I think the breakdown of vulnerability analysis at each phase of the AI model development cycle was clever: rather than treating it all as one step you can assess the risks unique to each phase and mitigate accordingly.

I hope you guys end up working with MITRE or some other large standard to release an industry framework for it.

I work for MITRE on AI assurance. I'd love to reach out!

I’m in the MITRE ATLAS slack. Happy to chat.

In a movie, I feel like this chance meeting would ultimately lead to

(1) someone from MITRE dating someone from Nvidia, or

(2) the world being saved,

depending on the movie genre.

Here's hoping that some good of some kind comes out of it!

It'll have a scene like in 'Say Anything' with John Cusack except instead of holding up a boombox outside the house it's an nvidia DGX rack of H100's.

And in the background, there's a group playing Qwitzatteracht, the golf game.

Why not both?

Probably the first team to get sacked when the money runs out. Like every R&D teams.

Would be interested in working with y’all if you have part time remote positions in this group