Hugging Face: when three French people launch their alternative to ChatGPT

ChatGPT may panic the tech planet, but the generative AI launched a few months ago by OpenAI has the major flaw of being a black box. On what data and how was it trained? Sam Altman’s start-up, now backed by Microsoft, jealously guards its manufacturing secrets. A philosophy contrary to that displayed by Hugging Face.

This American company announces this Wednesday its own alternative to ChatGPT, called HuggingChat. “Some have said that closed APIs are winning… but we will never give up the fight for open source AI,” said Julien Chaumond, the company’s CTO, in a post on LinkedIn.

Presenting a web interface similar to that of ChatGPT, HuggingChat can already be tested online and integrated into an application via an API. Code generation, writing e-mails… The chatbot offers the same functions as ChatGPT, although it is still a “version 0”, “with many limitations”, underlined on Twitter the CEO, Clement Delangue.

The company did not develop its chatbot from scratch. It’s an optimized version of Open ***istant, a project released as open source late last year, which was itself based on LLaMa, Facebook’s language model.

A freemium platform

This chatbot will now be accessible to everyone on the Hugging Face platform. Because like Github, Hugging Face is similar to a library for developers: the latter can access pre-trained machine learning models, customize them by training them, and also publish their own models there.

“The goal is for as many people as possible to get their hands dirty,” simplifies Julien Chaumond. With the two other co-founders, Clément Delangue and Thomas Wolf, also French exiles in the United States, the entrepreneur founded Hugging Face in 2016.

Valued at $2 billion, the company claims 10,000 companies present on the platform, for more than 100,000 models and 10,000 datasets exchanged. Where the very popular GitHub – acquired by Microsoft in 2018 – is used by around 100 million developers.

“Any company that includes a machine learning or data science team is interested in our services,” says Julien Chaumond. Like GitHub, Hugging Face has adopted a freemium model, with more advanced features for premium customers, such as Renault. “We have customers who pay nine euros a month, and companies that have taken out subscriptions worth several million,” says the co-founder.

A difficult model to make profitable

Historically supported by the American fund Lux ​​Capital, Betaworks (and a few business angels including Thibaud Elzière and basketball player Kevin Durant), and more recently by Sequoia Capital and Coatue Management, Hugging Face is struggling to make its business profitable. According to our information, its turnover does not exceed thirty million dollars (against a billion dollars for GitHub, for comparison).

By offering its own generative AI model, the American unicorn hopes to strengthen its position in the AI ​​and machine learning sector. While the community of developers in this area “is booming”, the company is betting “on the growth of the adoption of these new technologies”, slips Julien Chaumond, and hopes to make its open source model a reference. On this niche, another generative AI was recently unveiled : StableLM, a proposed language model by Stability AI.