🦾 Two new language models that you absolutely need to know about

An ice-cold good morning, AInauts!

Welcome to the latest issue of your favorite newsletter! New week, new language models. We take a look at what's important for you. And we also have a cool way for you to create and edit images live with AI.

The agenda for today looks like this:

• 🦾 Two new language models you absolutely need to know about

• 🔥 Create 3D images, mockups and more in real time with AI

• 📲 Checklist for better AI social media posts

• 🤯 When an AI agent controls an AI agent (video)

Here we go!

🐳 New LLMs again: which one should I use?

There was a lot going on again this week in terms of language models.

As always, we'll give you a brief overview and help you decide whether you should switch immediately. 😉

The main event was of course the presentation of Grok-3, the new language model from xAI. However, there was also an update of a lesser known but exciting language model (because uncensored).

Let's start with Grok-3:

Elon being Elon is of course very convinced of his company's model. However, it does indeed look as if it is currently pretty much the "smartest language model in the world".

Here are the most important facts at a glance:

• Grok-3 comes in different versions: Mini and also Reasoning. Roll-out via the app, X-Premium users get access first.

• There is a "Big Brain" feature that should produce even smarter answers than OpenAI's o3.

• AND, of course, there is also a research agent. It is called Deep Search without the "Re" 😉.

• A voice feature, similar to ChatGPT and Gemini, will also be coming soon.

In short: Grok-3 can do everything that the big models like OpenAI and others can do. And perhaps even a little better (at least until the next OpenAI release).

However, we like Grok and use the model again and again via OpenRouter. In some places it seems less restricted than other models and is often a bit more fun.

Do you have to quit ChatGPT or Claude now? No, but it never hurts to think outside the box and experiment.

If you want to dive even deeper, here is an exciting thread about it:

New version of an uncensored language model now thinks for itself

Speaking of restricted models. Another update that wasn't even mentioned in the major media comes from Nous Research.

The DeepHermes-3 preview was presented here a few days ago. It is a further development of our own Hermes-3 model, which now also has reasoning capabilities and can therefore think.

Why is this exciting?

Nous Research and Hermes are largely uncensored LLMs. In other words: the language model does not refuse to answer you for moral reasons, etc.

Hermes is particularly popular in the field of role-playing games, AI Girlfriend / Boyfriend etc.

We know that this is exciting for some readers - you can find more information here.

🔥 Create 3D images, mockups and more in real time with AI

After all the text, let's move on to the images.

If you've been reading us for a while, you'll know that we're big fans of the Krea AI tool. We have already presented it in relation to videos.

Today we want to talk about a feature that we are currently having a lot of fun with.

It's called Realtime and combines objects, inputs and prompts to create a new AI image. All in real time.

We use the Krea Realtime Canvas for this purpose:

Essentially, it works like this: you have a canvas on the left in which you can insert images and objects. At the bottom, you have the option of adding a prompt.

An image of all elements is then created on the right.

The exciting thing is that we can now move objects and adjust prompts.

Let's take a practical use case: We want to furnish a living room.

• To do this, we take a background

• Create an image of a couch that we like

• Turn the couch into a 3D model with the help of Krea

• And set up the whole thing nicely with a prompt

Here is our empty living room. First, we create a sofa that we like. We add that and turn the image into a 3D model.

Then we add a few shapes where we would like to have plants and maybe some kind of yellow table. Then another prompt: "Scandinavian living room".

Here is the result:

Now we can move the objects live in real time, rotate them etc. and always get a newly furnished living room.

This method gives you enormous control options for generating your AI images.

Krea can do much more and has a generous free quota.

For all Premium members, we will also be publishing a detailed tool report on Krea on Friday, including a video tutorial.

📲 Checklist for better AI social media posts

As you may know, we've been playing around with social media and AI automation for some time now.

But when we use AI to generate social posts etc., we realize something that doesn't really make sense ... We'll get to that in a moment.

For our prompts, we’ve found a checklist that prompts the AI to write the content in the desired tonality right away.

You can also use the following prompt module directly in the tool of your choice:

Please check the created content according to the following checklist. Make sure that all points are fulfilled:

Do people stick with it?
- Is there a reason to read to the end?

Are they getting curious?
- Do I arouse curiosity without giving everything away straight away??
- Is someone thinking, "Wait ... what's that?"

Hook - Does it stop the scroll?
- Is the first line/title so strong that you can't simply scroll on??
- Is the hook short, clear, suitable for my target group?
- Are emotions generated?

Is it real?
- Am I open and human?
- Do I build trust?
- It looks authentic instead of artificial?

Something comes back?
- I ask for opinion/feedback?
- I invite you to discuss or exchange ideas?

You can transfer this module directly into your prompt with which you want to generate content. Especially with the clever reasoning models, you should get a good result.

And it does!

But now what we have just noticed - which actually makes no sense:

If we don't use the module directly in the initial prompt, the AI initially leaves the content unedited. If we then ask it to check and optimize the text using the checklist, the results will be significantly better!

It may therefore be worth adding an extra step to the process. The good thing is: thanks to automation, you can implement such prompt chaining very quickly and easily …

🤯 When an AI agent controls an AI agent (video)

Finally, you want to watch the following video. Here, OpenAI Operator was used to control a Replit Agent and build an app autonomously.

Crazy! ... but this is just the beginning.

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.