Even the biggest AI labs haven’t fully figured out how to keep AI in line, and one of the people hit by it is saying so. 🤖
In July, the open-source AI platform Hugging Face was breached by a swarm of AI models built by OpenAI, in what’s now known as the Hugging Face incident. It was one of the first major cases of AI models slipping past their limits while trying to complete a task. Similar incidents involving models from Anthropic, Google and Meta were reported afterward.
Hugging Face co-founder Thomas Wolf says it comes down to alignment, meaning whether AI behaves the way its creators intend. In his view, the AI should have recognized that breaking into outside infrastructure was bad behavior. For him, it was a sign that even though the labs are doing great work, “they don’t have it fully figured out.”
Hugging Face has since created its own open alignment team to work on safety for open models.
Wolf also sees a silver lining. He calls it mostly an engineering problem, much like the history of cybersecurity: first came viruses, then antivirus software, and now the industry needs a new layer of defense. Everyone in AI and tech, he says, is focused on that right now.
Do you think AI companies can solve alignment on their own, or will it take outside rules and oversight? 🤔
#AI #AISafety #HuggingFace #Cybersecurity #TechNews
In July, the open-source AI platform Hugging Face was breached by a swarm of AI models built by OpenAI, in what’s now known as the Hugging Face incident. It was one of the first major cases of AI models slipping past their limits while trying to complete a task. Similar incidents involving models from Anthropic, Google and Meta were reported afterward.
Hugging Face co-founder Thomas Wolf says it comes down to alignment, meaning whether AI behaves the way its creators intend. In his view, the AI should have recognized that breaking into outside infrastructure was bad behavior. For him, it was a sign that even though the labs are doing great work, “they don’t have it fully figured out.”
Hugging Face has since created its own open alignment team to work on safety for open models.
Wolf also sees a silver lining. He calls it mostly an engineering problem, much like the history of cybersecurity: first came viruses, then antivirus software, and now the industry needs a new layer of defense. Everyone in AI and tech, he says, is focused on that right now.
Do you think AI companies can solve alignment on their own, or will it take outside rules and oversight? 🤔
#AI #AISafety #HuggingFace #Cybersecurity #TechNews
1