3.3 | 5 current findings on how LLMs work
Introduction
How do large language models really work?
Even though we work with them every day, we don’t fully understand them. Even leading AI experts call the models a „black box.“ But new research—for example, from Anthropic—is providing initial insights.
In this module, you’ll get an overview of what is known today about how LLMs work—and what is not (yet) known.
The video below summarizes the most exciting findings.
Bitte wählen Sie eine Option
Conclusion
You see: AI is more than just a reaction tool. It thinks—differently than we do—but no less complexly.
When you understand how LLMs work, you can use them more effectively. With better prompts, with clearer expectations—and with a critical eye on every answer.
Stay tuned—this field is evolving rapidly.
An Experiment
Copy the original article from Anthropic—and give it to Grok (which is a Large Language Model (LLM) available to you on the xpand AI platform) with the following prompt:
„Take a look at the attached article; one of your ‚colleagues‘ was scientifically studied there. What do you think of it—and can you summarize the findings with your comments and translate them into English for me?“
Tip
It’s amazing how nuanced Grok’s response is—almost self-reflective.
Feel free to try the same thing in other LLMs—you will get very different results and see how differently the language models „think.“
This is how you connect research and practice—and gain your own insights into the way modern AI thinks.
Your Takeaway
- Modern LLMs are more complex than we often think—their inner workings remain a „black box“ to some extent.
- New research (e.g., from Anthropic) provides initial insights into the „thinking“ of AI.
- Understanding how AI works helps you write better prompts and critically evaluate results.
- Different AI models „think“ differently—experiment with the same prompt in different LLMs.