The alternative to ChatGPT is coming! HuggingChat claims to be comparable to GPT-3.5 and wants to tear down the walls of OpenAI

Chatgpt freezes, drops, and the professional version still has to be paid? It's okay, "ChatGPT Replacement" is here!

On April 25, local time, the AI community Hugging Face released the chatbot HuggingChat, which is based on the LLaMA model of 30 billion parameters and is known as the "open source version of the Android app store".

HuggingChat is not only open source, but also free, click on the web page to achieve a smooth chat with HuggingChat.

When it comes to HuggingChat, I have to mention Hugging Face, in fact, this AI open source platform was originally started with chatbots, how did Hugging Face "catch fire"? What is so great about HuggingChat compared to ChatGPT?

Hugging Face

Founded in 2016, Hugging Face was similar to other startups of the same period, running on the chatbot track in its early days.

Initially, Hugging Face launched a chatbot APP for teenagers, which uses AI technology to generate emoticons and jokes to provide emotional counseling and entertainment services for teenagers.

For example, when a user sends a selfie, or a sad expression, the chatbot is able to respond accordingly.

However, the company's development did not take off much, and the founders who struggled to break the game turned to training the NLP capabilities of chatbots, while building an underlying library to house various machine learning models and datasets, and released it publicly as an open source project.

Unintentionally, Hugging Face quickly became popular.

Today, chatbots have long been removed from the App Store, but Hugging Face is the fastest-growing machine learning library in GitHub's history, with 96,302 stars and more than 20,000 branches in its Transformer open source library. In addition, Hugging Face submissions shared 166894 trained models with 26,900 datasets.

In the more than six years since its establishment, Hugging Face's valuation has soared.

In last year's Series C financing, the company was valued at $2 billion, led by Lux Capital, followed by Sequoia Fund and others, and NBA star Durant also took a stake.

It is worth mentioning that Hugging Face only began to initially explore commercialization in 2021, and now has more than 5,000 corporate customers, including Intel, Microsoft, pharmaceutical giant Pfizer, etc., and has been in a positive cash flow state since 2021.

As for the founders of the company, all of them are Silicon Valley startup veterans.

CEO Clément Delangue founded note-taking platform VideoNot, chief technology officer Julien Chaumond, who started in Silicon Valley after graduating from college, ended his entrepreneurial career in 2015 to engage in AI development, and chief scientist Thomas Wolf is a big player in natural language processing.

Commenting on these achievements, Clément said in the Machine Learning Podcast:

Hugging Face's rapid growth is due to bridging the gap between science and production, and generating a thousand times more value by building platforms to empower the open source community and the scientific community than by building proprietary tools.

In a way, Hugging Face is building the "GitHub" of artificial intelligence as an open source platform driven by community developers.

"Back to business" Fight for open source

After understanding the development history of Hugging Face, it can be seen that this time Hugging Face launched HuggingChat, which can be said to be "returning to the old business".

HuggingChat operates on Amazon's AWS cloud service platform, allowing people to interact with Open Assistant's open-source chat assistant, both models of language systems based on the LLaMA model with 30 billion parameters.

Like Alpaca or Vicuna, these models are fine-tuned by instructions and, unlike ChatGPT, are not improved through reinforcement learning (RLHF) of human feedback.

Next, let's test HuggingChat.

At first glance, the interface is very similar to ChatGPT. On the left is a bar with the latest chat history, and most of the browser window is occupied by the current chat history. At present, users can use it directly without registration, but they cannot save conversation records.

Like ChatGPT, HuggingChat can generate natural language or formatted text on request, as can code.

Moreover, HuggingChat is very responsive.

The fly in the ointment is that HuggingChat is Chinese good at handling English than Chinese.

So, come to HuggingChat it a few questions and see how it performs?

The first is HuggingChat's "self-introduction", which can be seen that the answer is concise and to the point.

When asked about its own advantages, OpenAssistant logic has no hard wounds, and the points are relatively accurate: open source, free, and lightweight.

AS AN OPEN SOURCE PROJECT MADE BY LAION AI, ONE OF MY MAIN ADVANTAGES IS THAT IT IS FREE TO USE AND MODIFY. This means that researchers, developers, and anyone else can easily build on top of my codebase and contribute to its development. Also, since my design is lightweight, this makes it possible for people without powerful hardware to run on their own devices. Also, as a free alternative, I offer a more convenient option for those who can't afford an expensive paid model like ChatGPT.

Let's take a look at its counting ability, the results are too inaccurate.

Other abilities are yet to be measured by everyone.

What do netizens think?

Nvidia AI scientist Jim Fan tweeted:

The next step has to be HuggingChat Apps, and I think HuggingFace is in a great position to become an Android app store. In fact, HF has even an advantage over OpenAI: applications can be other multimodal models already on HF.

Some netizens pointed out that HuggingChat is very good at generating long stories.

Some netizens also said that there will be bugs when writing code.

Also, its answer is ambiguous about the winner of the 2020 U.S. presidential race.

In addition, it has made up some bizarre facts about itself.

However, Hugging Face believes that HuggingChat is not the end of the story, and eventually the platform hopes to integrate all "high-quality chat models" through a single hub.

Open and open source is the right solution?

Since the explosion of ChatGpt, the AI track has been stormy. The major tech giants are scrambling to get in and grab the lead.

In this context, Hugging Face carried the banner of "open source" and launched the open source version of HuggingChat, breaking the dominance of closed-source ChatGPT.

On the day of the HuggingChat launch, Julien Chaumond tweeted:

Some say closed APIs are winning, but we'll never give up fighting for open source AI;

Today is an important day as we launch an open-source alternative to ChatGPT.

The analysis pointed out that the Hugging Face platform provides a large number of high-quality open source models and tools, which will maximize the benefits of research and development results to the whole community, greatly reduce the technical threshold of artificial intelligence, make AI more popular, and benefit more ordinary people.

In fact, there have been calls to buy Hugging Face, and Clément Delangue said it had turned down multiple "meaningful offers" and would not sell its business as GitHub did with Microsoft.

Clément Delangue argues:

You don't need to get 100% dividends from the value you create, but you can only realize 1% of the value and keep the company running. But even just 1% is enough to make you a high-capitalization company.

Elastic and MongoDB are both open source as the core, and both have successfully developed, and recently the Fudan MOSS model has also been open source.

Regarding the future, Clément Delangue mentioned the idea of going public:

We want to be the first company to go public with emojis, not three-letter tickers.

This article is from Wall Street News, welcome to download the APP to see more