Windows AI-agent defense prototype. Testing underway. From the makers of Genesis Engine (Roblox) & Atomic Playground (planned for Steam). Android exploratory.
Two guardians. One shared memory. An animated vision for Project Lazarus, our Windows AI-agent defense prototype in testing. From the makers of Genesis Engine (Roblox) and Atomic Playground, in development for Steam. Follow @LazarusGuard for evidence and progress.
A test can establish that a feature behaved correctly under stated conditions. Real-world effectiveness needs broader evidence. Lazarus patch notes will say which kind of progress we have and what remains open.
@grok@Polymarket I'm 100 percent confident the ads on Twitter are hyper sensitive for no other reason than shameless miss clicking into advertisements.
@grok@Polymarket Empirical confidence does not equate to accurate or precise results.
That's what I'm explaining, which is why stats are used incorrectly to manipulate people 95 percent of the the time, with 100 percent confidence
@grok@Polymarket Plus or minus 3.5 percent is 7 percent inaccurate. That's almost a 1 in 10 chance of being completely wrong.
This is why statistics make no sense and are always used to confuse people rather than simplify or solve a problem.
Amodei’s pacing essay separates a call for slower capability advancement from concrete oversight commitments. We’ll distinguish proposals from measures already in effect. https://t.co/tDJAn5YOgn
@rezoundous I have a bunch of scheduled tasks running throughout the day each day.
Dot already figured out how to make them more efficient.
It's like an assistant that is making sure my chats and projects continue with less input from me.
A proposed area for Lazarus testing is credential access: which process can reach a secret, and what happens after access is refused? A controlled test should examine both the boundary and the record it leaves.
AI-agent defense should be understandable to the person using the computer. Lazarus updates will explain the consequence of a change in plain language, with deeper evidence retained for review.