Why Your Mobile Network is Smarter Than Ever Thanks to AI: Look at These Companies Changing How You Use Your Mobile Phone!
June 24, 2024
AI-Enabled Scams Explode: Booking.com Security Boss Sounds the Alarm!
June 24, 2024
Which AI models are really “open-source”? This study shows which are which!
Not All ‘Open Source’ AI Models Are Truly Open
Recent analysis reveals that major technology companies like Meta and Microsoft are mislabeling their AI models as ‘open source’ without fully disclosing critical information about their technology. Researchers Mark Dingemanse and Andreas Liesenfeld found that many large language models (LLMs) do not provide comprehensive access to code and training data. This practice, known as “open-washing,” is particularly problematic as the European Union’s forthcoming AI Act will impose fewer regulations on models classified as open source, making the definition of openness more significant (Nature).
Smaller Players Leading in True Openness
The study, presented at the 2024 ACM Conference on Fairness, Accountability, and Transparency, evaluated 40 LLMs on 14 parameters, including code availability and data transparency. Surprisingly, smaller firms and academic groups, such as the team behind BLOOM, demonstrated higher levels of openness compared to big tech companies. The findings highlight that many models, including Meta’s Llama and Google DeepMind’s Gemma, are only “open weight,” meaning they allow limited external use without full access to underlying data or customization options.
Transparency Essential for Scientific Integrity
The lack of transparency in training data is a major concern, with many models providing only vague descriptions of their datasets. This opacity undermines the scientific principle of reproducibility, as researchers need detailed information to replicate and build upon existing models. The study emphasises the importance of true openness for scientific innovation and accountability, warning that without clear definitions and rigorous standards, the label of ‘open source’ could be exploited, compromising both scientific progress and ethical standards in AI development.
(Visit Nature for the full story)
*An AI tool was used to add an extra layer to the editing process for this story.
