3
3
Singapore
2 years of experience
I used to work as a data product manager, but now I have transitioned into full-stack development.
This is a life simulation game that simulates the most important events that occur each year in a player's life. Players can customize their character's attributes and talents. CLAUDE AI, considering the style of the game, predicts the character's life. The game style can be close to reality or depict a fantastical future or past. The motivation behind creating this game is that people's attention is often focused on the near future, and humans are inherently not very adept at long-term planning. This game extends the perspective of viewing things to the span of an entire life. The backend of this project is Flask and the frontend is Vue, with most of the development carried out by me.
Claude provides an API that accepts up to 100k tokens. AutoGPT faces challenges with memory and retrieval, which you can learn about at https://docs.agpt.co/challenges. Therefore, we can use the Claude API, which accommodates 100k tokens, to run AutoGPT and avoid these memory and retrieval challenges. So, I've modified the AutoGPT source code. I have replaced the chat API with Claude and disabled memory. Another benefit of using AutoClaude is that it's free! The Claude API does have a rate limit for each request, but there's no limit on usage frequency. Therefore, we can run the Claude API continuously.
We are attempting to modify auto_gpt to auto_bard. When Bard generates JSON, it sometimes produces abnormal JSON. Therefore, we are using OpenAI's AI function to ensure each JSON is correct. We have used the Claude API to accomplish the same task. You can see a similar project in another branch of this repository. We found that the performance of Claude is superior to that of GPT-3.5 Turbo, and GPT-3.5 Turbo is better than Bard. However, Bard's response speed is higher than the other models. We believe Bard's limitation lies in its input token limit, as it only supports up to 8192 tokens, which is quite short. If Bard could support more input words, it might be a good candidate for an agent