OpenAI fires contractors for using AI
Oh, the irony.
Image by Cybernews
Follow us
- OpenAI reportedly fired contractors who used AI while reviewing data meant to improve ChatGPT.
- The company bans contractors from using large language models, Grammarly, and AI translation tools for this work.
- OpenAI uses human reviewers to help prevent AI models from learning from synthetic data.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
OpenAI doesn’t want its contractors to use AI when training AI, as it defeats the purpose of making it more human-like.
OpenAI, the creator of ChatGPT, hired a number of contractors whose job is to read real users' AI prompts and other content to improve the AI agent’s answers, making them sound more human.
It was revealed that the company fired some contractors after discovering they had used AI to train the models.
The need for human revision helps AI models to stop them from “model collapse,” an AI learning process during which the models learn from synthetic data generated by other models.
This process causes new language models to degrade and produce lower-quality answers.
OpenAI hired people who would halt this process. However, by using AI-generated information themselves, the workers did the opposite.
Using AI to train AI is something the company will immediately fire a person for, a contractor told 404Media, adding that many contractors have been caught doing this.
It was recently reported that OpenAI hired contractors for its “Project Lily” to read real interactions between ChatGPT and its users.
Even without seeing usernames or other personal information, workers could still access the user’s memory summary.
No AI-based tools
The documents obtained by 404Media revealed that contractors hired to train AI are banned from using AI for their work.
The document for contractors hired to review other employees' work also stated that they should refrain from using AI detection tools because “they are not reliable.”
It was also specified that the contractors also can’t use Grammarly and AI translation.
Stay updated with our latest stories and follow us on social media
Be the first to discover new stories, ideas, and updates from our team.
One contractor who “turned to AI” for help was later presented with a termination letter stating that their employer had found issues with the “authenticity” of their work.
404Media spoke with 2 contractors employed by Mercor, an AI training company. It employs contractors who review material for ChatGPT.
A Mercor spokesperson confirmed that contractors are prohibited from using LLMs and that, once the company is sure a contractor used AI to complete their work, they are removed from the project.
One of the contractors interviewed by the outlet revealed that sometimes they would try to sabotage the AI training by choosing the worst responses.
