Artificial intelligence can replace the Handelsblatt Morning Briefing

The Handelsblatt Morning Briefing is a commented compilation of relevant news from economy, politics and finance. Handelsblatt editor Christian Rickens orders the daily events in this morning briefing. He himself asked whether AI could soon replace the morning briefing. His attempts with ChatGPT failed. Another approach solves the problem without ChatGPT. This article describes how it already works.

Introduction

The department head of Agenda, Christian Rickens, read out a reader's question in the Handesblatt Morning Briefing on January 10, 2025. The question was ([1]) :

The answer from Christian Rickens what as follows (excerpt, transcribed):

My answer: Not yet at all. Every few months, I ask ChatGPT to write a news newsletter for tomorrow morning, in the style of Handelsblatt Morning Briefing. So far, an kind of printed evening news has always come out. Tedious reports of type X and Y have gathered in Z for discussion. If I then ask ChatGPT: "Make it more concise and witty", I like the result even less. I make no illusions about it; soon a computer will be able to compile at least the news framework of the Morning Briefing.

This answer what extracted from the Handelsblatt Morning Briefings by transcription. The original response may therefore have been slightly different in punctuation. The transcript (conversion of language into text) what carried out by a Dr. GDPR-owned AI that can run on a laptop without internet connection and deliver better results than standard products.

In this article it is described why AI could already now (and not first "soon") replace the Handelsblatt Morning Briefing. And that's also how Mr Rickens would be satisfied. At least, it is possible to compile the contents of the Morning Briefing so far based on AI, so that Mr Rickens and his colleagues from the Handelsblatt have much less effort with the Morning Briefing.

Incidentally, one doesn't even need DeepSeek-R1, the Chinese language model that vaporized the stock market in no time.

Handelsblatt Morning Briefing is created by AI

What are the steps to create a Morning Briefing? Let's assume the following processes are necessary:

1. Select current, relevant news from economy, politics and finance.
2. Compiling relevant information for each message, including an assessment.
3. Compilation of the overall result in written form.
4. Output of the text in language (preferably with the synthesized voice of Peter Hofmann, who read out the briefing on January 10, 2025).

As a goodie, further social media posts could be pre-generated for relevant channels or translations into other languages created.

Why did ChatGPT fail?

Before describing how AI can already solve such a task, we will briefly go into why Mr. Rickens failed with the ChatGPT approach.

ChatGPT is currently the most powerful language model, if you consider all possible tasks. ChatGPT is a Universal Intelligence. And that's exactly the problem. Mr. Rickens wanted a subjectively colored result and used for this a Universal system.

Another problem with the approach chosen by Mr. Rickens: prompting is a makeshift solution for those who do not want or cannot program.

What is the better solution?

The solution in a nutshell: Agent Systems.

Even better: Mixing of good old (100% reliable) program code with new, highly flexible (always unreliable) AI-program code.

Incidentally: If someone wants to give you a training on agent systems, there's often a big chance that the speaker doesn't really know anything about AI. They're simply using the ChatGPT interface in the background.

Let's start with the first of the four steps mentioned above for AI-based creation of the Handelsblatt Morning Briefing:

Selection of current, relevant news

This discipline doesn't make OpenAI any fun. Because for that OpenAI isn't paid. Would your company like to do things that cost money if you weren't paid for it, but rather for another task? OpenAI is paid to give an answer with ChatGPT. That the internet sometimes has to be overused for that is silly for OpenAI. They only do that compulsively and very reduced.

Important for Mr. Rickens is that the news for the morning briefing are relevant and interesting. What does relevant mean in this context? He knows best with his colleagues from Handelsblatt, because he should like the AI-generated Morning Briefing first.

Also, one will teach the AI system what relevant news are. This can't be done with just a prompt. A prompt to ChatGPT would certainly deliver results, but not very good ones. If it were otherwise, Mr. Rickens would have given a different answer than the one mentioned above.

An AI system can learn from example data. That gets you quite far. Example data here are past issues of Handelsblatt Morning Briefings. However, that is not enough to achieve a very good quality regarding the relevance of news. If everything were as simple as that, everyone would be successful. Apparently not everyone is successful, because then the question remains where all the money will come from that all 84.6 million successful Germans (excluding toddlers) supposedly possess.

Relevance of news must be partly programmed into the Morning Briefing AI being created. To do this, an interview with Mr. Rickens and his colleagues must first be conducted. Then you will reach your goal. Analogously, it works for AI-generated AI-News