Switched around answer order in chapter 12 sections 2 and 3 so first response is not always correct

chapters/en/chapter12/2.mdx

@@ -142,15 +142,15 @@ In the next module, we'll get our hands dirty and dive into the DeepSeek R1 pape
 
 <Question
     choices={[
+        {
+            text: "Model, Data, Loss Function, and Optimizer",
+            explain: "These are components more commonly associated with supervised learning."
+        },
         {
             text: "Agent, Environment, Action, Reward, and Policy",
             explain: "Correct! These are the fundamental components that make up a reinforcement learning system.",
             correct: true
         },
-        {
-            text: "Model, Data, Loss Function, and Optimizer",
-            explain: "These are components more commonly associated with supervised learning."
-        },
         {
             text: "Input, Output, and Hidden Layers",
             explain: "These are components of neural network architecture, not specifically RL components."
@@ -162,18 +162,18 @@ In the next module, we'll get our hands dirty and dive into the DeepSeek R1 pape
 
 <Question
     choices={[
-        {
-            text: "It helps align models with human preferences and values",
-            explain: "Correct! RLHF uses human feedback to guide models toward more helpful, harmless, and aligned behavior.",
-            correct: true
-        },
         {
             text: "It makes models generate text faster",
             explain: "RLHF isn't primarily about improving generation speed."
         },
         {
             text: "It reduces the model's memory usage",
             explain: "RLHF doesn't focus on model efficiency or memory optimization."
+        },
+        {
+            text: "It helps align models with human preferences and values",
+            explain: "Correct! RLHF uses human feedback to guide models toward more helpful, harmless, and aligned behavior.",
+            correct: true
         }
     ]}
 />
@@ -202,18 +202,18 @@ In the next module, we'll get our hands dirty and dive into the DeepSeek R1 pape
 
 <Question
     choices={[
-        {
-            text: "To provide feedback on how well the model's responses align with desired behavior",
-            explain: "Correct! Rewards guide the model toward generating more helpful, truthful, and appropriate responses.",
-            correct: true
-        },
         {
             text: "To measure the model's vocabulary size",
             explain: "Rewards aren't used to evaluate vocabulary knowledge."
         },
         {
             text: "To determine the model's training speed",
             explain: "Rewards provide feedback on response quality, not training efficiency."
+        },
+        {
+            text: "To provide feedback on how well the model's responses align with desired behavior",
+            explain: "Correct! Rewards guide the model toward generating more helpful, truthful, and appropriate responses.",
+            correct: true
         }
     ]}
 />
@@ -222,18 +222,18 @@ In the next module, we'll get our hands dirty and dive into the DeepSeek R1 pape
 
 <Question
     choices={[
-        {
-            text: "A numerical score that measures the quality of a response",
-            explain: "Correct! Rewards provide feedback on response quality, guiding the model toward desired behavior.",
-            correct: true
-        },
         {
             text: "A function that generates responses",
             explain: "Rewards are feedback on response quality, not the generation process itself."
         },
         {
             text: "A model that evaluates the quality of responses",
             explain: "Rewards are feedback on response quality, not an evaluation model."
+        },
+        {
+            text: "A numerical score that measures the quality of a response",
+            explain: "Correct! Rewards provide feedback on response quality, guiding the model toward desired behavior.",
+            correct: true
         }
     ]}
 />

chapters/en/chapter12/3.mdx

@@ -231,18 +231,18 @@ In the next section, we'll explore practical implementations of these concepts,
 
 <Question
     choices={[
-        {
-            text: "The GRPO algorithm that enables learning from preferences with and without a reward model",
-            explain: "Correct! GRPO's key innovation is its ability to directly optimize for preference rectification, making it more efficient than traditional RL methods.",
-            correct: true
-        },
         {
             text: "Using more GPUs for training than any previous model",
             explain: "The paper's innovation is in its algorithmic approach (GRPO) rather than computational resources used."
         },
         {
             text: "Creating a larger language model than existing ones",
             explain: "The innovation lies in the training methodology and GRPO algorithm, not in model size."
+        },
+        {
+            text: "The GRPO algorithm that enables learning from preferences with and without a reward model",
+            explain: "Correct! GRPO's key innovation is its ability to directly optimize for preference rectification, making it more efficient than traditional RL methods.",
+            correct: true
         }
     ]}
 />
@@ -271,18 +271,18 @@ In the next section, we'll explore practical implementations of these concepts,
 
 <Question
     choices={[
-        {
-            text: "A process where the model recognizes errors, self-corrects, and explains its corrections",
-            explain: "Correct! The paper describes this as a four-step process: initial attempt, recognition of errors, self-correction, and explanation of the improvement.",
-            correct: true
-        },
         {
             text: "The point where the model reaches human-level performance",
             explain: "The 'Aha Moment' specifically refers to the model's ability to recognize and correct its own mistakes, similar to human problem-solving realizations."
         },
         {
             text: "When the model completes its training process",
             explain: "The 'Aha Moment' is about the model's emergent ability to recognize errors and self-correct during problem-solving, not about training completion."
+        },
+        {
+            text: "A process where the model recognizes errors, self-corrects, and explains its corrections",
+            explain: "Correct! The paper describes this as a four-step process: initial attempt, recognition of errors, self-correction, and explanation of the improvement.",
+            correct: true
         }
     ]}
 />
@@ -311,15 +311,15 @@ In the next section, we'll explore practical implementations of these concepts,
 
 <Question
     choices={[
+        {
+            text: "R1-Zero is smaller than R1",
+            explain: "The difference is in their training approaches (pure RL vs. multi-phase), not their model sizes."
+        },
         {
             text: "R1-Zero uses pure RL while R1 combines RL with supervised fine-tuning",
             explain: "Correct! As shown in the comparison table, R1-Zero uses pure RL training while R1 uses a multi-phase approach combining supervised fine-tuning with RL, resulting in better language consistency.",
             correct: true
         },
-        {
-            text: "R1-Zero is smaller than R1",
-            explain: "The difference is in their training approaches (pure RL vs. multi-phase), not their model sizes."
-        },
         {
             text: "R1-Zero was trained on less data",
             explain: "The key distinction is their training methodology: pure RL for R1-Zero versus a combined SFT and RL approach for R1."