This new Gemini feature is the best and worst thing that’s happened for my learning

I’m typically a generative AI skeptic, but this doesn’t mean I’m not open to new AI tools that could be useful. After all, I found ways to use NotebookLM in my daily life even though I’m not a student. I even use Gemini’s Live mode to practice my second language.

So when Guided Learning rolled out to the Gemini app, I wanted to test out whether it could turn my eagerness to learn new things into actual knowledge. The results were mixed, however.

I absolutely love learning new things, which is why I have apps that let me learn something new every day. However, I have also been very burnt out due to stress and learning has taken a backseat. When Google rolled out Guided Learning for Gemini, I decided to use it to dive into topics that I’ve been having trouble researching.

Self-hosting is one of these topics. The subject is vast and even though I’ve been self-hosting services for months, there’s still so much to learn. I have particularly been having trouble wrapping my head around certain networking concepts. Despite spending hours researching different tools, it was difficult to develop a clear understanding because I don’t have any experience in IT.

Once I saw that Guided Learning is now available, I saw this as my chance to really advance my knowledge. I would be able to ask questions I might be too embarrassed to ask colleagues and Redditors and I could add follow-up enquiries when an explanation needs to be simplified a bit more.

Up until now, I used queries through Brave’s search engine to learn about concepts. But I often end up getting derailed as I go between different websites or add additional queries. I also wanted to try out Google’s Learn About app, but it’s not available in my region yet.

The first session with Guided Learning was productive. Even though my prompt was very broad, the chatbot offered me some choices for which core concepts I’d like to start with. It provided a digestible breakdown of the topic and asked follow-up questions to check my understanding. It encouraged me when I got things right and it didn’t feel intimidating at all. I could also generate quizzes and flash cards to go over concepts and test my knowledge.

It made it easier for me to understand concepts like a reverse proxy versus a Cloudflare tunnel, along with their benefits. I was specifically able to ask it about which approach would be better for a beginner to implement.

I also eventually dove into the topic of central sensitization, which I wanted to expand my knowledge on due to its role in two conditions I have: fibromyalgia and chronic migraines. I have familiarity with the concept, but wanted to find out about its role and differences in each condition.

The chatbot was able to adapt to my questions and made sure not to present information in a way that was overwhelming. If I asked a complex question, it would handle concepts separately to make sure I understood them.

It was also able to easily switch between topics. I could start a session on self-hosting, then switch it to a completely different concept or topic within the same session. In this way, I’d say it is a little bit better than a real-life tutor because it has access to the internet as its knowledge base. This means I could essentially follow my bliss and ask it any questions on topics I found interesting. If I was talking to a person, it would be unlikely they had the knowledge of all the different topics I wanted to discuss.

I also like that it can generate diagrams to help with visual learning, although I did find the results mixed. The quizzes were the best part since they allowed me to apply my newly learned knowledge.

However, the honeymoon ended more quickly than I anticipated and small flaws quickly turned into major frustrations that limited my learning experience.

Google has poised Guided Learning as providing a collaborative, personal study companion that meets you where you are. It uses learning science to teach users in an expert-backed way. But in practice, I found Gemini’s approach restrictive and unwilling to adapt to my needs.

I immediately noticed the chatbot’s penchant for metaphors. While these can be helpful to explain concepts, I found that it over-used them and sometimes even mixed up its figurative examples in follow-up responses. I found some of its metaphors useful, but got irked by its over-reliance on them.

But what frustrated me the most is the chatbot’s focus on asking questions. Without fail, every response from the chatbot included a question — even if I said I knew nothing about a topic. At first, I thought this might be due to the open-ended nature of a prompt like “I want to learn more about self-hosting”. However, even if I asked the AI a direct question, it would skip past comprehensively explaining something, and focus on its own question instead.

For example, I asked it what the role of Substance P is in fibromyalgia. Instead of answering, it asked me what I know about Substance P and what its role in the body is. To lean into Gemini’s love of metaphors and similes, it’s as though you’re discussing a topic with someone who doesn’t know the answer to a question, so they turn the question on you instead.

Answering the question does not placate Gemini either. It will always have another question loaded. This can derail the conversation entirely. It’s also especially frustrating when you actually know nothing about a topic and have to repeatedly respond with “I don’t know” and other variations.

After saying “I don’t know” three times in a single session, I switched it up and started using responses like “Show me” and “Tell me”. But this stopped working almost immediately, with the chatbot chiding me to say that it wants to have a dialog to create collaborative learning.

I later asked it if it could ask me fewer questions. It said it understood and then proceeded to end off the response with another question. After the next response also had a question, I gave up on this approach.

This isn’t to say that the chatbot is completely useless, but if constant probing isn’t your favored method of learning, the questioning is tedious. Even if you answer the question, about a third of the response is focused on encouragement rather than the topic at hand. Another third is the question. So only about a third of the response is giving new information.

This also means that when you really went to dive deep into a topic, you’re often not going to get a detailed response. The chatbot just scratches the surface and is quickly waylaid by asking you what you know about the topic.

Don’t get me wrong, questions can be a useful way to figure out if a student understands a concept and allows them to critically think about it. It also allows you to extend your knowledge by thinking about the implications of what you have learned. This is why the algorithm uses this approach. Personally, I loved answering and discussing questions in my university classes as it created new ways to think about topics.

However, the approach the chatbot uses is a bit too rigid — especially if you’re learning about a new concept. These questions are great for revision, but not when you’re clueless about a topic. Teachers and lecturers often use the Socratic method to get students to engage with a topic and apply newly learned knowledge. But it’s usually when a debate is involved or foundational knowledge it already in place. These questions are used in moderation, not as a constant script the educator reverts back to.

I also found that the focus on questions meant that the chatbot often got stuck in loops by regenerating a previous response and asking you the same question over and over. I would have to exit the session to stop the loop, or change my answer even if the original was correct.

Another metaphor, if you will. Gemini’s Guided Learning was like a tantalizing serving of junk food. I got hyped up and excited to learn, but it didn’t keep me full for long. I craved something more substantial. It stoked my curiosity, but could not explore things to the extent I wanted to because it focused more on metaphors and questions than diving deeper into a topic.

This has motivated me to continue expanding my knowledge of various topics. But the unsatisfying nature of my interactions with the chatbot mean that I’m on the search for more robust solutions.

I’m glad that Gemini has reignited my curiosity and learning spirit during a period of burnout. However, I plan to find ways to actually sate my appetite for knowledge since Guided Learning left me frustrated overall. I do hope that the algorithm is refined so that it can be used to really dive into topics and adapt to a person’s learning needs, rather than following a rigid format.