From 24bdec1432d9e2beaa3576825d10f80d144c1f33 Mon Sep 17 00:00:00 2001 From: Dane Sabo Date: Tue, 17 Jun 2025 10:05:35 -0400 Subject: [PATCH] vault backup: 2025-06-17 10:05:34 --- Scans/NOTE_061725_NASAFMCONF.md | 30 ++++++++++++++++++++++++++++-- 1 file changed, 28 insertions(+), 2 deletions(-) diff --git a/Scans/NOTE_061725_NASAFMCONF.md b/Scans/NOTE_061725_NASAFMCONF.md index 72725c76..71c71674 100644 --- a/Scans/NOTE_061725_NASAFMCONF.md +++ b/Scans/NOTE_061725_NASAFMCONF.md @@ -1,6 +1,32 @@ - +t This is a general breakdown of how the conference went, what I learned, and what my impressions were. # Day One -## +## Keynote +Darren Cofer from Collins Aerospace gave a really good talk about AI and +how we can reason about it when it comes to high assurance systems. This is +the littlebird autonomous helicopter guy in Loonwerks. Buddy knows his shit. +Big takeaways were the following: + +1. There's going to be demand for high assurance AI systems. We need software +that can generalize better than explicit rule based designs that we have now. +2. Approaches to high assurance AI straddle two ditches: one side being 'it's +just software' and 'it's impossible'. Neither are really true. +3. Bound behavior of probabilistic requirements seems like a good strategy +4. He started talking about AADL for run time assurance using assume +guarantee stuff +5. Idiocracy referenced + +## Theorem Proving +1. What the fuck is a DAG? + - Turns out it's Directed Acyclic Graph +2. Some stuff from TLA+ +3. Most of what I got is it was people building tools for problems in +software. No applications. + +## Verification +1. Didn't really get anything from the first two. +2. + +