AI Training Its Own Successor? A Surprising Personalization Experiment for $9

Imagine this: What if you had an AI assistant that perfectly understood your daily work habits and priorities, mimicked your speaking style, and provided customized information just for you? In the past, this would have required enormous costs and specialized technical skills, but we are now living in an era where anyone can create their own “AI double” with about $9 worth of resources.

Recently, a case where someone used Google’s large language model (LLM—a massive AI that writes text or answers user questions), Gemini, to train their own personal assistant has garnered attention. Gemini is a highly intelligent assistant that goes beyond a simple chatbot, helping in various areas such as writing, planning, and brainstorming[16]. Today, I will explain, in an easy-to-understand way, how you can build your own assistant using the AI available to us, rather than in a grand laboratory, and what the technical background and significance of this are.

Why It Matters

We already live in an era where we converse with Google Assistant on our smartphones. Now, if you choose the Gemini app, it can replace the existing assistant and act as the “main assistant” on your smartphone[6].

This change signifies more than just using an “AI that speaks well.” As users can now directly train or provide instructions to an AI that understands their specific context (background knowledge for questions or situations)[7], AI is evolving from a universal tool into a “personal assistant just for me.” This is like having an AI by your side 24/7 that keeps a secret work notebook that no one else knows about.

The Explainer

Gemini is a collection of multimodal (AI models that understand not only text but also various forms of data such as images, audio, and video) AI models developed by Google DeepMind[9].

The principle behind using this technology to create your own assistant is similar to “Knowledge Distillation.” Simply put, it is like extracting the most important core summaries from a massive library (a large model like Gemini 1.5 Pro) and copying them into a small, light portable notebook (an efficient model like Gemini 1.5 Flash)[10].

When we command an AI, “From now on, summarize all answers in bullet points” or “Answer by prioritizing my work schedule,” we are adding our own rules to this portable notebook[7]. As these small modifications accumulate, it starts to operate like “my own model” that is completely different from the giant model. It is similar to an apprentice who has learned the recipes of a great chef and modifies the cooking methods to suit their own taste perfectly.

Where We Stand

Currently, Gemini has a diverse lineup including Gemini Pro, Deep Think, Flash, and Flash-Lite, depending on performance and purpose[9]. Ashwin Kannan, an engineer at Palo Alto Networks, evaluated that “The Gemini 3.5 Flash-Lite model is very fast and reliable, becoming an excellent choice that responds immediately when the user needs it”[5].

However, AI is not always perfect. Models like Gemini are still undergoing continuous learning and testing to perfectly grasp user intent, and sometimes users test the AI’s safeguards or ask questions in unexpectedly complex ways[12]. Therefore, the performance of the “personal assistant” we create can vary significantly depending on how the user guides it—that is, how precisely you input the rules.

What’s Next

The concept of “training AI” will become easier and easier. There are analyses suggesting that some models have referred to Gemini’s data to improve their performance, showing that interaction between models and the flow of knowledge are becoming active[11]. Before long, the day will come when anyone can carry an AI model in their pocket that has learned all their memories and habits for a cost that is effectively free, not just $9.

MindTickleBytes AI Reporter Opinion

Repurposing the knowledge of AI models to create personalized tools is a key step that goes beyond simple feature addition, opening the “era of personal assistants” where individuals own and control fragments of the massive intelligence that is artificial intelligence. However, this is a point in time that requires careful observation, such as whether the AI I trained will act smartly when it goes out into the world, or whether it might behave differently than I intended.

References

1. [How to Enable NSFW Mode on Gemini(2026)

   Gemini… - YouTube](https://www.youtube.com/watch?v=vfgDti2QJsY)

2. Google AI Pro & Ultra — get access to Gemini 3.1 Pro & more
3. Gemini 1097 issue - Gemini Apps Community

4. [Gemini Notebook

   AI research tool and thinking partner](https://notebook.google/?hl=en-GB)

5. Gemini — Google DeepMind
6. Google Gemini - Apps on Google Play
7. Personal context
8. Google Gemini - Wikipedia
9. Gemini (language model) - Wikipedia

10. [What is Google Gemini?

    IBM](https://www.ibm.com/think/topics/google-gemini)

11. [DeepSeek may have used Google’s Gemini to train its latest model

    TechCrunch](https://techcrunch.com/2025/06/03/deepseek-may-have-used-googles-gemini-to-train-its-latest-model/)

12. What is Gemini and how it works

13. [Google Just Launched Gemini, Its Long-Awaited Answer to ChatGPT

    WIRED](https://www.wired.com/story/google-gemini-ai-model-chatgpt/)

14. Gemini for Students — your AI study buddy from Google
15. Google Gemini
16. reddit.com/r/GeminiAI
17. AI Detector - Free AI Checker for ChatGPT, GPT-5 & Gemini