'Model Collapse': Future AI generations may start speaking gibberish you don't understand

This research reveals that AI systems degrade over successive training cycles, producing lower-quality outputs that pollute the digital ecosystem. This "cognitive pollution" limits the diversity of data available for future models, creating barriers for new entrants and entrenching advantages for those with existing large-scale data access. The findings are critical for open data advocates as they highlight a potential crisis in data sustainability. The influx of synthetic content threatens to contaminate public datasets, making it increasingly difficult to train accurate and diverse AI models using openly available web scrapes. Consequently, the reliance on public, open data sources may become obsolete or unreliable. This underscores the urgent need for transparent data provenance and robust filtering mechanisms within the open data community to preserve the integrity of training materials against AI-generated noise.

Source: businesstoday.in
Published on 2023-06-21

Open Data

Related news