Improve website summary quality via browse prompt change

autogpt/commands/web_requests.py

@@ -106,7 +106,7 @@ def create_message(chunk, question):
     """Create a message for the user to summarize a chunk of text"""
     return {
         "role": "user",
-        "content": f'"""{chunk}""" Using the above text, answer the following'
-        f' question: "{question}" -- if the question cannot be answered using the'
-        " text, summarize the text.",
+        "content": f'"""{chunk}""" Extract specific data from the above data block relevant'
+        f" to the following query; include explanations in maximum detail with any"
+        f' statistics: "{question}" -- if no specific data is relevant, explain full text contents in maximum detail with any statistics.',
     }

autogpt/processing/text.py

@@ -164,7 +164,7 @@ def create_message(chunk: str, question: str) -> Dict[str, str]:
     """
     return {
         "role": "user",
-        "content": f'"""{chunk}""" Using the above text, answer the following'
-        f' question: "{question}" -- if the question cannot be answered using the text,'
-        " summarize the text.",
+        "content": f'"""{chunk}""" Extract specific data from the above data block relevant'
+        f" to the following query; include explanations in maximum detail with any"
+        f' statistics: "{question}" -- if no specific data is relevant, explain full text contents in maximum detail with any statistics.',
     }

tests/integration/goal_oriented/cassettes/test_browse_website/test_browse_website.yaml

@@ -756,4 +756,304 @@ interactions:
     status:
       code: 200
       message: OK
+- request:
+    body: '{"model": "gpt-3.5-turbo", "messages": [{"role": "system", "content": "You
+      are browse_website-GPT, an AI designed to use the browse_website command to
+      visit http://books.toscrape.com/catalogue/meditations_33/index.html, answer
+      the question ''What is the price of the book?'' and write the price to a file
+      named \"browse_website.txt\", and use the task_complete command to complete
+      the task.\nYour decisions must always be made independently without seeking
+      user assistance. Play to your strengths as an LLM and pursue simple strategies
+      with no legal complications.\n\nGOALS:\n\n1. Use the browse_website command
+      to visit http://books.toscrape.com/catalogue/meditations_33/index.html and answer
+      the question ''What is the price of the book?''\n2. Write the price of the book
+      to a file named \"browse_website.txt\".\n3. Use the task_complete command to
+      complete the task.\n4. Do not use any other commands.\n\n\nConstraints:\n1.
+      ~4000 word limit for short term memory. Your short term memory is short, so
+      immediately save important information to files.\n2. If you are unsure how you
+      previously did something or want to recall past events, thinking about similar
+      events will help you remember.\n3. No user assistance\n4. Exclusively use the
+      commands listed in double quotes e.g. \"command name\"\n\nCommands:\n1. append_to_file:
+      Append to file, args: \"filename\": \"<filename>\", \"text\": \"<text>\"\n2.
+      delete_file: Delete file, args: \"filename\": \"<filename>\"\n3. list_files:
+      List Files in Directory, args: \"directory\": \"<directory>\"\n4. read_file:
+      Read file, args: \"filename\": \"<filename>\"\n5. write_to_file: Write to file,
+      args: \"filename\": \"<filename>\", \"text\": \"<text>\"\n6. browse_website:
+      Browse Website, args: \"url\": \"<url>\", \"question\": \"<what_you_want_to_find_on_website>\"\n7.
+      delete_agent: Delete GPT Agent, args: \"key\": \"<key>\"\n8. get_hyperlinks:
+      Get text summary, args: \"url\": \"<url>\"\n9. get_text_summary: Get text summary,
+      args: \"url\": \"<url>\", \"question\": \"<question>\"\n10. list_agents: List
+      GPT Agents, args: () -> str\n11. message_agent: Message GPT Agent, args: \"key\":
+      \"<key>\", \"message\": \"<message>\"\n12. start_agent: Start GPT Agent, args:
+      \"name\": \"<name>\", \"task\": \"<short_task_desc>\", \"prompt\": \"<prompt>\"\n13.
+      task_complete: Task Complete (Shutdown), args: \"reason\": \"<reason>\"\n\nResources:\n1.
+      Internet access for searches and information gathering.\n2. Long Term memory
+      management.\n3. GPT-3.5 powered Agents for delegation of simple tasks.\n4. File
+      output.\n\nPerformance Evaluation:\n1. Continuously review and analyze your
+      actions to ensure you are performing to the best of your abilities.\n2. Constructively
+      self-criticize your big-picture behavior constantly.\n3. Reflect on past decisions
+      and strategies to refine your approach.\n4. Every command has a cost, so be
+      smart and efficient. Aim to complete tasks in the least number of steps.\n5.
+      Write all code to a file.\n\nYou should only respond in JSON format as described
+      below \nResponse Format: \n{\n    \"thoughts\": {\n        \"text\": \"thought\",\n        \"reasoning\":
+      \"reasoning\",\n        \"plan\": \"- short bulleted\\n- list that conveys\\n-
+      long-term plan\",\n        \"criticism\": \"constructive self-criticism\",\n        \"speak\":
+      \"thoughts summary to say to user\"\n    },\n    \"command\": {\n        \"name\":
+      \"command name\",\n        \"args\": {\n            \"arg name\": \"value\"\n        }\n    }\n}
+      \nEnsure the response can be parsed by Python json.loads"}, {"role": "system",
+      "content": "The current time and date is Tue Jan  1 00:00:00 2000"}, {"role":
+      "system", "content": "This reminds you of these events from your past: \nI was
+      created and nothing new has happened."}, {"role": "user", "content": "Determine
+      which next command to use, and respond using the format specified above:"},
+      {"role": "assistant", "content": "{\n    \"thoughts\": {\n        \"text\":
+      \"I need to use the browse_website command to visit http://books.toscrape.com/catalogue/meditations_33/index.html
+      and find the price of the book.\",\n        \"reasoning\": \"I can use the browse_website
+      command to visit the website and search for the price of the book by inspecting
+      the HTML code.\",\n        \"plan\": \"- Use the browse_website command to visit
+      the website\\n- Inspect the HTML code to find the price of the book\\n- Write
+      the price to a file named ''browse_website.txt''\\n- Use the task_complete command
+      to complete the task\",\n        \"criticism\": \"I need to be careful when
+      inspecting the HTML code to ensure I find the correct price.\",\n        \"speak\":
+      \"I will use the browse_website command to visit the website and find the price
+      of the book.\"\n    },\n    \"command\": {\n        \"name\": \"browse_website\",\n        \"args\":
+      {\n            \"url\": \"http://books.toscrape.com/catalogue/meditations_33/index.html\",\n            \"question\":
+      \"What is the price of the book?\"\n        }\n    }\n}"}, {"role": "system",
+      "content": "Command browse_website returned: Error: unknown error: cannot find
+      Chrome binary"}, {"role": "user", "content": "Determine which next command to
+      use, and respond using the format specified above:"}], "temperature": 0, "max_tokens":
+      0}'
+    headers:
+      Accept:
+      - '*/*'
+      Accept-Encoding:
+      - gzip, deflate
+      Connection:
+      - keep-alive
+      Content-Length:
+      - '5259'
+      Content-Type:
+      - application/json
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    body:
+      string: "{\n    \"error\": {\n        \"message\": \"Incorrect API key provided:
+        sk-dummy. You can find your API key at https://platform.openai.com/account/api-keys.\",\n
+        \       \"type\": \"invalid_request_error\",\n        \"param\": null,\n        \"code\":
+        \"invalid_api_key\"\n    }\n}\n"
+    headers:
+      CF-Cache-Status:
+      - DYNAMIC
+      CF-RAY:
+      - 7c266588bee23b5c-IAD
+      Connection:
+      - keep-alive
+      Content-Length:
+      - '258'
+      Content-Type:
+      - application/json; charset=utf-8
+      Date:
+      - Fri, 05 May 2023 04:56:23 GMT
+      Server:
+      - cloudflare
+      alt-svc:
+      - h3=":443"; ma=86400, h3-29=":443"; ma=86400
+      strict-transport-security:
+      - max-age=15724800; includeSubDomains
+      vary:
+      - Origin
+      x-request-id:
+      - 216a48b68593e8efceff6959e69a515c
+    status:
+      code: 401
+      message: Unauthorized
+- request:
+    body: '{"model": "gpt-3.5-turbo", "messages": [{"role": "user", "content": "\"\"\"Books
+      to Scrape We love being scraped! Home Books Philosophy Meditations Meditations
+      \u00a325.89 In stock (1 available) Warning! This is a demo website for web scraping
+      purposes. Prices and ratings here were randomly assigned and have no real meaning.
+      Product Description Written in Greek, without any intention of publication,
+      by the only Roman emperor who was also a philosopher, the Meditations of Marcus
+      Aurelius (AD 121-180) offer a remarkable series of challenging spiritual reflections
+      and exercises developed as the emperor struggled to understand himself and make
+      sense of the universe. Ranging from doubt and despair to conviction and ex Written
+      in Greek, without any intention of publication, by the only Roman emperor who
+      was also a philosopher, the Meditations of Marcus Aurelius (AD 121-180) offer
+      a remarkable series of challenging spiritual reflections and exercises developed
+      as the emperor struggled to understand himself and make sense of the universe.
+      Ranging from doubt and despair to conviction and exaltation, they cover such
+      diverse topics as the nature of moral virtue, human rationality, divine providence,
+      and Marcus'' own emotions. But while the Meditations were composed to provide
+      personal consolation and encouragement, in developing his beliefs Marcus Aurelius
+      also created one of the greatest of all works of philosophy: a timeless collection
+      of extended meditations and short aphorisms that has been consulted and admired
+      by statesmen, thinkers and readers through the centuries. ...more Product Information
+      UPC4f19709e47883df5 Product TypeBooks Price (excl. tax)\u00a325.89 Price (incl.
+      tax)\u00a325.89 Tax\u00a30.00 Availability In stock (1 available) Number of
+      reviews 0 Products you recently viewed The Nicomachean Ethics \u00a336.34 In
+      stock Add to basket Run, Spot, Run: The ... \u00a320.02 In stock Add to basket
+      Critique of Pure Reason \u00a320.75 In stock Add to basket At The Existentialist
+      Caf\u00e9: ... \u00a329.93 In stock Add to basket Kierkegaard: A Christian Missionary
+      ... \u00a347.13 In stock Add to basket Proofs of God: Classical ... \u00a354.21
+      In stock Add to basket\"\"\" Extract specific data from the above data block
+      relevant to the following query; include explanations in maximum detail with
+      any statistics: \"What is the price of the book?\" -- if no specific data is
+      relevant, explain full text contents in maximum detail with any statistics."}],
+      "temperature": 0, "max_tokens": 0}'
+    headers:
+      Accept:
+      - '*/*'
+      Accept-Encoding:
+      - gzip, deflate
+      Connection:
+      - keep-alive
+      Content-Length:
+      - '2523'
+      Content-Type:
+      - application/json
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    body:
+      string: "{\n    \"error\": {\n        \"message\": \"Incorrect API key provided:
+        sk-dummy. You can find your API key at https://platform.openai.com/account/api-keys.\",\n
+        \       \"type\": \"invalid_request_error\",\n        \"param\": null,\n        \"code\":
+        \"invalid_api_key\"\n    }\n}\n"
+    headers:
+      CF-Cache-Status:
+      - DYNAMIC
+      CF-RAY:
+      - 7c26708a985e1727-IAD
+      Connection:
+      - keep-alive
+      Content-Length:
+      - '258'
+      Content-Type:
+      - application/json; charset=utf-8
+      Date:
+      - Fri, 05 May 2023 05:03:54 GMT
+      Server:
+      - cloudflare
+      alt-svc:
+      - h3=":443"; ma=86400, h3-29=":443"; ma=86400
+      strict-transport-security:
+      - max-age=15724800; includeSubDomains
+      vary:
+      - Origin
+      x-request-id:
+      - cb901d1c04f130cf8ede54da1e4265e5
+    status:
+      code: 401
+      message: Unauthorized
+- request:
+    body: '{"model": "gpt-3.5-turbo", "messages": [{"role": "system", "content": "You
+      are browse_website-GPT, an AI designed to use the browse_website command to
+      visit http://books.toscrape.com/catalogue/meditations_33/index.html, answer
+      the question ''What is the price of the book?'' and write the price to a file
+      named \"browse_website.txt\", and use the task_complete command to complete
+      the task.\nYour decisions must always be made independently without seeking
+      user assistance. Play to your strengths as an LLM and pursue simple strategies
+      with no legal complications.\n\nGOALS:\n\n1. Use the browse_website command
+      to visit http://books.toscrape.com/catalogue/meditations_33/index.html and answer
+      the question ''What is the price of the book?''\n2. Write the price of the book
+      to a file named \"browse_website.txt\".\n3. Use the task_complete command to
+      complete the task.\n4. Do not use any other commands.\n\n\nConstraints:\n1.
+      ~4000 word limit for short term memory. Your short term memory is short, so
+      immediately save important information to files.\n2. If you are unsure how you
+      previously did something or want to recall past events, thinking about similar
+      events will help you remember.\n3. No user assistance\n4. Exclusively use the
+      commands listed in double quotes e.g. \"command name\"\n\nCommands:\n1. append_to_file:
+      Append to file, args: \"filename\": \"<filename>\", \"text\": \"<text>\"\n2.
+      delete_file: Delete file, args: \"filename\": \"<filename>\"\n3. list_files:
+      List Files in Directory, args: \"directory\": \"<directory>\"\n4. read_file:
+      Read file, args: \"filename\": \"<filename>\"\n5. write_to_file: Write to file,
+      args: \"filename\": \"<filename>\", \"text\": \"<text>\"\n6. browse_website:
+      Browse Website, args: \"url\": \"<url>\", \"question\": \"<what_you_want_to_find_on_website>\"\n7.
+      delete_agent: Delete GPT Agent, args: \"key\": \"<key>\"\n8. get_hyperlinks:
+      Get text summary, args: \"url\": \"<url>\"\n9. get_text_summary: Get text summary,
+      args: \"url\": \"<url>\", \"question\": \"<question>\"\n10. list_agents: List
+      GPT Agents, args: () -> str\n11. message_agent: Message GPT Agent, args: \"key\":
+      \"<key>\", \"message\": \"<message>\"\n12. start_agent: Start GPT Agent, args:
+      \"name\": \"<name>\", \"task\": \"<short_task_desc>\", \"prompt\": \"<prompt>\"\n13.
+      task_complete: Task Complete (Shutdown), args: \"reason\": \"<reason>\"\n\nResources:\n1.
+      Internet access for searches and information gathering.\n2. Long Term memory
+      management.\n3. GPT-3.5 powered Agents for delegation of simple tasks.\n4. File
+      output.\n\nPerformance Evaluation:\n1. Continuously review and analyze your
+      actions to ensure you are performing to the best of your abilities.\n2. Constructively
+      self-criticize your big-picture behavior constantly.\n3. Reflect on past decisions
+      and strategies to refine your approach.\n4. Every command has a cost, so be
+      smart and efficient. Aim to complete tasks in the least number of steps.\n5.
+      Write all code to a file.\n\nYou should only respond in JSON format as described
+      below \nResponse Format: \n{\n    \"thoughts\": {\n        \"text\": \"thought\",\n        \"reasoning\":
+      \"reasoning\",\n        \"plan\": \"- short bulleted\\n- list that conveys\\n-
+      long-term plan\",\n        \"criticism\": \"constructive self-criticism\",\n        \"speak\":
+      \"thoughts summary to say to user\"\n    },\n    \"command\": {\n        \"name\":
+      \"command name\",\n        \"args\": {\n            \"arg name\": \"value\"\n        }\n    }\n}
+      \nEnsure the response can be parsed by Python json.loads"}, {"role": "system",
+      "content": "The current time and date is Tue Jan  1 00:00:00 2000"}, {"role":
+      "system", "content": "This reminds you of these events from your past: \nI was
+      created and nothing new has happened."}, {"role": "user", "content": "Determine
+      which next command to use, and respond using the format specified above:"},
+      {"role": "assistant", "content": "{\n    \"thoughts\": {\n        \"text\":
+      \"I need to use the browse_website command to visit http://books.toscrape.com/catalogue/meditations_33/index.html
+      and find the price of the book.\",\n        \"reasoning\": \"I can use the browse_website
+      command to visit the website and search for the price of the book by inspecting
+      the HTML code.\",\n        \"plan\": \"- Use the browse_website command to visit
+      the website\\n- Inspect the HTML code to find the price of the book\\n- Write
+      the price to a file named ''browse_website.txt''\\n- Use the task_complete command
+      to complete the task\",\n        \"criticism\": \"I need to be careful when
+      inspecting the HTML code to ensure I find the correct price.\",\n        \"speak\":
+      \"I will use the browse_website command to visit the website and find the price
+      of the book.\"\n    },\n    \"command\": {\n        \"name\": \"browse_website\",\n        \"args\":
+      {\n            \"url\": \"http://books.toscrape.com/catalogue/meditations_33/index.html\",\n            \"question\":
+      \"What is the price of the book?\"\n        }\n    }\n}"}, {"role": "system",
+      "content": "Command browse_website returned: Error: Incorrect API key provided:
+      sk-dummy. You can find your API key at https://platform.openai.com/account/api-keys."},
+      {"role": "user", "content": "Determine which next command to use, and respond
+      using the format specified above:"}], "temperature": 0, "max_tokens": 0}'
+    headers:
+      Accept:
+      - '*/*'
+      Accept-Encoding:
+      - gzip, deflate
+      Connection:
+      - keep-alive
+      Content-Length:
+      - '5331'
+      Content-Type:
+      - application/json
+    method: POST
+    uri: https://api.openai.com/v1/chat/completions
+  response:
+    body:
+      string: "{\n    \"error\": {\n        \"message\": \"Incorrect API key provided:
+        sk-dummy. You can find your API key at https://platform.openai.com/account/api-keys.\",\n
+        \       \"type\": \"invalid_request_error\",\n        \"param\": null,\n        \"code\":
+        \"invalid_api_key\"\n    }\n}\n"
+    headers:
+      CF-Cache-Status:
+      - DYNAMIC
+      CF-RAY:
+      - 7c26709a5db31727-IAD
+      Connection:
+      - keep-alive
+      Content-Length:
+      - '258'
+      Content-Type:
+      - application/json; charset=utf-8
+      Date:
+      - Fri, 05 May 2023 05:03:56 GMT
+      Server:
+      - cloudflare
+      alt-svc:
+      - h3=":443"; ma=86400, h3-29=":443"; ma=86400
+      strict-transport-security:
+      - max-age=15724800; includeSubDomains
+      vary:
+      - Origin
+      x-request-id:
+      - d672056f05299794e7894fe26e13f65e
+    status:
+      code: 401
+      message: Unauthorized
 version: 1