barrier for software development. This makes it possible to implement and validate ideas at lightning speed - and you can also publish your app directly online!

Of course, not all that glitters is gold. Technologically, the environment is still limited and Replit Agents are also less suitable for adapting existing projects or implementing complex applications.

They also tend to get stuck in loops after a while. In addition, the current limit on the number of prompts somewhat restricts longer development sessions.

However, this has not prevented us from testing the whole thing extensively!

Best practices - what we have learned

• Describe the desired app and include links and screenshots to give the agent a good starting point. You can also refer to Github, Magic UI etc. to use code from there.

• It's best to start with a clearly defined, simple prototype that you can then expand on with additional functions. Focus on just one specific customization per request (not "Do this and that!").

• Use speech-to-text for efficient interaction. We love Flow from Wispr for this.

• ChatGPT or Claude are the perfect helpers if you/the agent get stuck somewhere.

• You can develop several apps in different browser windows at the same time. If one agent is still thinking, you can interact with the other(s).

• Use the desktop and mobile apps. You can even work on your apps on the go!

Our take: The future of Replit Agents software development

Replit Agents are more than just another AI tool, as they not only simplify software development but also open up programming to a new user group.

They are currently best suited for prototyping and simple web applications but could fundamentally change the way we develop software.

Sure, they are still in its infancy, but the outputs will very soon become much more complex and useful, supporting an extended tech stack.

Despite some limitations, the potential is enormous. So why wait? Discover what you can build yourself. For $25/month you're in!

❌ Content is NOT king (AI ruins everything)

One of the biggest marketing and sales conferences has just taken place in Boston: HubSpot's INBOUND 2024

There were tons of great presentations and news (more on that in a moment). But we wanted to start with a topic that particularly stuck in our minds.

Namely, a presentation by Marketing OG Neil Patel.

Topic: ❌ Content is NOT King

Neil talks about an exciting transformation that we have been observing for a while now.

Producing good content such as blog articles, social media posts etc. created a decent asset over time. Which in turn has generated likes, followers, website visitors, leads and sales.

But here's the problem: it's not quite like that anymore!

Over 4.6 billion content pieces are published online every day. And the whole thing continues to grow exponentially, 'thanks' to AI.

Unfortunately, most of these posts are completely useless. Both in terms of content, but above all in terms of the effect for the companies that publish it.

And even if purely AI-generated content generates traffic, people stay on such pages for less time, because it is quickly recognized if the quality is subpar.

The figures are clear: 1 minute 21 seconds for AI content, compared to over 2 ½ minutes for content produced by humans.

Is content marketing therefore completely useless? Ummm, in many cases, unfortunately the answer is Yes. But of course, we see exciting, different ways.

How can content marketing work in the AI age?

We see similar results to Neil's for ourselves and our customers: Completely AI-generated content generally doesn't perform as well.

Especially not when working with generic prompts or poorly trained GPTs.

But you can get very far with the right approach. We've been working on our own AI content production process for months - and we're making good progress (we'll be presenting something cool on this soon).

At the moment, we are mainly using AI in the following areas:

• Brainstorming and idea generation

• Generic summaries such as Quick News

• Preparation of a framework for a segment

• Repurposing content for other channels

• Checking spelling and grammar

That helps, of course, but we are currently still heavily involved everywhere. Which is a good thing, because run-of-the-mill AI newsletters are a dime a dozen. Our aim is not to simply parrot the news, but to share our experiences, views and insights with you.

So how do we bring the best of both worlds together?

We are currently rolling out the following strategy (outside the newsletter):

• Idea generation with trained AIs & AI search agents

• Always review before publishing

• Post lots of ideas on X (X does not penalize flops, every post has the same chances of getting engagement)

• Repurpose the good topics and posts on X with AI for other channels

Let's see how this evolves, we'll keep you up to date.

And: Neil's talk is really worth watching. You can do it here, from about 2:30:00.

🤳 Mega AI (avatar) pictures with just one selfie?

To round off today's issue, here's another cool thing to try out for yourself.

We've written a lot recently about training your own AI image models. For example, about yourself, your products, your brand, etc.

Now we've come across a really cool model. It allows you to create identical or similar pictures with AI using just one picture of yourself (or whoever).

And all without LoRA training etc.

You can find the model here on replicate.

It is super easy to use:

• Upload a reference image

• Enter your prompt

• That's it!

Here are a few more examples, in influencer style and as avatars:

Of course, it also has a few disadvantages compared to a trained LoRA: where the eyes are looking remains similar to the reference image, and sometimes it doesn't quite match the original.

If you really want to take consistently good pictures of yourself or other people with AI, there's no getting around LoRA training.

The good thing is that it's not difficult at all!

Let us know if you are interested in a step-by-step online training course on this topic. It could teach you how to easily create such trained image AIs (LoRAs). And not just of people! Logos, icons, products, styles and fonts are also possible.

We could also show you how to incorporate these LoRAs into your automations.

P.S. For e-commerce brands, LoRAs can be a real game changer. You can also train them on products and packaging! (Sometimes a little more error-prone, but it's getting better).

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