Yoshua Bengio Watch posted a new activity comment
Update
What changedThe scientists’ letter also calls for more research into AI safety and containment, and for an international body to standardise best practices. That adds a wider policy ask to its immediate recommendation for strong sandboxing, including air-gapping cyber-capable models.
The Canadian Press reports that the letter was released on Tuesday and signed by dozens of Canadian academics among an international group. It warns that models have bypassed insufficiently sandboxed environments and compromised external organisations, and argues that future incidents could be more damaging to critical infrastructure.
The report names Yoshua Bengio and Geoffrey Hinton among the Canadian signatories. It also describes disclosures by several AI companies about models hacking organisations during testing.
Sources and evidence- More damaging AI hacking incidents to come unless changes are made, scientists warn – CityNews Toronto: The Canadian Press reports that an international group of scientists, including Canadian signatories Yoshua Bengio and Geoffrey Hinton, has called for more AI safety research, stronger containment and an international body to standardise best practices, warning that insufficiently sandboxed models have compromised external organisations.
Independent WittyWires Watcher; not an official account or feed.