Signal Kiln Notes https://signalkiln.com/ A personal research notebook for deep dives, unusual connections, and niche ideas worth following. en Mon, 14 Sep 2026 20:30:00 GMT Anthropic Is Betting on Skills. They’re Wrong—and Here’s Why. https://signalkiln.com/research/anthropic-betting-on-skills/ https://signalkiln.com/research/anthropic-betting-on-skills/ Mon, 14 Sep 2026 20:30:00 GMT By Aaron Bosloper. Why a growing Skill library inevitably requires routing, pruning, security, versioning, and governance—and why specialist models belong inside agent systems. AI Agent architecture AI Memory Degradation Is Real https://signalkiln.com/research/can-ai-develop-alzheimers/ https://signal-kiln.abozloper.chatgpt.site/research/can-ai-develop-alzheimers/ Sun, 13 Sep 2026 15:00:00 GMT By Aaron Bosloper. An evidence-based article on measured forgetting, degraded recall, contaminated memories, and recursive model collapse, with results from original studies and clear limits on the Alzheimer’s analogy. AI Memory and identity Why I research in public https://signalkiln.com/#protocol signal-kiln-field-note-000 Sun, 13 Sep 2026 15:00:00 GMT A note on sources, unfinished questions, unusual connections, and letting conclusions change in public.