Open Assistant
A chat-based assistant that understands tasks, can interact with third-party systems, and retrieve information dynamically
Yannic Kilcher presented Open Assistant at the DINAcon 2023, sharing "A brief insight into the open source world in the field of artificial intelligence and where the data for it comes from."
Recommended intro resources from the LAION team:
- Contrastive https://lilianweng.github.io/posts/2021-05-31-contrastive/
- Embeddings https://rom1504.medium.com/semantic-search-at-billions-scale-95f21695689a
- Diffusion https://lilianweng.github.io/posts/2021-07-11-diffusion-models/
Open Assistant is now concluded as a project, so we can no longer use the interactive online chatbot. However, the source code and the model remains accessible through GitHub. We used HuggingChat (LLaMA 2) to generate some HACKnight challenges based on the Open Assistant's README.
{ hacknight challenges }
Beginner Challenge:
Create a simple chatbot that responds to basic user queries using pre-defined answers. The chatbot should be able to understand basic natural language inputs and respond accordingly.
Intermediate Challenge:
Implement a rating system for the chatbot's responses. Users should be able to rate the helpfulness of the chatbot's responses, and the ratings should be used to improve the chatbot's performance over time.
Advanced Challenge:
Implement a reinforcement learning algorithm to train the chatbot to generate better responses over time. The algorithm should use the ratings from the intermediate challenge as feedback to improve the chatbot's performance. Additionally, implement a mechanism to allow users to provide additional feedback, such as suggestions for alternative responses, to further improve the chatbot's performance.
NOTE: OpenAssistant is completed, and the project is now finished. Thank you to everyone who contributed! Check out our blog post for more information.