Research
Building Human-Centered AI
Systematic evaluation of CAF 5-2.2D against validated Theory of Mind methodologies (Kosinski et al., 2024; Wimmer & Perner, 1983). Results: 100% Tier 1 accuracy on ~70B substrate versus documented 88% GPT-4 ceiling; 93% on novel Tier 2 cognitive integration rubric. Full test battery documentation included.
The AI industry is convinced capability scales with parameters. But what if the bottleneck is architecture, not size? Sub-frontier models can match frontier performance - when the infrastructure doesn't fight you. Here's the evidence.
Intractable doesn't mean unknowable, it’s messy. And messy is engineerable. AI is now one of the biggest chunks of the U.S. economy, built on technology the industry can't explain. Here's the tractable alternative."
Crafted Logic Lab wasn't the plan. I set out to write science fiction about artificial minds—and ended up building a company that creates them instead. This is the story of how that happened.