Tech companies need raw data to train their AI chatbots. As demand grows, they’re paying workers to generate fresh training data, often with low-quality contracts. To cut corners and minimize errors, many workers are using other AI chatbots to supply the data needed for their own projects. Experts warn that this “AI cannibalism” can destabilize large language models.
Employees like Alice say it’s easy to pass off AI-generated data as their own, as long as they scrub out chatbot tics. However, many workers feel forced into using LLMs due to the fear of making mistakes and losing their income. They use multiple AI chatbots in tandem to create scenarios and files, despite feeling guilty about it.
This phenomenon could have severe consequences for the AI industry, highlighting the irony that companies are being outmaneuvered by the workers they hired to generate data for them. If companies want high-quality data, they need to offer quality contracts that prioritize worker well-being, rather than exploiting them with low pay and short-term work.
Source: https://futurism.com/artificial-intelligence/ai-companies-contractors-slop-data-training