Business

OpenAI Fires Contractors for Using AI to Train ChatGPT

OpenAI has dismissed an unspecified number of contractors who used artificial intelligence tools to evaluate and train ChatGPT, a move aimed at preventing catastrophic model collapse.

Computerworld AI1 day agoBusiness
Image: Computerworld AI

OpenAI has terminated the contracts of multiple workers tasked with reviewing user prompts to improve ChatGPT. According to a report from 404 Media, these contractors were caught using generative AI tools to complete their work, directly violating the company's strict guidelines. The exact number of dismissed workers remains undisclosed, and OpenAI has declined to comment on the situation.

The contractors were explicitly forbidden from using any automated assistance. OpenAI's terms of service for these workers banned AI detection tools like GPTZero, as well as writing assistants like Grammarly and AI-powered translation software. The guidelines mandated that all feedback, comments, and reviews must be written entirely by humans to ensure genuine human input.

The strict prohibition stems from fears of "model collapse," a degenerative process where AI models trained on AI-generated data gradually lose quality and perform worse than their predecessors. One contractor told 404 Media that using AI is "pretty much the one thing that will get you kicked off ASAP" from the platform, which employs thousands of reviewers.

For AI practitioners and developers, this development highlights the critical importance of data provenance and the challenges of maintaining high-quality training pipelines. Relying on outsourced human feedback is a standard industry practice for reinforcement learning, but ensuring that humans are actually doing the work is becoming increasingly difficult. As synthetic data threatens to pollute training sets, organizations must implement rigorous auditing to protect their models from degradation.

This is our own summary of reporting by Computerworld AI

More in Business