OpenAI employs hundreds of people to analyze ChatGPT user requests. Their mission is to improve the quality of the answers ChatGPT provides to its users: people evaluate and analyze bot-generated comments. 404 Media found that many hired experts have been fired for using artificial intelligence in training OpenAI models. Here’s the irony: A company whose main goal is to encourage people to use artificial intelligence at work is firing employees who use the same technology.
Image source: unsplash.com
404 Media gained access to OpenAI’s internal documents and spoke with three contractors involved in various company projects, which could employ more than 10,000 people. By all accounts, inspectors are prohibited from using artificial intelligence in their work, and documents obtained by 404 Media confirm this.
One contractor said he saw how people “They use artificial intelligence all the time and get fired for it all the time; that’s probably the only reason they can fire you out of hand.”. According to him, “In a group of thousands of people, there are many violators.”
“Don’t use tools to detect artificial intelligence or use artificial intelligence yourself. – the document says. — Do not use GPTZero or any other AI detection tool. They are unreliable. Reviewers are also prohibited from using artificial intelligence (including Grammarly and machine translation tools) to review submissions or write reviews or comments. “
Contractors who review the work of other performers are tasked with identifying telltale signs of artificial intelligence use. These include repetition of words, idiosyncratic punctuation (e.g., excessive use of em dashes), and task completion at a suspiciously fast pace. These instructions instruct them not to report their reasons for suspecting the use of artificial intelligence: “If they know exactly what you’re focusing on, it’ll be easier for them to hide it. Evaluate the overall situation, not individual characteristics.”
One contractor said he used artificial intelligence in training OpenAI models and produced a document he called a termination notice. It said the employer had discovered the problem “Authenticity” his job. “I needed a little help, so I turned to artificial intelligence, which ultimately got me fired, – he said. “I don’t get any pleasure from my work and I don’t feel like I’ve made any contribution to society.”
Mercor hires performers for OpenAI. “We hire experts for their expertise and ability to make informed decisions – qualities necessary for the further development of artificial intelligence. Our contract strictly prohibits the use of large language models to complete projects, and we strictly enforce this requirement. We invest heavily in tools and systems to detect technical abuse and monitor experts’ compliance with project rules and contract terms. If an expert is confirmed to have used artificial intelligence to complete a task, we will immediately remove him from the project work.”” said a company representative.
There have also been reports of performers deliberately choosing the worst answers because they wanted to disrupt the model’s training process. “At first I felt guilty about doing this kind of work; – one of them admitted. — I either choose randomly without looking at the results, or I choose deliberately. [худший] options. I have no idea how important this is, since the neural network’s results are evaluated by hundreds of other people, but I still feel like I’m being paid to make the AI worse. “
Experts point out that some AI models have begun to show signs of so-called “model collapse,” in which the quality of the work of AI trained on text created by other AIs gradually deteriorates. Ironically, the people hired to prevent such problems themselves train OpenAI models using the responses generated by the AI.
If you find an error, select it with your mouse and press CTRL+ENTER.
