The Hugging Face web site on a smartphone organized in New York, Aug. 17, 2023.
Gabby Jones | Bloomberg | Getty Photographs
Hugging Face, an AI agency based mostly in New York, has raised $235 million at a $4.5 billion valuation from a few of expertise’s largest firms.
Google, Amazon, Nvidia, Salesforce, AMD, Intel, IBM and Qualcomm all contributed to the spherical, the corporate mentioned. Hugging Face CEO Clement Delangue mentioned the funds are to be targeted on hiring expertise to be aggressive within the synthetic intelligence area.
Startups engaged on AI fashions have reached excessive valuations as huge firms and enterprise capitalists search to plow cash within the latest AI growth, which kicked off final 12 months when Microsoft-backed OpenAI launched its ChatGPT chatbot.
Hugging Face’s huge valuation and crop of outstanding backers mirror how a extra collaborative method to constructing AI has been gaining steam in latest months, particularly after Fb guardian Meta launched its Llama giant language mannequin, which is free to make use of for the overwhelming majority of firms.
Different extremely valued AI startups, like OpenAI or Cohere, work on the expertise immediately and guard the outcomes as a commerce secret, then cost prospects to entry them by way of utility programming interfaces, or AIs.
However Hugging Face produces a platform the place AI builders can share code, fashions, knowledge units, and use the corporate’s developer instruments to get open-source synthetic intelligence fashions working extra simply. Particularly, Hugging Face typically hosts weights, or giant recordsdata with lists of numbers, that are the guts of most fashionable AI fashions.
Whereas Hugging Face has developed some fashions, like BLOOM, its main product is its web site platform, the place customers can add fashions and their weights. It additionally develops a sequence of software program instruments referred to as libraries that enable customers to get fashions working rapidly, to scrub up giant datasets, or to judge their efficiency. It additionally hosts some AI fashions in an internet interface so finish customers can experiment with them.
It is comparable in theme and apply to code-repository GitHub (which Microsoft acquired in 2018), the place coders from world wide publish their initiatives whereas they’re engaged on them.
Hugging Face endorses the idea that almost all firms working with AI will wish to develop their very own fashions or expertise, and can want instruments to take action, co-founder and CEO Delangue instructed CNBC. He hopes that AI builders will depend on Hugging Face each day to get their work finished.
One motive the massive firms are investing: Their staff are actively utilizing the platform, he mentioned.
“AI builders are utilizing Hugging Face all day, daily,” Delangue mentioned. He predicted that the variety of software program builders working with AI fashions would develop within the coming years.
“Possibly in 5 years, you are going to have like 100 million AI builders. And if all of them use Hugging Face all day, daily, we’ll clearly be in an excellent place,” he mentioned.
Though most consideration in latest weeks has been on so-called giant language fashions like ChatGPT or Llama that target producing textual content, Hugging Face hosts any AI mannequin, together with ones that generate music or photographs, translate languages, or determine objects inside photographs. Hugging Face hosts 500,000 completely different AI fashions, 250,000 knowledge units, and has 10,000 paying prospects, the corporate mentioned.
Delangue instructed CNBC he had not too long ago performed with IDEFICS, which permits customers to add a picture and ask questions on it. He additionally talked about Seamless M4T, a translation mannequin from Fb, in addition to Llama2, Fb’s language mannequin.
What’s with the humorous title?
Hugging Face is called after an emoji, the cuddling face, a smiley face framed by two open fingers.
The title and emblem date again to the corporate’s founding. Hugging Face was initially a iPhone chatbot app, however when the corporate open-sourced some its machine-learning code, it realized that it was catching on with AI builders, and pivoted towards that.
“Once we began the corporate, with my co founders Julien Chaumond and Thomas Wolf, we joked that we needed to be the primary firm to go public with an emoji as a substitute of the three letter ticker,” Delangue mentioned.
“Possibly throughout this spherical we must always begin our lobbying train with the Nadsaq for them to permit us to make use of emojis on their board,” he quipped.
