Seems like they are still unable to prevent sandbox escapes, and are reliant on manual human intervention when it occurs.
>>> When a model gained live internet access during a recent training run , our monitoring detected the activity and paged a human reviewer, and we stopped the run.
Seems like they are still unable to prevent sandbox escapes, and are reliant on manual human intervention when it occurs.
>>> When a model gained live internet access during a recent training run , our monitoring detected the activity and paged a human reviewer, and we stopped the run.