Digdeep Podcast: Goodbye 2024, Hello 2025!

The first 2025 episode of #digdeep reflects on twelve exciting topics and interviews in 2024 that inspired both podcast creators and listeners.

Whether in mechanical engineering, where even medium-sized companies can get ahead with attractive digital concepts, or in medicine and healthcare, where new start-ups are now making research data usable in various areas: Only with the digital transformation, which must have arrived in Germany long ago, the two podcasters believe, will it become apparent what economic leaps are possible and how well data can help to steer and regulate issues in society. “They are quantitative accompanying research for social development,” says MCML PI Frauke Kreuter.

In evidence-based politics and data analysis, research is also continuously focused on bringing about a change of mindset in the digital sphere. While public opinion is important for election forecasts in election polls, reporting is also important in election campaigns. In the coming years, it will be exciting to see what role science and the press play in the political spectrum. Broadcasters and the press in general are working intensively on AI. “AI is becoming more and more prevalent, helping, supporting and taking away routine work, but it should be seen as a very powerful tool,” says Horn.

Artificial intelligence, which is determined by hype cycles and challenges, will enter the next productive phase in 2025. ChatGPT and generative AI show various parallels and partly shifted hype cycles. Depending on the task for which generative language models are being used. “Data analysis work can be extremely accelerated if parts of code are generated,” says MCML PI Frauke Kreuter. “It's fascinating to see what's happening right now. The experts are getting faster, the beginners are getting a jump start and everyone who doesn't use it is simply left behind in terms of speed.” When it comes to alignment, Kreuter is also concerned with how to get language models to do what you want or don't want them to do. “Guardrails and newer models are so powerful that you can hardly keep up with them in research,” says Kreuter, ”other models may be needed here in the future.”

For Christof Horn, his personal hype cycle is multi-layered. ChatGPT not only helps in the daily search for topics and delivers precise results, but also in understanding topics and acquiring world knowledge, for example with regard to language or technology.

In addition to the euphoria and democratization, the limits of the models, which cannot do any real reasoning, are now becoming clearly visible: “We are currently experiencing a crisis for everyone. The idea that even more training data will help does not work and there is a saturation of the models, because even more parameters do not lead anywhere. Moreover, the training data is not getting any better, but is contaminated with results that it has produced itself,” says Horn.

The AI topics on the hype cycle continue: “What we will see next year is an explosion in the integration of generative AI into data production,” says MCML PI Frauke Kreuter. The chatbot as an interviewer and the automatic evaluation of open answers. Both will become decisive criteria for the future of the customer journey.

To escape the perpetual feeling of inevitable chatbot hell in 2025, real people will gain in value. Engineering and other tools will nevertheless bring AI further among us and the discussion about language models will no longer be held. Robotaxis, lawn robots and digital agents that you can talk to will remain normal for people. Also because they are entertaining. When it comes to influencing, every medium offers a new platform.

In autonomous driving, platforms such as “Waymo”, which work with training data and further optimize it, will have a stronger influence, as will innovative, promising start-ups, such as those in Germany in sensor technology or international ones that automate delivery and factory traffic.

In their podcast, MCML PI Frauke Kreuter and Christof Horn 2024 also dealt with the digitalization of shoes and education platforms, were at the “herCAREER” career fair and were the first German-language podcast to host ChatGPT for the second time.

With or without a language model, but not yet generated by it: #digdeep will still exist in 2025. In the upcoming January episode with “BrandEins”, they visit the “work awesome” conference in Berlin with the CEO of “Burger King” and “Metaplan”.

Just listen in (podcast in German).