ChatGPT could replace AI trainers, study says

News on April 14, according to foreign media reports, researchers from the University of Zurich recently found in a new paper that ChatGPT can outperform crowd workers in text annotation tasks , that is, labeling text used to train artificial intelligence systems.

The researchers provided ChatGPT with a sample of 2,382 tweets and asked it to classify the text based on relevance, topic, stance, problem or solution framing, and policy framing. The researchers concluded that by using ChatGPT, they were able to achieve higher accuracy and consistency across codes. Best of all, they discovered they could save money by using AI-powered ChatGPT, which is 20 times cheaper than paying a human on Mechanical Turk.

This research adds to a larger discussion about how rapidly evolving artificial intelligence language models, such as OpenAI’s GPT family, will impact work. Researchers at OpenAI argued in a recent paper that the introduction of GPT could impact at least 10% of tasks performed by 80% of the U.S. workforce. However, automating human annotators is particularly problematic because this is already a precarious group of workers—an outsourced workforce that performs tasks for pennies on the dollar for big tech companies.

Of course, although some technology giants have made rapid progress in the field of artificial intelligence and achieved great achievements, the reality is that all their artificial intelligence models rely on human power.

Tech companies use tens of thousands of employees to manually label and filter content in AI model datasets. That's because AI often isn't able to recognize the nuances of an image yet, especially when it's still training. Even when AI models are deployed, they rely on human user interaction to fine-tune and identify model shortcomings.

OpenAI, the creator of ChatGPT, reportedly paid Kenyan workers less than $2 an hour to make its chatbot safer to use.

The above is the detailed content of ChatGPT could replace AI trainers, study says. For more information, please follow other related articles on the PHP Chinese website!