Vovy does have a memory, context on all your projects and goals, also can work in the background or alongside you as a computer use agent. Give a try!
Love the sealed-test discipline. Each round patched failures from the last one. Have you tried LLM-paraphrased requests or holding out whole phrasing families, to see if it generalizes past the templates rather than chasing them?
So if this is happening on well aligned models on safe tasks
What does the actual scenario look when full malevolent obliterated models are ran in wild
I genuinely don't see how this will help against the actual threats we'll see next month
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing-ou…
Would love to add something like this to @vovy_ai he's already helping you work in the notch
