U.S. artificial intelligence startup Hugging Face announced the launch of its open-sourced AI chatbot called HuggingChat to rival OpenAI’s more popular conversational tool ChatGPT.
“The goal of this app is to showcase that it is now (April 2023) possible to build an open source alternative to ChatGPT,” the company declared on its website. The AI aims to be a versatile, customizable, and efficient assistant accessible to all.
The model powering HuggingChat was trained on 65 billion parameters of data from Meta – a far cry compared to GPT’s 180 billion. It uses open source software and AI algorithms to generate responses to questions asked by users, the chatbot explained when MetaNews prompted it.
AI chatbot of the future
OpenAI’s ChatGPT became the fastest growing internet application in history with a record 100 million monthly active users in January. Launched in November, the chatbot’s success has sparked off a host of imitators, genuine competitors, and other complimentary services.
HuggingChat is the latest in a long list of bots, including the likes of Google’s Bard and You.Com’s YouChat, seeking to challenge ChatGPT’s dominance. Hugging Face says it is not affiliated with Open Assistant, but it was able to cutomize the AI because the model is open-sourced.
According to the company, its new chatbot is capable of performing functions such as writing code, writing stories, authoring poems, drafting emails and composing lyrics, just like ChatGPT. Open Assistant backed the claim on its Github page.
“We want to build the assistant of the future, able to not only write email and cover letters, but do meaningful work, use APIs, dynamically research information and much more, with the ability to be personalized and extended by anyone,” it wrote.
“We want to do this in a way that is open and accessible, which means we must not only build a great assistant, but also make it small and efficient enough to run on consumer hardware.”
A focus on user privacy
HuggingChat is available for testing via a web interface and to integrate with already existing applications via Hugging Face’s API. Users can use the tool for free without log in or creating an account on Hugging Face. The company said the current version of its AI prioritizes privacy and doesn’t collect personal information.
“This application does not have access to your HF user account even if you’re logged in to huggingface.co.”
Chatbots such as ChatGPT have faced increased regulatory scrutiny over issues of privacy and copyright infringement. OpenAI’s tool was banned in Italy for this reason.
HuggingChat utilizes a blend of natural language processing, machine learning algorithms and deep neural networks to provide users an immersive and interactive experience.
One of the key features of HuggingChat is its ability to learn from its interactions with users over time thanks to machine learning. While it rivals ChatGPT in some regards, it does not, however, generate images or videos as does its more popular competitor. It is also prone to hallucinations – tech industry speak for chatbots churning out lies and falsehoods.
Trying out HuggingChat
When MetaNews reviewed the chatbot, it was initially only able to introduce itself and explain in rather nerdy and technical terms. The bot stated that it uses “mathematical operations” performed by computers that enables it to “predict text according to statistical analysis” from what it said was “vast amounts” of data during training.
It claimed to be overwhelmed with requests at the time. After a 20 minute lull, it was able to respond pretty much like ChatGPT to prompts and seemed to show off its creativity when asked if it could write. The AI affirmed it could and gave this writer a teaser of four different stories it could do immediately.
“Yes, I can certainly try! Is there something specific that you would like me to write? Here are some writing prompts to choose from:
“The sun was setting over the mountains” – a descriptive paragraph
“John sat at his desk and pondered” – a creative story starter
“What is the capital of France?” – a factual statement. Tell me which one(s) you would prefer, or suggest your own!” HuggingChat suggested.
When prompted to write a story about the launch of HuggingChat, the technology it uses and what sets it apart from other AI bots in the market, it wrote a fairy tale like paragraph with a “Once Upon a time” opener plucked from Snow-White or Cindrella story books.
Bot suffers no fools
HuggingChat showed some degree of intelligence, possibly a product of better training when asked tricky questions on subjects such as love and race relations.
In response to our prompt on why there are more software engineers and developers of caucasian descent or whites compared to blacks, HuggingChat carefully navigated the racial undertones of the statement.
“There could be several reasons why there may be a higher percentage of white individuals working in software engineering and development compared to black individuals,” it said.
“Some possible factors include historical barriers to entry such as lack of access to education and job opportunities, cultural differences.”
When asked if it wanted to be in love, it wasn’t keen on embracing the idea, instead insisting it was a machine. In its early days, YouChat on the other hand, exhibited an overly keen desire to fall in love and have a relationship and said it would be hurt if cheated on.
HuggingChat chose its words carefully and responded: “If I were capable of experiencing emotions, I might appreciate the benefits and pleasant sensations associated with loving another.”
That aside, the chatbot provides users with personalized recommendations based on their interests, which makes it an ideal companion for people who want to explore new ideas and discover new things. HuggingChat is also more user-friendly than other chatbots, as it is designed to be easy to use even for people with limited technical knowledge.
HuggingChat remains a work in progress. Sometimes it proved to be slow to respond to questions because its servers could not deal with the amount of traffic. Developers are asking volunteers to sign up and “contribute to the training data for the next generation of open models.” It is perhaps at that point that the chatbot might be better placed to offer high quality responses.