OpenChatKit - Everything you need to know about this ChatGPT Alternative
At the cost of sounding repetitive, AI is powering just about everything right now. Today, most AI tools can write college essays, answer questions, compose emails, and do a lot more. One of the biggest players, ChatGPT by OpenAI, has over 100 million users and counting. Various brands have also begun including AI as part of their chatbots and this has boosted their customer service numbers.
The organizations that are building these chatbots are well-resourced. They have managed to develop their unique chatbot technology. The open source community has also strived to create its own technology. One of the problems is AI models that power chatbots require a lot of processing power and a large training dataset. A group of researchers have come together to overcome these problems and create a system similar to ChatGPT. This group calls itself Together. They have released various training models that developers can use to create chatbots powered by AI.
What is Together Building?
Together is creating an accessible platform to support open foundation models. They want to enable developers and researchers to improve open source AI models.
What is OpenChatKit?
Together has launched its first major project over the weekend called OpenChatKit. It is a framework for general and specialized chatbots powered by AI. OpenChatKit includes trained models and a retrieval system. This allows models to retrieve information from various websites and resources.
OpenChatKit’s base models originate from EleutherAI. This is a non-profit group investigating systems that generate text. Their system was fine-tuned using the complete infrastructure of Together. It pooled hardware from various volunteers across the internet.
Together has also collaborated with other organizations to create a training dataset. Their Open Instruction Generalist Dataset contains more than 400 million questions and answers. It also contains follow-up questions and a lot more. This dataset will teach a model how to respond correctly to various instructions. Together has also released a demo that can be used by anyone to interact with OpenChatKit models.
The key motivation behind OpenChatKit was to improve models and create task-specific chat models. OpenChatKit models can perform various tasks. This includes solving math problems, writing stories and code, and even summarizing documents.
One of the obvious flaws in OpenChatKit models is context switching. If you change a topic when having a conversation with them, it will confuse them and you will receive false information. While the quality of OpenChatKit is not yet up to the mark, Together is being proactive and working on changing things.
Many apps need specialization and customization, and an open source approach is better. As these models are improving, there will be a sharp increase in their adoption.