The week was buzzing with announcements from Google AI Workspace and Microsoft Copilot 365—rightfully so, as I’m eager to see how those tests unfold! However, another intriguing development emerged: a new, fully open-source, testable trained model called OpenChatKit, available on GitHub.
OpenChatKit is an open-source project released under the Apache-2.0 license, designed for developing both specialized and general-purpose chatbots for various applications. Essentially, it’s a conversationally tuned large language model fine-tuned from GPT-NeoX-20B with 43 million instructions. The strengths of OpenChatKit include:
- Summarization and contextual question answering,
- Information extraction, and
- Text classification.
According to its creators, there are still areas in need of refinement:
- Knowledge-based closed question answering,
- Coding tasks,
- Context switching, and
- Creative writing or longer responses.
One of the most interesting aspects for me is that the project features an extensible retrieval system, which allows the chatbot to incorporate regularly updated or custom content into its responses. This means it can continuously pull in fresh information from specified sources (like Wikipedia or some business data source), keeping its knowledge up to date.
You can give it a try on Hugging Face (I’ve also added the link to the AI Tools section). Here, you’ll find tools for parallel testing across multiple LLMs!
Source: Together
