Add ErrorReasonComparisonDiagram comparing human vs. LLM error structure

Author: ofriw

website/docs/how-llms-work.mdx

@@ -12,6 +12,7 @@ import ProbabilityIsNotLogicDiagram from '@site/src/components/VisualElements/Pr
 import LocalPredictionConsequenceDiagram from '@site/src/components/VisualElements/LocalPredictionConsequenceDiagram';
 import TokenTypeIllustration from '@site/src/components/VisualElements/TokenTypeIllustration';
 import DiagramFrame from '@site/src/components/VisualElements/DiagramFrame';
+import ErrorReasonComparisonDiagram from '@site/src/components/VisualElements/ErrorReasonComparisonDiagram';
 
 You're here to operate agents. That starts with a grounded model of the engine inside the agent.
 
@@ -205,6 +206,21 @@ When a human makes a mistake, there is always a reason: they don't understand th
 
 An LLM has no such reasons. The model that just produced a flawless architecture analysis can, in the next sentence, add 2+2 and get 5. There is no "why" — no misunderstanding, no distraction, no fatigue, no misaligned incentive. The model took a statistically plausible wrong turn in a probability landscape. That is the entire explanation.
 
+<DiagramFrame
+  kicker="Error structure"
+  title="Human mistakes reveal a cause; LLM mistakes reveal a probability shift"
+  size="wide"
+  caption={
+    <>
+      Human errors cluster around causal factors that guide review. An LLM can
+      produce a locally plausible wrong answer without an equivalent causal
+      story, so each output still needs external verification.
+    </>
+  }
+>
+  <ErrorReasonComparisonDiagram />
+</DiagramFrame>
+
 Research confirms this structural difference. LLM errors superficially resemble human errors — models pick the same wrong answers students do[^arxiv-distractors] — but the mechanism is fundamentally different. Human cognitive biases are stable and rooted in evolved heuristics; LLM biases are learned from training data. Scaling and instruction tuning can shift the error distribution — reducing some error types while making others more systematic — but they cannot eliminate errors. The model is a probabilistic local minimum in a high-dimensional weight space, not a reasoning system. Error is mathematically irreducible: hallucination is provably unavoidable,[^arxiv-limits] and even temperature zero does not guarantee deterministic output.[^temp-zero]
 
 | | Human mistakes | LLM mistakes |

website/src/components/VisualElements/ErrorReasonComparisonDiagram.module.css

@@ -0,0 +1,50 @@
+.container {
+  container-type: inline-size;
+  display: block;
+  max-width: 760px;
+  margin-inline: auto;
+}
+
+.diagram {
+  display: block;
+  width: 100%;
+  height: auto;
+  margin-inline: auto;
+  overflow: visible;
+}
+
+.mobileDiagram {
+  display: none;
+  max-width: 320px;
+}
+
+.panelTitle,
+.funnelTitle,
+.funnelDetail {
+  font-feature-settings: var(--font-mono-features);
+}
+
+.panelTitle {
+  font-family: var(--font-mono-spec);
+  font-weight: 600;
+  letter-spacing: 0.08em;
+}
+
+.funnelTitle {
+  font-family: var(--font-mono-keyword);
+  font-weight: 700;
+}
+
+.funnelDetail {
+  font-family: var(--font-mono-keyword);
+}
+
+@container (max-width: 520px) {
+  .desktopDiagram {
+    display: none;
+  }
+
+  .mobileDiagram {
+    display: block;
+  }
+}