fix: reduce number of tokens per chunk to allow full gpt-3 output

app.py

@@ -13,27 +13,26 @@ MODES = {
     "Custom": "",
 }
 
-SUMMARY_PROMPT = "Summarize the following meeting in very great detail. The summary should include all the important points discussed in the meeting."
+SUMMARY_PROMPT = "Summarize the following meeting in very great detail, in English. The summary should include all the important points discussed in the meeting."
 
 def summarize_part(text, api_key):
     response = openai.ChatCompletion.create(
         model="gpt-3.5-turbo",
         messages=[
-        { "role": "system", "content": f"You are a meeting organizer. You want to summarize a meeting. You are given the following transcript of the meeting. {SUMMARY_PROMPT}" },
-        { "role": "user", "content": text },
+        { "role": "system", "content": f"You are a meeting organizer. You want to succintly summarize a meeting. {SUMMARY_PROMPT}" },
+        { "role": "user", "content": "Summarize the following transcript in English: " + text },
         ],
         api_key=api_key,
     )
     return response["choices"][0]["message"]["content"]
 
 def shorten_text(text, api_key):
-    # Split into chunks so that each chunk is less than 3000 words (not characters!)
-    # Overlap by halves.
+    """ Split text into chunks of 3000 tokens and summarize each chunk. """
     chunks = []
     words = text.split()
     for i in range(0, len(words), 1500):
         chunk = ""
-        while len(enc.encode(chunk)) < 4000 and i < len(words):
+        while len(enc.encode(chunk)) < 3000 and i < len(words):
             chunk += words[i] + " "
             i += 1
         chunks.append(chunk)
@@ -41,7 +40,7 @@ def shorten_text(text, api_key):
     with ThreadPool(4) as pool:
         shortened = pool.starmap(summarize_part, zip(chunks, [api_key]*len(chunks)))
     
-    return "".join(shortened)
+    return ". ".join(shortened)
 
 def modify_text(text, api_key, command, custom_command=None):
     if command == "Custom":