A contribution by Sophie Hundertmark.

Sophie Hundertmark is an expert in the practical use of artificial intelligence with a focus on chatbots, AI strategies and responsible technology integration. She is a researcher and lecturer at the Lucerne University of Applied Sciences and Arts and is currently writing her dissertation in the field of Conversational AI at the University of Fribourg. As a consultant, she supports companies, administrations and educational institutions in the introduction of effective AI solutions. Find out more about Sophie Hundertmark on Linkedin.

I recently shared a post about how much I’ve been happy with Gemini lately, often more than ChatGPT. The reason? To me, it feels like Gemini can remember “more context”. I looked into this in more detail for this article and found a helpful and exciting overview! You can find the source here: Overview of the major language models (LLMs).

What are these LLMs anyway and why is everyone talking about them?

Before we go into detail, let’s briefly clarify what we’re actually talking about here: LLMs are the “brains” behind many of the AI applications you may already be using – from text generators to chatbots, like ChatGPT, to smart assistants. Imagine an LLM is like a giant encyclopedia that not only knows words and facts, but also understands how sentences are constructed, how people communicate and the meaning behind words. They are trained to recognize patterns in huge amounts of text data and thus understand and generate human language themselves. The fascinating thing is that these models not only reproduce information, but can also write new, creative and coherent texts based on their training knowledge. For us in our day-to-day work, this makes many language-related tasks incredibly easier and more efficient – whether we’re drafting emails, brainstorming or creating content.

The secret of “context” – Why Gemini “notices” more

This is where I had my aha moment with Gemini! When I say Gemini remembers “more context”, I’m talking about something very important in the world of LLMs: the context window. Imagine you are talking to someone. If this person remembers everything you have discussed in the last ten minutes, the conversation will be fluid and coherent. But if she can only remember the last two sentences, it will be difficult to have an in-depth conversation, won’t it?

It’s similar with LLMs. Every conversation you have with an LLM consists of a series of “tokens”. A token is the smallest unit of information that an LLM can process. This can be a word, a punctuation mark or even just part of a word. The context window of an LLM now determines how many of these tokens it can “remember” at the same time in order to understand your current request and generate a suitable response. The larger this window is, the more of our previous conversation the model can keep an eye on.

If we look at the Exploding Topics table, it quickly becomes clear why I have this impression:

• Gemini 2.0 Pro, for example, has a context window of 2,000,000 tokens!
• Gemini 2.5 Pro is also huge with 1,000,000 tokens (soon even 2,000,000).
• In comparison, many ChatGPT models (such as GPT-o4-mini, GPT-o3) often have “only” 200,000 tokens or even GPT-4o and GPT-4o mini with 128,000 tokens.

As you can see, the difference is huge! And that’s exactly why my interactions with Gemini often feel so much more fluid and in-depth. When I’m working on a topic over a longer period of time or when I’m faced with complex tasks, Gemini is better able to take into account everything we’ve previously discussed and thus provide me with more precise and relevant answers. It’s as if it’s really listening and remembering my thoughts.

The “Knowledge Cutoff” – When was AI born?

Another important point that appears in the overview is the Knowledge Cutoff Date. I also had a comment about this on LinkedIn that might be of interest to many of you. This “knowledge cut-off” simply refers to the date up to which the LLM was trained. All information created after this date is unknown to the model due to its original training.

Here, too, it is worth taking a look at the table:

• Many Gemini models (e.g. Gemini 2.0 Pro) have a knowledge cut-off in August 2024 or even January 2025 (Gemini 2.5 Pro).
• Some GPT models (e.g. GPT-4.1) have a cut-off in May 2024 or October 2023 (GPT-4o).

This means that the basic knowledge of the model is only current up to this key date. However, this is where a crucial aspect comes into play that you must pay attention to: Some LLMs, such as Gemini, are not limited to this “training date” when it comes to up-to-date information. You have the ability to access current data via the Internet. This is a game changer! Even though their knowledge from the training may be “old”, they can use the web search to incorporate the latest information into your answers. This is extremely important, especially for us who work in a fast-moving environment, as it means we can always stay up to date without having to wait for the next training round of the model.

What does all this mean for you in practice?

As practice-oriented users, you benefit enormously from these developments. A large context window means:

• Better conversation management: You can have longer, more complex conversations without the model losing the thread.
• More coherent results: If you have a longer text created or are working on a project, the results remain more consistent and coherent across the various requests.
• Less repetition: You don’t have to keep repeating the same information to the model because it remembers more.

And the ability to access current data means for you:

• Always up to date: You get answers based on the latest information, even if the basic knowledge of the model is older.
• More reliable information: Particularly important when it comes to fast-moving topics, market analyses or current events.

I hope this little “deep dive” into the world of LLMs was as enlightening for you as it was for me! It is fascinating to see how the technology behind these models influences and improves our daily work. If you have any questions or would like to share your own experiences, please get in touch at any time! I am always happy about the exchange.