Skip to content
TheChatArchiveThe Chat Archive
Claudeclaude-sonnet-4-20250514· AnthropicHugging Face LMArena Human Preference 140K

graph TD; %% === 1. DEFINE ALL NODES === classDef stage…

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with o4-mini-2025-04-16’s answer.
  1. User

    graph TD; %% === 1. DEFINE ALL NODES === classDef stage fill:#f8f9fa,stroke:#333,stroke-width:2px,stroke-dasharray: 3; classDef io fill:#e6f3ff,stroke:#333,stroke-width:2px;

    Input["Input Tensor <br/> (Batch, Seq_Len, 17, 3)"];
    
    S1_A["Flatten Keypoints <br/> (Batch, Seq_Len, 51)"];
    S1_B["Linear Embedding <br/> (51 -> d_model)"];
    
    S2_A["Add Positional Encoding"];
    S2_B["N x Transformer Encoder Layers <br/> (Multi-Head Self-Attention)"];
    S2_C["Mean Pooling over Sequence <br/> (Temporal Aggregation)"];
    
    S3_A["MLP Classifier <br/> (Linear -> ReLU -> Linear)"];
    
    Output(("Output Logit <br/> (Batch, 1)"));
    
    %% === 2. GROUP NODES INTO STAGES ===
    subgraph "Stage 1: Input Processing"
        S1_A; S1_B;
    end
    
    subgraph "Stage 2: Temporal Feature Extraction"
        S2_A; S2_B; S2_C;
    end
    
    subgraph "Stage 3: Classification Head"
        S3_A;
    end
    
    %% === 3. DEFINE ALL CONNECTIONS ===
    Input --> S1_A;
    S1_A --> S1_B;
    S1_B --> S2_A;
    S2_A --> S2_B;
    S2_B --> S2_C;
    S2_C --> S3_A;
    S3_A --> Output;
    
    %% === 4. APPLY STYLES ===
    class Input,Output io;
    class S1_A,S1_B,S2_A,S2_B,S2_C,S3_A stage;
    

    can you make the arrows from stage to stage only connected to the outer blocks not the inner ones?

  2. Claude Assistant

    Here's the modified version with arrows only connecting between stages at the outer level:

    graph TD;
        %% === 1. DEFINE ALL NODES ===
        classDef stage fill:#f8f9fa,stroke:#333,stroke-width:2px,stroke-dasharray: 3;
        classDef io fill:#e6f3ff,stroke:#333,stroke-width:2px;
    
        Input["Input Tensor <br/> (Batch, Seq_Len, 17, 3)"];
        
        S1_A["Flatten Keypoints <br/> (Batch, Seq_Len, 51)"];
        S1_B["Linear Embedding <br/> (51 -> d_model)"];
        
        S2_A["Add Positional Encoding"];
        S2_B["N x Transformer Encoder Layers <br/> (Multi-Head Self-Attention)"];
        S2_C["Mean Pooling over Sequence <br/> (Temporal Aggregation)"];
        
        S3_A["MLP Classifier <br/> (Linear -> ReLU -> Linear)"];
        
        Output(("Output Logit <br/> (Batch, 1)"));
    
        %% === 2. GROUP NODES INTO STAGES ===
        subgraph "Stage 1: Input Processing"
            S1_A --> S1_B;
        end
        
        subgraph "Stage 2: Temporal Feature Extraction"
            S2_A --> S2_B --> S2_C;
        end
        
        subgraph "Stage 3: Classification Head"
            S3_A;
        end
    
        %% === 3. DEFINE STAGE-TO-STAGE CONNECTIONS ===
        Input --> S1_A;
        S1_B --> S2_A;
        S2_C --> S3_A;
        S3_A --> Output;
    
        %% === 4. APPLY STYLES ===
        class Input,Output io;
        class S1_A,S1_B,S2_A,S2_B,S2_C,S3_A stage;
    

    The key changes:

    • Moved the internal connections (S1_A --> S1_B and S2_A --> S2_B --> S2_C) inside their respective subgraphs
    • Only the stage-to-stage connections (Input → S1_A, S1_B → S2_A, S2_C → S3_A, S3_A → Output) cross between the grouped sections

    This creates a cleaner flow where arrows only connect the entry and exit points of each stage.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.