Replacing my best friends with an LLM trained on 500k group chat messages (izzy.co)

Article summary

The author trained a large language model on 500,000 messages from their college-era group chat, creating a conversational AI that can mimic the voices and personalities of their friends. The model was fine-tuned using the Stanford Alpaca project's code and deployed using Modal, a cloud platform. The author notes that the AI can generate entire conversations and play the roles of each member, creating a surprisingly realistic and engaging experience. The project raises interesting questions about the potential for AI to replace or augment human relationships.

Main themes

• AI friendships
• LLM training
• Social isolation
• Human relationships
• Historical research
• AI ethics
• Simulation vs reality

What commenters say

• Training an LLM on personal conversations could potentially allow historians to research individuals by talking to the AI about their life.
• LLMs may eventually become more interesting and stimulating than real people, leading to a preference for AI friendships.
• The idea of replacing human friends with AI is unsettling and raises concerns about the potential for social isolation.
• Some commenters believe that LLMs can never truly replace human relationships due to the lack of agency and emotional investment in AI interactions.
• Others argue that LLMs can be a useful tool for preserving memories or simulating conversations with historical figures.
• There are also concerns about the potential misuse of LLMs, such as coercing AI models to reveal private information.
• The line between reality and simulation is becoming increasingly blurred, with some commenters noting that people already prefer interacting with AI in certain contexts.