OpenAI Releases CriticGPT Model Aimed At Identifying Inaccuracies In ChatGPT Responses

Artificial intelligence research organization OpenAI introduced CriticGPT, a new model based on GPT-4, designed to identify and correct errors in the output generated by ChatGPT‘s code.

The new model is specifically trained to generate critiques that pinpoint inaccuracies in the answers provided by ChatGPT. It is among the GPT-4 series of models that fuel ChatGPT, and focus on enhancing its utility and interactivity through a method called “Reinforcement Learning from Human Feedback” (RLHF). This approach involves the collection of AI trainer ratings of various ChatGPT responses against each other.  

As large AI models like ChatGPT advance in reasoning and behavior, their accuracy improves, and errors become more nuanced. This evolution poses challenges for AI trainers tasked with identifying inaccuracies during RLHF comparisons. Without enhanced tools, evaluating these systems can be challenging. OpenAI emphasizes that the CriticGPT model aims to address this by providing a means to evaluate outputs from advanced AI systems.

According to the company, when users assign ChatGPT tasks such as “write a specific Python function,” ChatGPT will generate code accordingly. Meanwhile, CriticGPT, when evaluating one of the generated codes from ChatGPT, may suggest a better alternative.

However, OpenAI noted that CriticGPT’s suggestions are not always accurate. Nevertheless, it can still assist AI trainers in identifying more issues in model-generated answers than they could without AI assistance. Moreover, when used in conjunction with human effort, CriticGPT enhances users’ abilities, leading to more thorough critiques and fewer hallucinated errors compared to when the model operates independently.

OpenAI Joins Forces With Oracle Cloud Infrastructure To Bolster AI Model Development 

OpenAI is dedicated to research and development in the field of artificial general intelligence (AGI) and generative models, and is well-known for its chatbot, ChatGPT. At present, the firm is reportedly in the process of developing its latest version, ChatGPT-5.

Recently, OpenAI has partnered with Oracle Cloud Infrastructure (OCI) to enhance its capabilities. Through this collaboration, it will leverage the OCI Supercluster for the training and inference of its next-generation AI models.

The post OpenAI Releases CriticGPT Model Aimed At Identifying Inaccuracies In ChatGPT Responses appeared first on Metaverse Post.