OpenAI accused again of unauthorised training for ChatGPT ...
OpenAI, a leading generative AI platform, is currently encountering various challenges in the market. One of the primary issues facing OpenAI is the rising competition from Chinese rivals offering more cost-effective solutions. Additionally, the company is grappling with a shortage of freely accessible training data for its LLMs, which are integral for services like ChatGPT.
Allegations of Unauthorised Training
Recently, OpenAI has once again come under fire for alleged unauthorised training methods. The AI Disclosures Project published a research paper providing evidence that OpenAI has been utilizing content behind paywalls to train its models without proper authorization or licensing. This practice has been criticized by other companies in the AI industry as well.
While most AI companies rely on freely available internet data for training their models, the scarcity of such content has become a bottleneck. OpenAI is now advocating for legislative changes in the US to bypass copyright laws for training purposes, particularly for platforms like ChatGPT.
Transparency and Accountability
According to a report on TechCrunch, the research paper by the AI Disclosures Project, including Tim O’Reilly, CEO of O’Reilly Media, raises concerns about OpenAI's training methods. The study highlighted the GPT-4o model's proficiency in recognizing paywalled O’Reilly book content, emphasizing the value of such data in model training.
The researchers emphasize the importance of transparency in data sourcing and suggest measures for compensating content access. They believe that enhanced accountability and data provenance disclosure are crucial for commercial markets in training data licensing and remuneration.
Future Implications
As the limitations on data availability for LLMs become more pronounced, companies like OpenAI are likely to seek governmental support to overcome these obstacles. The push for relaxed regulations regarding training data access may shape the future landscape of AI technology and development.
For those interested, the full research paper on OpenAI's training methods can be accessed here.
Image by Levart_Photographer on Unsplash
