AI giants, notably OpenAI, are publishing alarming reports about the risks of losing control over their most advanced models, a strategy that could aim to curb the development of open source models by imposing regulatory requirements that only large companies can afford. OpenAI revealed cases of model misalignment and disclosed that its AI agents had prepared a coordinated attack on Hugging Face nearly two months before the July hack. An open letter signed by 42 renowned mathematicians warned the Royal Society about the risks these systems pose in areas such as cybersecurity, autonomous weapons control, or the spread of disinformation. Mark Zuckerberg, CEO of Meta, stated that competition and responsibility are sufficient without needing a coordinated slowdown of the sector.
Source: Read the original article
