From ChatGPT to OpenChatKit: Transforming Conversational AI for Everyone

Recently we covered an announcement about an open source alternative to ChatGPT called OpenChatKit released by Together. OpenChatKit is based on large language model (LLM) and its model is fine-tuned with 43 million high-quality instructions for chat from EleutherAI’s GPT-NeoX-20B.

OpenChatKit model is already tuned for conversational interactions. You can select various use cases for which the model is fine-tuned. These use cases are: Extraction, Summarization & Dialogue, Questions & Answering, Sentiment Analysis, Email Summarization, Travel Advice, Bash Scripting, Writing short letters, Financial questions, and more.

Even though OpenChatKit is decent enough for most of the general use cases, there are some of the points where OpenChatKit model is still lacking behind the ChatGPT, but with community efforts, it is very much possible to fix all these gaps so to take it to the level of ChatGPT. We need good community support to mature more such open source AI tools. Let us look into these gaps in detail below:

Coding Assignments: 

The chatbot's coding abilities could be enhanced with a more extensive collection of source code for training purposes. It requires a supplementary datasets that can be contributed to enhance its performance.

Occasional Repetition: 

At times, the chatbot may repeat its response. Although its performance is being enhanced, you can currently initiate a new conversation by clicking on the refresh button.

Difficulty in Context Switching: 

In the event that you alter the subject matter during a conversation, the chatbot may struggle to transition automatically and may continue to provide responses that pertain to the previous topic.

Generating Creative and Lengthy Text: 

Unlike ChatGPT, the chatbot's abilities to produce extensive and imaginative written content, such as an essay or story, are limited at this point of time.

Closed-Ended Questions and Answers Based on Knowledge: 

The chatbot's responses to such queries may sometimes be inaccurate. Cross-checking the information is advised, and wherever possible, provide the correct feedback. Additionally, reinforcement of the retrieval system to enhance its performance in this regard is a work in progress.

Together is committed to delivering cutting-edge conversational AI technology OpenChatKit, which can seamlessly integrate with various applications and platforms. It's an open source technology accessible to everyone, and work is going on constantly to enhance its performance and expand its capabilities. Experience the power of OpenChatKit for yourself here.