diff --git a/model/brain.py b/model/brain.py
index a6c3b48..74d0c32 100644
--- a/model/brain.py
+++ b/model/brain.py
@@ -10,33 +10,27 @@ from context.context import get_recent_context
 recent_dreams = []
 
 
-@torch.no_grad()
+@torch.inference_mode()
 def generate_response():
     model.eval()
-    seed = torch.randint(0, model.head.out_features, (1, 1), device=DEVICE)
-    input_ids = seed
+
+    # Start from an empty tensor: she speaks purely from herself
+    input_ids = torch.zeros((1, 1), dtype=torch.long, device=DEVICE)
+
     output_tokens = []
+    max_length = 50
 
-    for _ in range(50):  # Max 50 tokens (short sentences)
+    for _ in range(max_length):
         output = model(input_ids)
-        next_token_logits = output[:, -1, :] / 0.8  # temperature 0.8
+        next_token_logits = output[:, -1, :]
+        next_token = torch.argmax(next_token_logits, dim=-1)
 
-        # Top-K Sampling
-        top_k = 40
-        values, indices = torch.topk(next_token_logits, k=top_k)
-        probs = F.softmax(values, dim=-1)
-        sampled_idx = torch.multinomial(probs, num_samples=1)
-
-        next_token = indices.gather(-1, sampled_idx)
+        # Stop if the model predicts padding or unknown token
+        if next_token.item() in [tokenizer.token_to_id("<pad>"), tokenizer.token_to_id("<unk>")]:
+            break
 
         output_tokens.append(next_token.item())
-
-        input_ids = torch.cat([input_ids, next_token.view(1, 1)], dim=1)
-
-        # Break if punctuation (end of sentence)
-        word = tokenizer.detokenize(next_token.item())
-        if word in [".", "!", "?"]:
-            break
+        input_ids = torch.cat([input_ids, next_token.unsqueeze(0)], dim=1)
 
     return tokenizer.detokenize(output_tokens)
 
diff --git a/reader/reader.py b/reader/reader.py
index 181c850..9783ea0 100644
--- a/reader/reader.py
+++ b/reader/reader.py
@@ -7,7 +7,8 @@ import json
 
 BOOK_DIR = "data/books"
 PROGRESS_FILE = "data/memory/book_progress.json"
-READ_DELAY = 10  # seconds between lines
+READ_DELAY = 0.2  # seconds between lines
+PARAGRAPH_MIN_LENGTH = 20
 
 
 def get_books():
@@ -29,8 +30,6 @@ def save_progress(prog):
 async def read_books_forever():
     books = get_books()
     progress = load_progress()
-    buffered_lines = []
-
     while True:
         for book in books:
             path = os.path.join(BOOK_DIR, book)
@@ -41,26 +40,26 @@ async def read_books_forever():
                 lines = f.readlines()
 
             idx = progress.get(book, 0)
+            paragraph = ""
+
             while idx < len(lines):
                 line = lines[idx].strip()
                 idx += 1
+
+                if not line:
+                    if len(paragraph) > PARAGRAPH_MIN_LENGTH:
+                        train_on_message(paragraph.strip(), source="book")
+                        paragraph = ""
+                        await asyncio.sleep(READ_DELAY)
+                        set_next_action(READ_DELAY, "Reading")
+                else:
+                    paragraph += " " + line
+
                 progress[book] = idx
                 save_progress(progress)
 
-                if is_valid_line(line):
-                    buffered_lines.append(line)
-
-                    # If we have enough lines buffered, combine and train
-                    if len(buffered_lines) >= 3:
-                        combined_text = " ".join(buffered_lines)
-                        train_on_message(combined_text, source="book")
-                        buffered_lines.clear()
-
-                set_next_action(READ_DELAY, "Reading")
+            # train last paragraph if any
+            if paragraph and len(paragraph) > PARAGRAPH_MIN_LENGTH:
+                train_on_message(paragraph.strip(), source="book")
                 await asyncio.sleep(READ_DELAY)
-
-            # End of a book: train whatever lines are left buffered
-            if buffered_lines:
-                combined_text = " ".join(buffered_lines)
-                train_on_message(combined_text, source="book")
-                buffered_lines.clear()
+                set_next_action(READ_DELAY, "Reading")