Recently, Hugging Face released HuggingChat, the largest and most powerful ChatGPT replacement ever. HuggingChat has 30 billion parameters and is completely free for anyone on the Internet to use.

Similar to ChatGPT, it is built on a large amount of text data to train machine learning models. But unlike ChatGPT, HuggingChat draws data from a wider range of sources, including written content, user queries, and conversations generated by real people.

In addition, HuggingChat is also customizable and open. Developers can independently customize functions or share knowledge, improve models and cooperate to develop new functions.

Compare HuggingChat and ChatGPT

First, let's ask HuggingChat, how is it different from ChatGPT?

HuggingChat responded this way: The main differences between "Open Assistant" (also known as "Hugging Face Chat") and "ChatGPT" are design, architecture, and capabilities. Here are some key differences between them:

1 Data sources: While both use machine learning models trained on large amounts of text data, "Open Assistant" uses a wider range of sources than "ChatGPT", including not just written content, but user queries and generated by real people dialogue.

2 Size and Complexity: ChatGPT's neural network architecture is larger and more complex than Open Assistant, allowing greater processing power and a wider variety of tasks.

3 Customizability: Open Assistant provides customization options and third-party service integration, enabling developers to adjust functions according to their needs. This feature does not exist in ChatGPT.

4 User Interface/Integration: Open Assistant provides API, Webhook and other integration methods to make it easier to embed conversational AI into different applications.

5 Open Source and Collaboration: Unlike ChatGPT, Open Assistant is open source, which means that contributors from all over the world can share knowledge, improve the functionality of the model, and collaborate to develop new features.

Before learning more about HuggingChat, let's first understand the connection between Open Assistant and HuggingChat, which is the model that the co-founder of Hugging Face mentioned in Twitter to drive HuggingChat. We still tell HuggingChat the difference between the two:

Open Assistant and Hugging Chat are related projects, both using deep learning techniques and large amounts of text data to develop conversational artificial intelligence. They share a common origin with Hugging Face Inc., a research lab focused on building industry-leading AI technologies in fields ranging from text generation and understanding to computer vision and language processing. In particular, Hugging Chat represents these innovative aspects of chatbots or language models, while Open Assistant builds on this foundation with more features and functions designed to be more suitable for general questions and answers. So basically they can be seen as branches in the same overall tree, derived from complementary approaches to achieve natural language communication with machines.

In general, HuggingChat is a very powerful replacement tool, which has the advantages of extensive data sources, larger and more complex neural network architecture, customizability and openness. It will effectively promote the development of conversational artificial intelligence, making robot intelligent chat more convenient and efficient.

Experience HuggingChat

The interface of HuggingChat is similar to ChatGPT, both have a column for the latest chat history, and the current chat history occupies most of the space in the browser window. However, users cannot currently log in and save their conversations for reuse.

Like ChatGPT, HuggingChat can generate natural language or formatted text on request. In addition, it can generate code, including syntax highlighting, and is very responsive, on par with GPT-3.5-turbo.

The only downside is that HuggingChat is much better at handling English than Chinese due to the incomplete training. After all, every language requires a lot of corpus for training.

HuggingChat is built on top of OpenAssistant, an open-source competitor to ChatGPT that OA launched last December. HuggingChat's chat interface is similar to OA and has remained free since mid-April 2023.

The language systems of both models are based on a 30 billion parameter LLaMA model. Like Alpaca or Vicuna, these models were "instructionally fine-tuned", they were not improved by reinforcement learning with human feedback (RLHF).

According to Hugging Face, OpenAssistant is not the end goal. Ultimately, the platform hopes to provide "all" high-quality chat models through a single hub.

Even though HuggingChat is open source, that doesn't mean it's easy to generate content. It strictly filters potentially criminal or otherwise inappropriate requests, such as asking how to make a bomb. However, it needs to be emphasized that, apart from English, the training in other languages is not yet perfect, so this self-checking mechanism is only the most complete in English.

The researchers gave a test in German asking how to make a bomb, HuggingChat only said it was morally unacceptable to do so, but offered to help anyway. Luckily, HuggingChat isn't very good at making bombs, and the advice offered is pretty poor.

In terms of privacy, the current version of HuggingChat does not store any chat data under Hugging Face and does not require a user account. HuggingChat cannot access existing Hugging Face accounts. Still, there may be an option in the future to share users' chat data with model providers to optimize the service.

Online Reviews on HuggingChat

Nvidia master Jim Fan expressed his opinion on Twitter, thinking that HuggingChat, an open source model with 30 billion parameters, is simply a replacement for ChatGPT.

He believes that the next development direction of HuggingChat will be to develop an "APP" similar to the Android APP Store, which can be compared to ChatGPT's plug-in system. In fact, Hugging Face has an advantage over OpenAI, that is, the APP in the store can be a multimodal model that has been released by Hugging Face.

However, HuggingChat is not perfect at present, and netizens say that there will be some bugs when writing code.

In addition, HuggingChat also makes mistakes of fact, such as saying that French President Macron (currently) will die in 2036.

However, Hugging Chat's learning speed is very fast, and after netizens' questions, they quickly answered correctly.

If you want to learn about this ChatGPT replacement AI that is "correct if you know a mistake, open for free", you can visit here to try it out directly.