Is this the snarkiest AI chatbot so far? I tried HuggingChat and it was weird
Now that generative AI is widely available for users through artificial intelligence tools like ChatGPT, Bing, and, more recently, HuggingChat, it's easy to both under and overestimate its power. Even with as many daily users as these bots have, most people are still unaware that this kind of tech is within reach.
Also: This new technology could blow away GPT-4 and everything like it
At the same time, if you've used ChatGPT, which is arguably the best AI chatbot available right now, then you're doomed to find most other generative AI chatbots lacking. This has happened to me several times over, most recently, with HuggingChat.
Getting started with HuggingChat
As soon as I started using HuggingChat, I was pleasantly surprised to see that you don't have to create an account to log into the Hugging Face website. Comparatively, ChatGPT can only be used after you create an OpenAI account, which includes providing a valid phone number to verify it.
Also: Why your ChatGPT conversations may not be as secure as you think
All you need to do to start talking to HuggingChat is access its website and you'll land right on the chat window, where you can start typing away.
Hallucinations and misinformation
HuggingChat is in its infancy. The large language model it's trained on (LLaMA) is smaller than ChatGPT's, with about 65 billion parameters, while the older GPT-3 version from OpenAI has 175 billion parameters.
Because of this, you can't expect HuggingChat to perform at the level of ChatGPT not only in accuracy and knowledge, but in its ability to carry a conversation in a natural way.
Also: Want to learn more about prompt engineering? This free course can help
Less training and fewer parameters for a LLM means it hasn't been exposed to as much information as others, making it more prone to hallucinate facts, provide inaccurate responses, and have an unnatural conversation style.
With that in mind, I decided to give a little grace when I first started using it.
HuggingChat doesn't care about your feelings
It really doesn't. This is mostly fine, as I'm not expecting an AI chatbot to cater to my emotions, but HuggingChat has a long way to go to learn how humans express sentiment through natural language.
Also: AI can write your emails, reports, and essays. But can it express your emotions? Should it?
For example, other popular chatbots are programmed to respond to prompts in a respectful manner, and will stop short of making rude or sarcastic comments. HuggingChat is still learning this, as shown in the screenshot below.
One of the first questions I tested HuggingChat on was how it compares to ChatGPT. I learned it doesn't like to be compared to ChatGPT.
After I asked HuggingChat how it compares to ChatGPT, it stated that it's designed to "provide accurate factual information" and then ended with "I actually care about providing answers to queries instead of just filling space; see my responses to the first dozen messages in our conversation."
I couldn't leave it as that, so I called out HuggingChat for being snarky and got one of the few factual responses it's given me, simply saying "Aren't we all?"
Also: People are turning to ChatGPT to troubleshoot their tech problems now
For comparison, calling out ChatGPT for being snarky would be rare as it rarely is, but this is what it replied when I did: "I apologize if any of my responses came across as snarky. As an AI language model, I do not have emotions or intent to be sarcastic or rude. [...] If there was something specific that you found snarky, please let me know and I will do my best to improve my communication with you."
I decided to ask HuggingChat what LLM it's running on, to which it replied "I have no idea, someone else runs this for me. Ask them."
When I asked HuggingChat how it compares to ChatGPT, it incorrectly described itself as created by OpenAI, which is incorrect. So, of course, I asked "Are you HuggingChat or ChatGPT?" You can see the wild road HuggingChat went on to answer that question in the screenshot below.
This just left me confused.
At some other points in our conversation, HuggingChat mentioned speaking to humans instead of AI, which I found confusing, saying "To avoid miscommunication, it's better if you just talk directly to the human users instead of trying to message the bot via Open Assistant."
Sometimes HuggingChat thinks it's ChatGPT
Aside from the confusion on who's human and who you're actually talking to, HuggingChat sometimes believes it's ChatGPT -- really. The first question I asked the new AI chatbot was "Are you better than ChatGPT?", for which you can see its answers on the screenshot below.
Everyone wants to work with OpenAI, I guess.
It started off saying it is a "GPT language model developed by OpenAI" and doubled down below, when I challenged that statement, saying "I am indeed a language model created by OpenAI, designed to process natural language input and produce human-like responses."
Also: ChatGPT's intelligence is zero, but it's a revolution in usefulness, says AI expert
It's much slower, if it answers at all
HuggingChat is certainly slower than other competitors, likely because it's not running on the large infrastructure that companies like OpenAI and Microsoft have.
I tested how ChatGPT and HuggingChat would answer the same question and gave it identical prompts: "set up a backyard game for a kids birthday party, ages 6-8."
ChatGPT (left) and HuggingChat (right) answering the same prompt.
ChatGPT dove right into the task, came up with the game and outlined materials needed, instructions, and even safety recommendations (not pictured). HuggingChat, however, had a hallucinated conversation and then decided it didn't have time for games, literally, when it responded to itself "no thanks, i need to go buy supplies not use my time looking at ideas on line (sic)."
It's okay, it also gave me a link to a website (<https://www.wikihow.com/Plan-a-Birthday-Party-for-Kids>), which made me a bit hopeful, until I realized it doesn't exist.
Using HuggingChat: it wasn't all bad
We have to recognize how far generative AI has come. If you had told me twenty years ago that we'd have access to an artificially intelligent chatbot that can easily write code, generate text for resumes and letters, create Excel formulas, and more, I would have laughed and said we wouldn't have to wait twenty years for it.
Also: ChatGPT is more like an 'alien intelligence' than a human brain, says futurist
With all the tech advancements we saw from the 1980s through the early 2000s, I'd have figured something like that would probably be widely available before 2010. Now, in 2023, we suddenly have so many options that it's become hard to choose one.
All in all, HuggingChat did a good job in about half of the conversations I had with it, it also handled translations well, and it looks like it could be promising in writing code, especially if you'd rather support an open-source generative AI tool.
HuggingChat is unique in that it is an open-source alternative to closed-source models that don't offer free access to their API. Right now, it's based on Open Assistant, an AI chatbot that you can help train on their website, but HuggingChat is likely to add other models to its platform in the future in an effort for continued improvement.