Stability AI’s StableVicuna is the First Chatbot Trained with Human Feedback
In Brief
Stability AI releases StableVicuna, the first large-scale open-source chatbot, which uses Reinforcement Learning with Human Feedback (RLHF).
StableVicuna is based on the Vicuna chatbot and uses a 13 billion parameter LLaMA model.
Stability AI has introduced its latest breakthrough in AI, StableVicuna, the first large-scale open-source chatbot trained with human feedback. The innovative chatbot is the brainchild of Stability AI, the company that created the popular open-source image model, Stable Diffusion, and the newest AI image generation algorithm, DeepFloyd.
StableVicuna is based on the Vicuna chatbot released in April, which uses a 13 billion parameter LLaMA model. What sets the Vicuna variant of Stability AI and Carper AI apart is its use of Reinforcement Learning with Human Feedback (RLHF). This method enables the model to improve continuously.
Stability AI suggests that chatbots are successful because of two training method types: instruction fine-tuning and reinforcement learning through human feedback. However, most existing chatbot models use only one of these methods and not both. Recently, datasets for RLHF training have become publicly available. Thus, along with a user-friendly training tool, this has enabled the creation of StableVicuna, which is the first large-scale chatbot model that incorporates both types of training.
StableVicuna incorporates text generation, simple math functions, and the ability to write code. It is comparable to other open-source chatbots in common benchmarks.
According to The Decoder, open-source chatbots fine-tuned with data from other chatbots risk amplifying existing errors and biases through repetitive training, causing an echo chamber effect. Fine-tuning data can also exacerbate hallucinations by introducing information not present in the original model.
Users can access a demo of the chatbot on HuggingFace. The company has also disclosed plans to provide StableVicuna through a chat interface in the future.
Read more:
Disclaimer
In line with the Trust Project guidelines, please note that the information provided on this page is not intended to be and should not be interpreted as legal, tax, investment, financial, or any other form of advice. It is important to only invest what you can afford to lose and to seek independent financial advice if you have any doubts. For further information, we suggest referring to the terms and conditions as well as the help and support pages provided by the issuer or advertiser. MetaversePost is committed to accurate, unbiased reporting, but market conditions are subject to change without notice.
About The Author
Agne is a journalist who covers the latest trends and developments in the metaverse, AI, and Web3 industries for the Metaverse Post. Her passion for storytelling has led her to conduct numerous interviews with experts in these fields, always seeking to uncover exciting and engaging stories. Agne holds a Bachelor’s degree in literature and has an extensive background in writing about a wide range of topics including travel, art, and culture. She has also volunteered as an editor for the animal rights organization, where she helped raise awareness about animal welfare issues. Contact her on agnec@mpost.io.
More articles
Agne is a journalist who covers the latest trends and developments in the metaverse, AI, and Web3 industries for the Metaverse Post. Her passion for storytelling has led her to conduct numerous interviews with experts in these fields, always seeking to uncover exciting and engaging stories. Agne holds a Bachelor’s degree in literature and has an extensive background in writing about a wide range of topics including travel, art, and culture. She has also volunteered as an editor for the animal rights organization, where she helped raise awareness about animal welfare issues. Contact her on agnec@mpost.io.
