🇨🇳 Chinas new language model surprises the world

AI-HOY AInauts,

Happy New Year! Welcome to the latest issue of your favorite newsletter. Here's what we have in store for you today:

• 🇨🇳 Chinas new language model surprises the world

• 🗣️ Communicate better with AI

• 📚 Book tip: The next stage of evolution

Here we go!

🇨🇳 Chinas new language model surprises the world

A new version of DeepSeek's language model has recently been released and has surprised the world on several levels.

DeepSeek-V3 is a language model out of China, and it can keep up with the big players.

And there are some reasons why it surprises everyone:

• As mentioned, it can keep up with the leading models in almost all benchmarks.

• However, DeepSeek-V3 is OpenSource! Which is great, because alongside Meta with the Llama series, another strong player is reducing the dependency on closed models such as OpenAI and Anthropic.

• Apparently, V3 was trained in just two months. The costs for this are estimated at around USD 5.5 million. Absolute peanuts when you compare this with the estimated USD 500 million cost of training for Metas Llama 3.1.

• The "mixture-of-experts" architecture makes the model both fast and inexpensive.

The most important fact for us users:

• DeepSeek-V3 outperforms the other players in the market when it comes to mathematics. In other words: whenever you work with a lot of numbers, it's worth taking a closer look.

Why we think it's so exciting:

DeepSeek-V3 is a great example of how innovation can be used to develop high-performance models, even without having the infinite resources of big tech.

It also shows why sanctions (in this case, the chip export restrictions imposed by the US on China) are only partially effective...

The difference between open source and closed models is very, very small. Which in turn benefits us all.

What is not so cool is that V3 often identifies itself as ChatGPT. Which suggests that a lot of training data was generated with ChatGPT...

You can try DeepSeek-V3 right here. If an account with a Chinese makes you uncomfortable, you can also use one of our favorite tools, OpenRouter, and pit it against the other models.

🗣️ Communicating better with AI

Let's move on to a topic that we thought about whether we wanted to cover it at all.

Because it's so simple. So trivial. Nothing special, really. A little reminder, if you like.

However, it helped us so much recently that it might help you too. It also describes one of our AI mindsets, which we apply as often as possible. Specifically, it's about communicating better - thanks to AI.

The situation: We recently had some important matters to settle in our company before the end of the year, and had to deal with tax consultants, lawyers, team, etc.

Especially in delicate situations, it is extremely important to communicate well and clearly. If emotions are also present, it becomes even more difficult.

The great thing is that today we can simply use AI to improve our texts. Especially in important situations, no email goes out without first being checked by AI.

We use AI to make our emails better, to simulate different reader groups, etc. Just post the text in your favorite AI, give a little context and ask a few smart questions.

Here are a few questions we would like to ask:

• What might be confusing / misunderstood in the text?

• Where should I add more context?

• What questions do you have after reading this text?

• What emotion does the text trigger?

• How could I formulate it better to achieve my goal X??

It's always crazy to see how AI improves texts in seconds. It makes things clearer and saves a lot of trouble, especially with sensitive topics.

Whether you use ChatGPT or Claude doesn't really matter. Of course, it helps if you have defined your writing style beforehand and the AI knows a little more about your background.

But even without much context, the small detour via the chat with the right questions can lead to significantly better results in communication.

As already mentioned: very simple, nothing special, but it helps us a lot. Maybe it will help you too.

P.S.: Here's a GPT that we built for ourselves (... and if you just want to check spelling, you'll find it here). It is in German, but works in any language. Have fun!

📚 Book tip: The next stage of evolution

Let's end the first issue of the year with something light-hearted.

The thousands of Out of Office autoresponders tell us that many people are still on vacation this week.

If you're reading the mail and are looking for a new book: We're reading this one right now!

He has an interesting outlook and vision for the next few years. We also like that it is rather positive and not just "doom and gloom" Terminator apocalypse.

If you don't feel like reading, check out this cool 2-minute robot video. Just for fun. You won't regret it. 😉 

So much for the first edition of 2025, and we are very grateful to have you with us. It's going to be an absolutely brilliant year, and we already have a lot of exciting topics in mind.

Here's to a great 2025 – may AI be with us!

Reto & Fabian from the AInauts

P.S.: Follow us on social media - that motivates us to keep going 😁!
Twitter, 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.