new file mode 100644
index 0000000..b5991f3
@@ -0,0 +1,31 @@
+---
+type: idea
+title: LLM behavior improvement
+tags: [ai, research, training]
+status: raw
+first_captured: 2026-03-16
+updated: 2026-04-10
+sources:
+ - sources/ideaflow/2026-03-16_noticed-that-claude-codex-were-kinda-failing-on.md
+ - sources/ideaflow/2026-04-09_these-days-llms-etc-often-behave-poorly-becuase.md
+ - sources/ideaflow/2026-04-09_some-sort-of-a-graph-think-for-llms.md
+---
+
+# LLM behavior improvement
+
+multiple captures around making LLMs better at their jobs:
+
+1. **codebase scanning for weak spots**: noticed Claude and Codex failing on Prisma — scan the codebase for things the LLM is not confident about, then note exactly how to deal with them in AGENTS.md. "kinda like spec but for weird parts of codebase."
+
+2. **acting without enough info**: LLMs often behave poorly because they act without enough information. could be solved with better prompting/telling users to give more info, or with a dataset training models to guess intent intelligently and refuse unspecced tasks.
+
+3. **graph thinking for LLMs**: current approaches crudely approximate human thinking (x then y then z) but miss backtracking and organization. proxy: linear text with organization indicators. noted counterpoint: attention mechanism already integrates everything non-linearly, "so possibly moot."
+
+connects to the [[agents-md-research|AGENTS.md research]] and the [[overnight-app-grinder|overnight app grinder]] which depends on LLMs being reliable.
+
+---
+
+## timeline
+
+- [2026-03-16] codebase scanning concept
+- [2026-04-09] acting-without-info problem, graph thinking exploration