# Clarethium research > Experiments on AI and human judgment. Each finding ships with what survived testing, what did not, and how to verify it. Source: https://clarethium.com/blog Machine-readable index: https://clarethium.com/blog/transmissions.json Sitemap: https://clarethium.com/blog/sitemap.xml ## Findings - [Why 'Don't Be Generic' Doesn't Work](https://clarethium.com/blog/dont-be-generic) (2026-07-12): Telling a model 'don't be generic' does nothing on its own; giving it specific anchors does. The gain is verifiability, not quality. - [Three Questions Before You Prompt AI](https://clarethium.com/blog/before-you-type) (2026-07-05): Three questions before you type structure most of the prompt. Specificity is the strongest single lever (Hedges g=1.34); 'be exceptional' alone does almost nothing. - [The Cheap Half of the Loop](https://clarethium.com/blog/cheap-half-of-the-loop) (2026-06-27): Coding pulled ahead because its feedback is cheap to check. That is one half of the loop. The same training trains out the other half, and the part that chooses the frame is the last thing to get cheap. - [Why AI Defaults to Generic](https://clarethium.com/blog/the-default) (2026-06-22): Every prompt technique is one move: make the default path expensive enough that the model leaves it. Specificity is the largest measured version. - [Past the Obvious](https://clarethium.com/blog/past-the-obvious) (2026-06-09): The original ideas arrive late, after the obvious ones clear. You cannot will yourself there, but you can force the move off the default path, borrow from a far domain, then let reality correct you. - [Three AIs, No Source, the Same Answer](https://clarethium.com/blog/source-is-the-substrate) (2026-05-25): Same model, same prompt. The source you paste, not the prompt you write, decides whether the numbers are real. - [You Can Only Evaluate What You Could Produce](https://clarethium.com/blog/ownership-test) (2026-05-19): You defend AI-shaped conclusions you cannot rebuild. The ten-minute test reveals which parts of your work are yours. The discipline that compounds: choose what to own, delegate what AI can verify. - [Satisfaction Turns Off Your Doubt, Not Your Detection](https://clarethium.com/blog/satisfaction-trap) (2026-05-11): You notice when AI argues with you. You do not notice when AI confirms you. Confirmation has no signature, so the impulse to question it never fires. Satisfaction is the trap. - [AI Amplifies What You Bring](https://clarethium.com/blog/ai-amplifies-what-you-bring) (2026-05-06): Same model, same task, two paragraphs of operator context, dramatically different output. The kit, the design history, and the principle for adapting it to your own situation. - [Frame Check](https://clarethium.com/blog/frame-check) (2026-05-05): Drop any document in. See which analytical perspectives it covers, which it skips, the voice, what evidence backs each numerical claim. Free, open source, useful from the first paste. - [How AI Makes You More Wrong With More Analysis](https://clarethium.com/blog/frame-trap) (2026-04-26): Five hours of analysis, increasingly sharp, increasingly wrong. The frame amplifies. What changed the outcome was the reframe, not more analysis. - [Stop Calling It Hallucination](https://clarethium.com/blog/stop-calling-it-hallucination) (2026-04-25): Hallucination is six or more distinct failure modes. Different mechanisms. Different solutions. Name the type first. - [The Decision That Was Never Made](https://clarethium.com/blog/answer-trap) (2026-04-25): AI resolved the uncertainty before your own thinking had time to finish. Resolution and decision are different things. - [Why Experts Miss What Beginners Catch](https://clarethium.com/blog/construction-trace) (2026-04-25): Generation builds the mental model that makes evaluation possible. Without it, evaluation collapses to surface features. - [What You Feel When AI Disagrees](https://clarethium.com/blog/disagreement-audit) (2026-04-25): Ask AI to oppose you. Four observable signals reveal whether you are evaluating or defending. The same defensive reaction fires on AI opposition as on human. - [Adding Information Often Doesn't Help](https://clarethium.com/blog/context-serves-search) (2026-04-20): Adding information to an already-thorough prompt produced near zero improvement. Three constraint sentences changed everything. - [Stop Polishing, Start Switching](https://clarethium.com/blog/ceiling-switch) (2026-04-16): The ceiling is per generation mode. Switch modes to access territory that iteration can't reach. - [Your Verdict Is In Before You Read It](https://clarethium.com/blog/first-read) (2026-04-07): The same defensive reaction that fires when a person disagrees with you fires when an AI does, and the first read lands before conscious evaluation begins. Speed is what hides it. - [Four Layers Produce Every AI Output](https://clarethium.com/blog/system-layer) (2026-04-03): Four layers produce every AI output. The company's system. Your system. Your prompt. The model. The model is the only one with a name. - [Same Technique, Opposite Results](https://clarethium.com/blog/constraint-paradox) (2026-03-24): The evaluative structure that produced precision on convergent problems actively harmed exploratory ones. Organizational structure helped both. - [For Behavior, the Model Is Rarely the Variable](https://clarethium.com/blog/attribution-error) (2026-03-24): The context determined whether behaviors existed at all. The model adjusted the volume. - [The Output That Feels Most Trustworthy Is Often the Least Reliable](https://clarethium.com/blog/trust-signals-are-inverted) (2026-03-23): The signals you use to judge AI trustworthiness are the same signals fabrication produces. - [The Most-Cited Finding Was Wrong](https://clarethium.com/blog/catching-your-own-overclaim) (2026-03-23): The most-cited effect across 90+ experiments was three effects stacked. Honest magnitude: 40% smaller. - [Why AI Can't Verify Its Own Work](https://clarethium.com/blog/self-check-illusion) (2026-03-23): The agent reported clean. The output was wrong. Same process generating and evaluating. - [How to Stop AI from Making Up Numbers](https://clarethium.com/blog/source-conditioning) (2026-03-23): Source material drops unsourced numbers from roughly half to single digits. Three steps. - [Most AI Numbers Are Unverifiable](https://clarethium.com/blog/fabrication-architecture) (2026-03-23): 77 to 100 percent of AI-generated numbers are temporally unstable. Source material fixes it. Prompts don't.