31 ChatGPT training dispatch workers were fired after failing to meet expectations

On April 23, according to insider reports and internal communication documents, more than 30 dispatch workers who helped train the language model behind the popular chatbot ChatGPT were fired in March this year. .

Screenshots of internal Slack chats show that as of March 16, San Francisco-based outsourcing company Invisible Technologies laid off 31 dispatch workers. However, OpenAI continues to hire across its company.

The screenshot also shows that hundreds of dispatch workers from Invisible Technologies, known as "advanced artificial intelligence data trainers", are working with OpenAI to help the latter train its GPT chatbot. The company's AI data trainers are responsible for improving the models' coding skills, enhancing their creative writing abilities, or training them to refuse to respond to certain topics, one dispatch worker said. The dispatch worker asked to remain anonymous because of a confidentiality agreement, but sources confirmed his identity and employment.

Invisible Technologies vice president of operations Kamron Palizban (Kamron Palizban) discussed the layoffs at an all-hands meeting in March. He said in the leaked meeting recording that OpenAI hopes to reduce the number of dispatch workers due to changes in business needs. Parizban also said at the meeting that many of the laid-off dispatch workers were working on projects that did not provide a high enough return on investment for OpenAI.

OpenAI slashes the number of dispatch workers

Invisible Technologies’ relationship with OpenAI provides a glimpse into the ChatGPT maker’s data training. For the most part, OpenAI has kept these trainings secret.

The contract adjustment between OpenAI and Invisible Technologies follows reports that the former has increased its headcount for six consecutive months. People familiar with the matter said that as of January this year, OpenAI had hired nearly 1,000 data annotation dispatch workers in Eastern Europe and Latin America.

Just two months before the layoffs at Invisible Technologies, Microsoft had just injected $10 billion into OpenAI. But Invisible Technologies is not the only outsourcing company working with OpenAI.

An investigation by Time magazine showed that in February 2022, Sama, an outsourcing company also based in San Francisco, learned that its data labeling employees in Kenya were censoring harmful content such as sexual abuse, hate speech and violence. Later, the partnership with OpenAI was terminated.

In a statement to Time, an OpenAI spokesperson explained: “Classifying and filtering harmful text and images reduces the amount of violent and pornographic content contained in the training data to the lowest possible level. The minimum necessary steps and help create tools that can detect harmful content."

Jobs of Artificial Intelligence Trainers

According to dispatch workers at Invisible Technologies , the most basic responsibilities of an AI trainer include reviewing conversations between the AI ​​and its users to identify potentially illegal, privacy-invasive, offensive, or error-filled messages. Dispatch workers interviewed described their daily lives as follows:

After starting a shift, they first open the internal work browser and view the team's task list. They might click on a task like "Have a conversation about a random topic while browsing is disabled," and then enter a query in the message box.

After submitting a query, OpenAI’s model will generate four responses. Dispatch workers evaluate each response by opening a drop-down menu and selecting the type of error present, such as factual errors, spelling or grammatical errors, or harassment. The dispatch workers then rated the severity of the error on a scale of one to seven, with seven representing a "basically perfect" answer.

Next, the dispatch worker must craft a perfect response and submit it to complete the task. The dispatch worker said the results will be sent to quality inspectors at OpenAI and Invisible Technologies. This process needs to be repeated for each subsequent task.

Camren Parizban, vice president of operations at Invisible Technologies, said of OpenAI at the conference: “They are at a stage where they are about to have more clarity on where they are going. Grace Matelich, partner and operations manager at Invisible Technologies, said in a recorded meeting that the company based its firings on performance metrics such as "quality" and "throughput" of completed tasks. dispatch workers who performed poorly.

Underperforming dispatch workers and those who were hired but did not "meet certification thresholds" were fired, though some were given the option to move to different OpenAI teams, Mattledge said. He also said: "If you are still here today, I want you to know that it is because I believe in your ability to complete the task well." (Xiao Xiao)

The above is the detailed content of 31 ChatGPT training dispatch workers were fired after failing to meet expectations. For more information, please follow other related articles on the PHP Chinese website!