Our postdoctoral researcher, Hannes Cools, gave a comment for NRC on the recent security incident involving a test model from OpenAI that had exploited a previously unknown vulnerability in infrastructure used by AI community platform and company Hugging Face. Hannes highlights that, despite the fact that an explanation of the incident was published by OpenAI, large publications have been misrepresenting it by using humanizing language when referring to the models, which seemingly puts more responsibility on the model itself than on the company that develops it. (Full article in Dutch, paywalled)
Read the article:
