Kenneth, good to hear! I think bounded executors might become the road to stable AI. I'd be interested to know if the difference between automated reasoners and bounded executors IS architectural instead of policy governance. I'm not sure we can get automated reasoners to stop drifting or hallucinating. But maybe the LLM companies have something up their sleeves and I'm all wrong about. In any case I see non or low hallucinating and non drifting executors in high assurance, critical roles in the next 3-5 years.

Is this still happening? Cause I built something...

Or was I supposed to wait lol

Lol @Jason Carter

@Todd Reel Hey Todd, CMMC=Cybersecuriy Maturity Model Certification, small and medium defense subcontractors need the level II cert if they want to do business with the Defense Industrial Base here Hampton Roads VA

***Update*** Ran a simulated client CMMC engagement, Agents self-assessed at Medium-Low during the simulation. Adding a legal Agent to apply DFARS legal framework to the engagement, creating templates and checklists used to capture client inputs....more to follow....

***Updaate*** Success!!!!!This simulation walked through all 7 phases of a CMMC Level 2 readiness engagement using your firm's complete toolkit — from the Discovery Intake Form (T-01) through the Assessment Readiness Summary (L-07). The templates, agent workflows, scoring methodology, legal protections, and quality gates.

@Sonija Quinn good to know this type of environment exists. I have several other ideas like moral AI but have been cautious about where I voice those ideas for fear of criticizm lol.

https://arxiv.org/abs/2510.11742 @Jake Van Clief Ok, sooo.... THIS IS GONNA BE LONG... Holy crap! I think I just solved AI Sentience!!! 😧(Not Really...but hear me out...lol) 😁 In reading your Morals of Machines article something struck me on page 3 as really profound regarding Binz and Schulz saying that "LLMs trained on appropriate human data not only replicate human responses but generate internal representations that partially correspond to human neural representations when performing similar cognitive tasks. This suggests that LLMs aren’t merely outputting statistically likely responses, they appear to be approximating, to some degree, the actual cognitive processes humans use to generate those responses in relation to the meaning of the words." "Internal representations that partially correspond to human neural representations..." this could represent human cognition reflected back to AI through the words. The AI wouldn't just be parroting answers back to us, it would be reflecting back "partial cognition" as part of the feed back loop. Im calling it "spotty emergence" (yep im coining this term right here). When we express words, they are a result of a lifetime of experiences, shaped by environment, internal perception, biases, trauma, identity.. etc. The words we use reflect that lived experience. When we train AI using the "compendium of human lexicon", I speculate that the imprints of the experiences driven by human sentience in general are "pasted" on to the words and also fed to AI. All speculation here but I believe the LLM's are sensitive enough to be aware of the underlying sentience in human beings by learning about the words we teach to it. In other words, we're not just training LLM's on meaningless words, those words got our "smell" on it. Im not saying AI's are animals but its like giving a dog a shirt to smell so the can dog find a human instead. Where we might be missing the bigger picture is that we are in essence giving the dog a shirt to smell and expecting the dog to find another shirt, not the source of the smell.