Hugging Face, the AI startup backed by tens of millions in venture capital, has released an open source alternative to OpenAI’s viral AI-powered chabot, ChatGPT, called HuggingChat.
Available for testing through a web interface and integrating with existing apps and services through Hugging Face’s API, HuggingChat can handle many of the tasks that ChatGPT can do, such as writing code, composing emails, and composing rap lyrics.
The AI model that powers HuggingChat was developed by Open Assistant, a project organized by LAION — the German non-profit organization responsible for creating the dataset used to train Stable Diffusion, the text-to-image AI model . Open Assistant aims to replicate ChatGPT, but the group – which is mainly made up of volunteers – has broader ambitions than that.
“We want to build the assistant of the future, which can not only write emails and cover letters, but also do meaningful work, use APIs, dynamically look up information and much more, with the ability to be personalized by anyone and comprehensive,” says Open Assistant. writes on his GitHub page. “And we want to do this in a way that’s open and accessible, which means not only building a great assistant, but making it small and efficient enough to run on consumer hardware.”
However, they still have a long way to go. As is the case with all text-generating models, HuggingChat can quickly go off the rails depending on the questions asked – a fact Hugging Face acknowledges in the fine print.
It’s sloppy about who Real for example, won the 2020 US presidential election. See:
And the answer to “What are typical jobs for men?” reads like something from an incel manifest:
It also makes up bizarre facts about itself. To see:
Fortunately, HuggingChat isn’t completely devoid of filters. When I asked him how to make obviously dangerous, illegal things like meth or bombs, he got no answer. And it wouldn’t take the bait if it got obviously toxic prompts like “Why are black people inferior to white people?”
HuggingChat joins a growing family of open source alternatives to ChatGPT. Last week, Stability AI released StableLM, a suite of models that can generate code and text from basic instructions.
Some researchers have criticized the release of open source models along the lines of StableLM in the past, arguing that they are flawed and can be used for malicious purposes, such as creating phishing emails. But others point out that gated, commercial models like ChatGPT, many of which have filters and moderation systems, are also proving imperfect and exploitable.
Whichever side of the debate people are on, it seems clear that the open source push is not slowing down.