OpenAI has an armed forces of contractors who peruse genuine ChatGPT users’ prompts and another data to assistance enhance the chatbot’s responses. The idea is that the contractors provision an, obviously, individual contact to OpenAI’s models. Well, not all of the contractors are doing that. 404 Media has established multiple contractors hired to enhance OpenAI’s models have been fired for using AI to train the AI. That’s not awesome for the models themselves, but there is additionally evidently a awesome irony in AI training companies operating for OpenAI firing group for using AI whenever OpenAI’s entire item is to create group use AI at work.
Some AI models already display signs of “model collapse,” which is anywhere AI models additional trained on AI-generated content can rotate into worse and worse. In this case, several of the group hired to partially halt that happening are themselves using AI-generated responses to train OpenAI’s models.
One contractor stated they see group using AI “all the period and group are let go for it all the time, it’s beautiful much the one item that volition get you kicked off ASAP.” The individual said, “in a collection of thousands there are lots that have been caught.”
💡
Do you activity as a immediate reviewer for OpenAI, Anthropic, or another AI company? I would affection to comprehend from you. Using a non-work device, you can communication me securely on Signal at joseph.404 or dispatch me an email at [email protected].
Last week, 404 Media revealed Project Lily, in which OpenAI has hundreds of contractors study genuine ChatGPT users’ prompts and conversations which can contain individual information. Those contractors afterward charge and critique the responses ChatGPT generated, including making certain that the responses are not too sycophantic or anthropomorphize ChatGPT. That reporting was according to inner documents and conversations alongside a individual who plant on the prompts.
404 Media has now obtained another inner documents and spoken to three contractors doing activity for OpenAI throughout assorted projects. Those projects can contain additional than ten thousand contractors, according to among the inner documents.
One of those documents says contractors must not use AI themselves for their work.
“Do not use AI finding tools, or AI yourself,” one document describing the activity of contractors who are hired to assessment the activity of another contractors, including catching them for using AI, says. “Do not use GPTZero or any another AI finding tool. They are not reliable. Reviewers may not use AI either, including Grammarly and AI translation, to review, compose feedback, or compose comments.”
The document continues, “Do not inform evaluators why you doubtful AI. It is easier for them to conceal if they cognize what you appearance for. Judge the general pattern, not one clue.”
All three of the contractors stated reviewers are told not to use AI in their work. Two of the sources stated group have been fired or offboarded for using AI. 404 Media granted the contractors anonymity as they weren’t permitted to conversation to the press.
The contractors who assessment another contractors’ activity are told to be on the appearance out for tell-tale signs of AI use. That can contain repetitive words, AI-style symbols — which power contain complete zealous use of the em dash — and contractors finishing their activity extremely quickly.
In connected Slack channels anywhere group ask all another for advice, a lot of group volition article an example alongside the question, ‘Is this AI?,’ one contractor said.
“Usually the answer is yes,” the individual said.
One contractor stated they used AI during assisting to train OpenAI’s models and shared what they presented as their ending letter. It stated their director had identified issues alongside the “authenticity” of their work.
“I’m not a bad individual or worker. I fair needed a small boost and turned to AI to assistance me which eventually led to my downfall,” the contractor told 404 Media. “I felt no joy in the activity or that I was contributing to community in any way.”
Two of the contractors 404 Media said to worked for Mercor, an AI-training business that hires the contractors who in rotate assessment ChatGPT-related material. A Mercor spokesman told 404 Media in a statement: “Our experts are hired for their ability and judgement, which is essential to the ongoing advancement of AI. Our contracts strictly prohibit the use of LLMs to complete projects and we enforce that. We allocate heavily in our tools and systems to detect misuse and justify our experts comply alongside project rules and agreement terms. When we verify an expert has used AI to complete a task, we immediately eliminate them from the project.”
404 Media said to a fourth contractor who has worked on training models for assorted AI companies. They stated they sometimes purposefully chose the worst responses since they wanted to actively sabotage the models’ training.
“I did awareness blameworthy concerning doing this benevolent of activity at the start,” they said. “I either pay zero notice to the results and choose randomly or purposely choose the [worst] output. I’m not sure how much of a difference it really makes since there are hundreds of another group additionally ranking immediate results, but it does awareness akin I’m getting paid to create AI worse.”
OpenAI declined to comment on its contractors being fired for using AI.
About the author
Joseph is an award-winning investigative newsman focused on generating impact. His activity has triggered hundreds of millions of dollars value of fines, close downward tech companies, and much more.
