OpenAI has launched CriticGPT, powered by GPT-4, to help identify errors in code generated by ChatGPT. The tool assists AI trainers in spotting mistakes more efficiently, although it is not 100% accurate and faces challenges like handling complex tasks and occasional hallucinations. OpenAI plans to integrate CriticGPT into their RLHF labeling pipeline to enhance ChatGPT’s reasoning abilities and reduce errors. The tool provides critiques to highlight inaccuracies in ChatGPT responses, improving trainers’ error identification process. While CriticGPT is still a work in progress, OpenAI aims to address its shortcomings and enhance its capabilities for GPT-4 training. Additionally, Oxford researchers have used semantic entropy to assess AI-generated outputs for quality and hallucination detection. As AI models advance, NVIDIA’s CEO suggests that coding may not be a viable career option in the future, pointing towards alternative fields like biology, education, manufacturing, or farming.
Source link
Source link: https://www.windowscentral.com/software-apps/nvidia-ceo-might-be-right-about-coding-being-dead-because-of-ai-openais-new-criticgpt-model-identifies-chatgpts-programming-mistakes-better-than-ai-trainers
OpenAI introduces CriticGPT to detect ChatGPT’s coding errors. #AI
