Coinworld.com Report:
You may be familiar with the popular ChatGPT. You may also know of other impressive chatbots like Claude, Reka, or Yuan AI. But you may have missed out on a very adorable chatbot: one that is free, open-source, and potentially more capable than ChatGPT in specific tasks such as image generation, document parsing, video editing, and more.
Hugging Chat is the open-source alternative to ChatGPT. It comes from the team at Hugging Face, a platform that has gained popularity among open-source AI researchers and developers. If you’re looking for a conversational AI that doesn’t have paywalls, complicated account registrations, or concerns about data usage, then Hugging Chat is the perfect solution – did we mention it’s free?
In contrast to popular skepticism, Hugging Chat allows users to choose their text generation capabilities from a variety of open-source AI models. It supports eight of the best open-source LLMs, including Llama 3.1 70B, Nvidia’s Nemotron 70B (which outperforms GPT-4o and Claude 3.5 Sonnet in certain benchmark tests), Llama 3.2 Vision, and Cohere’s Command R Plus (designed specifically for conversational interaction and long-context tasks).
Feature Showdown: Hugging Chat vs. ChatGPT
At first glance, Hugging Chat and ChatGPT may seem to offer similar functionality. They are both designed for various scenarios, ranging from casual conversations to in-depth technical support. But upon closer inspection, there are some notable differences.
Variety of Available Models
The first thing I noticed in my analysis is how versatile these two chatbots are in terms of text capabilities. With Hugging Chat, users can choose from eight different LLMs built by different developers, considering various technologies and use cases. While GPT-4o may excel at everything, users have the option to choose a model that is specifically tailored to their needs; for example, Nemotron excels at complex reasoning tasks, while Llama 3.1 is better at creative writing.
On the other hand, ChatGPT Plus users can only choose from OpenAI’s models based on the company’s guidelines and writing style. Currently, users can only choose between GPT-4o (more creative) and OpenAI o1 (more suited for inference skills).
Assistants: OpenAI GPT’s Open-Source Version
Another similar feature is the assistants. Think of them as custom GPTs from OpenAI, but more flexible. These assistants act as specialized agents for different tasks, ranging from casual conversations to more niche AI queries.
Hugging Chat is more versatile than ChatGPT and allows you to add impressively useful tools. You can have your chatbot browse the web, retrieve URLs, generate images using Flux (considered a top-tier image generator currently available), clone voices, or parse RAG documents (retrieval-augmented generation).
Meanwhile, ChatGPT’s custom GPT is great for its simplicity. With just simple language, you can adjust your personalized AI assistant, and ChatGPT will customize the agent to perform well in terms of performance.
Considering this, you can reward this point based on your user type and what you want to achieve:
– If you’re a beginner who wants to use existing agents, Hugging Chat wins as it offers many highly functional choices.
– If you’re a beginner who wants to create your own agent, ChatGPT wins as you can essentially have a natural language conversation with your chatbot, making it understand what you want the agent to do.
– If you’re an expert user who is capable of coding but just wants to play with custom agents, then both Hugging Chat and ChatGPT offer quite a lot of choices. So, in this case, it’s a tie.
However, technical experts seeking to build custom AI assistants will find Hugging Face’s approach to have a clear advantage. The open-source framework allows developers to directly modify the agent’s code, adjusting specific parameters that ChatGPT does not allow users to modify, such as temperature, top-P, top-K parameters, and penalties. While ChatGPT users have to rely on prompts and hope the system interprets their intent correctly, Hugging Chat developers can precisely tune the assistant’s behavior by accessing and modifying underlying parameters.
Community Tools: Steroids Your Chatbot Didn’t Know It Needed
Hugging Chat’s unique feature lies in its community tools, which significantly enhance its capabilities by allowing users to directly plug in various tools into the Language Model (LLM). This feature enables users to create and use tools from the Hugging Face Hub – a community-developed repository of custom models, making it easier to integrate different functionalities into the chatbot experience.
Therefore, for example, users can configure their chatbot to do many things beyond text generation. Users can enable over 30 different tools.
Some of the most interesting ones include:
– URL Extractor for retrieving text content from specific URLs
– Document Parser for parsing content from PDF files (also known as RAG or retrieval-augmented generation)
– Image Generator using Flux – considered a killer open-source MidJourney
– And other fun ones like an image editor, voice cloner, photo background remover, chart analyzer, Python code tester, and strangely enough, a Roblox 3D asset generator.
OpenAI behaves differently. Instead, it embeds everything into its own interface, where users can activate specific functionalities based on the models they call. For example, they do provide a document parser available with GPT-4o, but not with OpenAI o1, and they can generate images due to native integration with DALL-E 3, something Hugging Chat doesn’t support.
In conclusion, Hugging Chat is more versatile, but OpenAI’s approach provides a more streamlined experience. Users don’t need to activate anything; they just have to ask their chatbot to do something, and it will respond accordingly (or decline).
How to Use Hugging Chat
Good news: using Hugging Chat is very straightforward. It’s free (did we mention that?), it’s open-source, and you don’t even need to create an account – although you can if you want to save conversations. Hugging Face claims an account is required, but we managed to log in without one, no problem. It’s that simple…though there will be some limitations in terms of usage.
All you need to do is visit huggingface.co/chat, choose a model from the list of open-source AI options, and you’re ready to go. The interface? It’s as intuitive as it gets. You have an input text box, and that’s it – no need to navigate clunky menus or hidden features. You type, Hugging Chat responds, and the conversation flows effortlessly.
However, creating an account isn’t a bad idea. It will allow you to access new models that can be run locally if you enjoy playing with open-source models. Plus, if you like what you see, you can get a Pro account for $9 per month – almost half the price of a ChatGPT Plus subscription – and support the project. It comes with additional benefits like early access to upcoming features, 5x usage quota, higher GPU queue priority, some social media perks, and a cute badge to proudly display on your profile picture.
By default, Hugging Chat starts with Llama-3.1-70B-Instruct. To use a different model, simply click on the control panel next to the “Current Model” tab.
It will open a popup window that allows you to select a new model on the left side and start a new chat by clicking a button. This button is located on the right side of the window.
Once you click the button to start a new chat, the popup window will close, and the UI will return to the initial screen with the chatbot ready for a new interaction.
That’s it. Now you can chat with your model.
Advanced Feature: Using and Creating Assistants
To select a new assistant, users must choose the option on the lower-left corner menu. Once there, a new section will open, allowing users to choose their preferred assistant.
Remember, some assistants may be incompatible with all LLMs, so be sure to explore the assistants and then choose a compatible LLM or simply select your model and choose an assistant from the compatible and available pool of options.
To do this, simply click on the box below the “Assistant” heading and select your desired assistant.
If you want to dive deeper into building your custom assistant – essentially an open-source version of GPT – simply click the “Create Assistant” button in the top right corner. This will open a window similar to the OpenAI custom GPT interface:
Here’s what each field does:
– “Avatar” is a thumbnail that showcases your character. Unlike ChatGPT, you can’t ask the chatbot to create one for you; you need to manually upload a photo.
– “Name” is basically what you want your assistant to be called.
– “Description” is a brief description of your assistant’s programming capabilities. This will help other users in the community understand if your creation meets their needs.
– “Model” is an area to pay attention to. Essentially, the LLM will power the behavior of your assistant. Make sure to choose a model that suits the task you envision for your assistant – so if you want an coding assistant, you might want to use something like Nemo. If you want to deal with image analysis, choose Llama 3.1 Vision, and if you just want quick document summaries, choose Phi.
– “User Begin Message” are just examples of how you can start interacting with the assistant. For example, you can write “Explain this as if I were five years old” or “Generate a detailed description of a realistic image based on this setup.”
– You also have the option to allow your assistant access to the internet. This has been an advantage for ChatGPT, as its custom GPTs work offline.
If you choose to allow your assistant access to the internet, you must choose one of three options:
– Web Search will basically scour the entire web, just like it’s doing a Google search for your task.
– Specific Domains will perform web searches based on certain domains. For example, if you only want the best information about cryptocurrencies and emerging technologies, you can limit news results to Decrypt.co, which focuses on cryptocurrency news.
– You can also choose to search for information from specific links, so it will retrieve information from the URL list you provide.
By default, your assistant is set to work offline only.
“Custom Instructions” or system prompts are the most important part of the configuration, where you place all the information and instructions that will shape your assistant’s behavior.
Unlike ChatGPT, you can set additional advanced features such as temperature or penalties to determine the model’s creativity when speaking and how much effort it puts into selecting different words.
Boosting Hugging Chat’s Potential with Tools
The community tools themselves are a league of their own, giving Hugging Chat a real advantage in terms of versatility compared to competitors.
To use the tools, simply click the button on the bottom left and select the desired functionality. These are not model-specific, so don’t worry about which model to choose.
Once you select the desired tools, simply click the button to return to the main UI and start a new chat. It’s located next to the Hugging Chat logo on the top left.
Now, when you click on available tools, choose which tool to use for a specific interaction. When done, unselect and continue chatting with your AI chatbot as usual. Note that some tools may take some time – possibly too long – and some tools may have errors. So be prepared for potential hiccups.
Our Verdict
Unlike its competitors, Hugging Chat doesn’t lock features behind paywalls. Whether you’re a developer or just someone looking to avoid ChatGPT fees, Hugging Chat offers an enticing package without any corporate nonsense. It’s free, flexible, and once you get the hang of it, easy to use – a no-brainer for anyone who wants AI without being controlled or restricted by closed software.
However, as with all things open-source, you have to be prepared to get your hands dirty.
Hugging Chat is a strong contender. If you’re tired of locked features and outdated training data, give this open-source alternative a try. It may not beat Claude or ChatGPT in terms of creativity or simplicity, but it’s a great choice worth emulating and exploring (did we mention it’s free?).
Editor: Andrew Hayward
