Home OpenAI launches CriticGPT to catch ChatGPT errors

OpenAI launches CriticGPT to catch ChatGPT errors


  • OpenAI launched CriticGPT to help spot mistakes in code written by ChatGPT, improving accuracy and usefulness.
  • Humans using CriticGPT for reviewing code outperformed those without it 60% of the time in an experiment.
  • CriticGPT still has limitations, like handling only short answers and discrete errors, and being prone to AI hallucinations.

OpenAI has launched a new tool called CriticGPT, a new artificial intelligence (AI) large language model (LLM) designed to assist humans in spotting mistakes in code written by ChatGPT.

In an effort to improve the accuracy and usefulness of its GPT models, OpenAI uses a technique called Reinforcement Learning from Human Feedback (RLHF), where humans review and critique output from the AI models. CriticGPT is designed to aid the AI trainers in the RLHF process.

According to OpenAI, “When people get help from CriticGPT to review ChatGPT code they outperform those without help 60% of the time,” in an experiment that showed a human AI trainer preferred critiques from a team of human-plus-CriticGPT over a human without the AI tool 60% of the time.

The major benefit of pairing a human and an AI to provide feedback in the RLHF process is that CriticGPT can write more comprehensive and thorough critiques than a human operating alone, and humans can step in when the model is reporting hallucinated errors, or when CriticGPT is reporting too many unhelpful “nitpicks”.

How was CriticGPT trained?

CriticGPT was also trained using OpenAI’s RLHF methodologies. Trainers took code written by ChatGPT and then inserted errors into the code. They then wrote example feedback that a model could provide if it had found the inserted error.

The trainer then compared output from the new model, looking at multiple critiques of the modified segment of code and seeing when the model had correctly identified and explained the error inserted.

What are the limitations of CriticGPT?

The training data so far for CriticGPT has focused on only short answers written by ChatGPT and further research and development will be required for it to address longer and more complex outputs. Additionally, it is not free from the dreaded AI hallucinations that plague LLMs.

Further work will also need to be carried out before the model can identify and critique dispersed errors – it currently can only handle discrete errors that appear in one place.

What’s next for CriticGPT?

OpenAI intends to integrate CriticGPT into its RLHF pipeline and then go on to scale the work further and increase its utility.

About ReadWrite’s Editorial Process

The ReadWrite Editorial policy involves closely monitoring the tech industry for major developments, new product launches, AI breakthroughs, video game releases and other newsworthy events. Editors assign relevant stories to staff writers or freelance contributors with expertise in each particular topic area. Before publication, articles go through a rigorous round of editing for accuracy, clarity, and to ensure adherence to ReadWrite's style guidelines.

Ali Rees
Tech journalist

Ali Rees is a freelance writer based in the UK. They have worked as a data and analytics consultant, a software tester, and a digital marketing and SEO specialist. They have been a keen gamer and tech enthusiast since their childhood in are currently the Gaming and Tech editor at Brig Newspaper. They also have a Substack where they review short video games. During the pandemic, Ali turned their hand to live streaming and is a fan of Twitch. When not writing, Ali enjoys playing video and board games, live music, and reading. They have two cats and both of…

Get the biggest tech headlines of the day delivered to your inbox

    By signing up, you agree to our Terms and Privacy Policy. Unsubscribe anytime.

    Tech News

    Explore the latest in tech with our Tech News. We cut through the noise for concise, relevant updates, keeping you informed about the rapidly evolving tech landscape with curated content that separates signal from noise.

    In-Depth Tech Stories

    Explore tech impact in In-Depth Stories. Narrative data journalism offers comprehensive analyses, revealing stories behind data. Understand industry trends for a deeper perspective on tech's intricate relationships with society.

    Expert Reviews

    Empower decisions with Expert Reviews, merging industry expertise and insightful analysis. Delve into tech intricacies, get the best deals, and stay ahead with our trustworthy guide to navigating the ever-changing tech market.