HuggingChat, a great open-source alternative to ChatGPT

Online now there are dozens of alternatives to ChatGPT based on the GPT-3.5 language model (or the GPT-4 successor, in the best cases) with some more specific functionality depending on the focus of the developers. For example, in early May 2023 we observed the case of AI concept, a productivity tool designed for bloggers and marketing and content creation workers. Today, however, we will talk about HuggingChata free and open source alternative developed by the US company Hugging Face. Let’s see what it is able to do and the most evident differences between this solution and the one offered by OpenAI.

What is HuggingChat?

HuggingChat is a collaboration between Hugging Face and LAION, a German non-profit organization otherwise known as Large-scale Artificial Intelligence Open Networkwhose job is to “make ML models, datasets, and any AI-related source code more accessible to the general public”, pulling them from the web following their use to train models already known as Stable Diffusion and Imagen. In the specific case of HuggingChat, the partnership manifests itself in the use of the LLaMa model created by LAIONfor which HuggingChat essentially serves as a user interface.

The key advantage over ChatGPT is its open source nature: this means that users can access every minimum information regarding the internal mechanisms of the AI, replicating or improving them thanks to their experience. Furthermore, this way it is possible to make sure that the generative AI does not violate the privacy of the users, ergo gain unwanted and unauthorized access to sensitive data or stored information shared by the consumer for content generation.

During the month of April 2023 the estimates published by the staff behind HuggingChat spoke of training based on about 30 billion parameters, a much lower figure than the 175 billion for GPT-3.5 and the even richer database fed to the successor GPT-4. Despite this, with these numbers the Hugging Face chatbot is in all respects the open-source AI chatbot with the best possible performance.

How does it work?

Those who have used ChatGPT will feel at home: theuser interface Of HuggingChat it is very familiar to the one visible in the case of the OpenAI chatbot, as the text box is positioned right in the lower central part of the window, while on the left is the message history. Once you type the prompt in the input box, in a few seconds the AI ​​will start generating the response. Nothing new, really.

Do you want to use the chatbot to develop a game? No problem. The same goes for more technical and creative requests, from writing stories to drafting emails and more technical paragraphs to authoritative texts. It is also interesting to note that when the site is opened, HuggingChat immediately displays the language model used and allows access to the pages from which to download both the model itself and the dataset on which it was trained. Finally, the user interface is downloadable via GitHub to understand how it works and also use it locally, after downloading the LLaMa and the dataset.

In terms of quality, however, ChatGPT is naturally the best tool: generally, HuggingChat shares discrete quality output with the user, with even more evident defects compared to the OpenAI signed counterpart, including the generation of apparently coherent answers but based on false data or wrongly manipulated by artificial intelligence, eventually reaching nonsensical and incorrect conclusions.

What is causing this problem? Most likely from the training methodology adopted by the developers, ie learning reinforced by human feedback. According to this method, it is human beings who train the artificial intelligence through a series of questions and answers, correcting some outputs in order to make them suitable for sharing with a greater number of users. For this reason, HuggingChat could provide well-founded answers about prejudices included in user feedback.

Specific use cases

As for answers for more specific queries, HuggingChat delivers better coding performance and returns the complete code without detailed instructions, spending less time sharing them as output. Or again, the Hugging Face chatbot differs from the competition thanks to advanced features on the personalization front: every user, especially teenagers, will be able to establish an “authentic bond” with artificial intelligence, although the developers have not exactly explained how .

Finally, having spent much less time developing and relying on far fewer parameters than ChatGPT, regardless of the use case it is clear the lack of refinement compared to the GPT language model. If nothing else, it remains open-source, ergo visible and analyzable in its entirety by anyone, even those who don’t have much experience in the sector.

Simply put, HuggingChat is an excellent alternative to OpenAI’s solution for anyone looking for one Free, transparent and efficient AI, “settling for imperfection”. Now that ChatGPT can use the InternetHowever, some might ultimately favor the generative AI that launched the industry into hyperspace, although the extraction of information from the web is actually only accessible by paying for the Plus subscription.