Summary Simulating my friend Philippe. Philippe digital twin | by Santiago Ortiz | Jun, 2023 | Medium medium.com
3,202 words - html page - View html page
One Line
The author explores the concept of simulating individuals using digital twins, specifically discussing the case of simulating their friend Philippe using ChatGPT and emphasizing the potential applications and limitations of this technology.
Slides
Slide Presentation (7 slides)
Key Points
- Simulating a person using chatgpt requires managing external information and feeding the model with relevant parts of the person's writings.
- Chatgpt can simulate well-known figures like Aristotle, but simulating less well-known individuals like Philippe is more challenging.
- The relevancy of text chunks for a given question can be measured by comparing their semantic similarity to the question.
- Large Language Models like chatgpt create meaning by placing words and texts in embedding spaces, allowing for comparisons of semantic similarity.
- The process of finding relevant information for chatgpt prompts is becoming popular and can be used for various purposes, such as collective knowledge interfaces.
- The author has implemented this approach in projects involving social media conversations and introducing complexity thinking in education.
- Interacting with systems through language allows for powerful functionalities and eliminates the need for complex interfaces.
- The objective is to make people feel they are talking with Philippe while understanding how the language computation process works.
Summary
439 word summary
Santiago Ortiz has created a digital twin of his friend Philippe using ChatGPT. The goal is to make people feel like they are talking with Philippe and to understand the language computation process. The simulation does not require an interface, only an input text box. Santiago is using this technology to create interactive interfaces and educational tools. He is interested in introducing complexity and non-linearity into the classroom. Santiago also discusses the use of Large Language Models (LLMs) and their ability to generate meaningful text. He visualizes the text chunks using embeddings and explains how LLMs measure semantic similarity. Santiago demonstrates how he uses ChatGPT to simulate Philippe by providing a long prompt that contains 50 texts written by Philippe. He selects the most relevant chunks for a given question and uses them as source material for generating answers. The simulation allows Santiago to capture Philippe's writing style and ideas. In this article, the author discusses the concept of simulating a person using a digital twin. They explain that the system can detect and retrieve relevant information in real time to provide logical and meaningful answers. The author mentions using this methodology for various purposes such as academic research, literary analysis, and email assistants. They emphasize that this architecture is already becoming standard and provide references for further reading.
The author goes on to explain the mechanics of the simulation, highlighting the multidimensional space for exploring social media conversations, reading books, asking questions, and gaining insights. They mention that the simulation can be extended in any direction and provides a solution for simulating less well-known individuals.
The author mentions that they trained the model with texts from Philippe and that the general problem is that the model lacks awareness of who Philippe is and his ideas. They highlight that the simulation can combine and recombine written information at convenience.
The author discusses the magic of being able to access Aristotle's thoughts through big language models like chatgpt. They mention that time is a fundamental reality and raise questions about its essence and whether it flows or exists as an illusion.
The author introduces the idea of simulating Aristotle using chatgpt and asks whether it can simulate his ideas and writing style. They mention that they used the system to simulate their friend Philippe and provide a link to read some of the conversations.
In conclusion, the author shares their project of simulating a friend using chatgpt and highlights the potential of using a person's texts to create a digital twin. They emphasize the need for managing external information and mention the limitations of chatgpt in receiving very long prompts.