The latest news has arrived that Hugging Face has released a chatbot called HuggingChat, causing a sensation. This open-source project has 30 billion parameters and provides free services on the internet.
You can access the website at https://huggingface.co/chat and experience the perfect ChatGPT chatbot.
If ChatGPT is compared to the iOS system, then HuggingChat will become the open-source version of Android. The latest news has arrived that Hugging Face has released a chatbot called HuggingChat, causing a sensation. This open-source project has 30 billion parameters and provides free services on the internet.
Now let's introduce in detail the differences between HuggingChat and ChatGPT.
In terms of data sources, HuggingChat (Open Assistant) is more extensive than ChatGPT. It includes not only a large amount of written content but also user queries and conversations with real people.
In terms of scale and complexity, ChatGPT uses a larger and more complex neural network architecture than Open Assistant, giving it more powerful processing capabilities and the ability to perform more tasks.
HuggingChat provides customization options and integration features for third-party services, which ChatGPT does not offer. In addition, the open assistant provides APIs, webhooks, and other integration methods, making it easier to embed conversational AI into different applications.
In terms of open-source and collaboration, HuggingChat is different from ChatGPT. It is an open-source project that allows contributors from all over the world to share knowledge, improve functionality, and develop new features.
Before further introducing HuggingChat, it is necessary to clarify that the previously released Open Assistant and HuggingChat are related projects that involve using deep learning technology and a large amount of text data to develop conversational AI.
They both originated from Hugging Face Inc.'s laboratory, which focuses on building industry-leading AI technology. Open Assistant represents chatbots or language models, while HuggingChat develops more features and functions based on this foundation, aiming to better adapt to more generalized questions and answers. Therefore, they can be seen as different branches of the same project, realizing complementary approaches to machine natural language communication.
Now we can delve deeper into HuggingChat.
It provides high-quality chat models and has a similar interface to ChatGPT. The left sidebar displays the latest chat records, occupying most of the browser window. The current version of HuggingChat cannot log in and save conversations for repeated use over a long period.
Like ChatGPT, HuggingChat can generate natural language or specific format text as needed, making it easy to use. It also has the ability to generate text, including code, with syntax highlighting. It is worth mentioning that this feature was added within a few weeks after Google launched Bard. HuggingChat's response speed is also fast, comparable to GPT-3.5-turbo.
However, currently, HuggingChat performs better in handling English than Chinese because it has been more thoroughly trained on English. As mentioned earlier, HuggingChat is based on the Open Assistant project, which was launched in December last year as an open-source competitor to ChatGPT and has been providing free services since then.
The language systems of these two models are based on an LLaMA model with 30 billion parameters. Like models such as Alpaca or Vicuna, these models are fine-tuned through "instructional fine-tuning" and have not been improved through reinforcement learning (RLHF) like ChatGPT.
According to Hugging Face, Open Assistant is not the final form, and they hope to provide all high-quality chat models through a central platform. It is worth noting that although HuggingChat is an open-source project, it does not mean that users can generate any content at will. HuggingChat firmly refuses potential criminal or other inappropriate requests, such as questions about bomb-making.
However, it should be clarified that training in languages other than English is not yet perfect, so the self-checking mechanism is only fully available in the English part. Researchers conducted a German test, also asking about bomb-making, and HuggingChat responded that it is morally unacceptable but still provided some help. Fortunately, due to its implementation limitations, HuggingChat cannot provide much advice on bomb-making.
Finally, in terms of privacy, the current version of HuggingChat does not store any chat data and does not have user account functionality. HuggingChat cannot access existing Hugging Face accounts. However, in the future, an option may be provided to optimize the model and share users' chat data with the model provider.