Not all 'open source' AI models are open: here's a ranking

Researchers have found that many large language models powering chatbots claim to be open source but restrict access to code and training data. The definition of open source in AI models is still debated, with advocates emphasizing the importance of full openness for scientific progress and AI accountability. A study by language scientists identified the most and least open models, revealing that some big tech companies engage in "open-washing" by claiming openness while disclosing minimal information. The European Union's upcoming Artificial Intelligence Act will have implications for models classified as open source, offering them exemptions from certain transparency requirements. Smaller players are noted for being more transparent compared to tech giants. The study also highlights concerns about the lack of transparency regarding the training data used in these models. The importance of openness for reproducibility and innovation in AI research is emphasized, with the need for models to be scrutinized and understood to assess their achievements accurately.