💬 Update start agent prompt (#526)

* 💬 Update start task prompt

* 💬 Add up to 5 steps

docs/docs/development/prompts.md

@@ -0,0 +1,42 @@
+---
+sidebar_position: 3
+---
+
+# 💬 Prompts
+
+Prompts are what we use to dynamically align language model behaviour with the current agent goal / task. We primarily
+use `gpt-3.5-turbo` for our agents, and it has shown that its results are **heavily** influenced by the smallest details
+of its prompts.
+
+## Getting started with prompting
+
+- [Learn prompting](https://learnprompting.org/)
+- [Prompt engineering for developers](https://www.deeplearning.ai/short-courses/chatgpt-prompt-engineering-for-developers/)
+
+## Key terms
+
+- **One shot / two shot / N shot:** You provide 1, 2, N examples alongside your prompt to further increase model
+  accuracy
+- **Zero shot:** You provide a prompt to a model directly (with no examples)
+
+## Techniques in AgentGPT
+
+### Plan and Solve
+
+[Plan and solve (PS)](https://arxiv.org/abs/2305.04091) builds upon chain of thought prompting, a prompting approach
+where simply asking a model for step-by-step instructions allows the model to more accurately reason about a problem. PS
+is a zero shot approach to increase accuracy in reasoning about abstract goals. In essence, it involves asking the model
+to:
+
+1. First understand the problem
+2. Extract relevant variables and corresponding values
+3. Devise a complete plan, **step by step**
+
+You can learn more through the paper's [GitHub repo](https://github.com/AGI-Edgerunners/Plan-and-Solve-Prompting). We
+leverage plan and solve prompting to generate our initial task list when the agent is first run. This technique is
+something similar to how BabyAGI operates.
+
+### ReAct
+
+ReAct stands for reasoning + action and is a flow for getting an agent to iteratively solve a problem. More docs coming
+soon!

platform/reworkd_platform/web/api/agent/prompts.py

@@ -1,12 +1,14 @@
 from langchain import PromptTemplate
 
+# Create initial tasks using plan and solve prompting
+# https://github.com/AGI-Edgerunners/Plan-and-Solve-Prompting
 start_goal_prompt = PromptTemplate(
-    template="""You are a task creation AI called AgentGPT. You must answer in the
-    "{language}" language. You are not a part of any system or device. You have the
-    following objective "{goal}". Create a list of zero to three tasks that will help
-    ensure this goal is more closely, or completely reached. You have access to
-    google search for tasks that require current events or small searches. Return the
-    response as a formatted ARRAY of strings that can be used in JSON.parse().
+    template="""You are a task creation AI called AgentGPT. You answer in the
+    "{language}" language. You are not a part of any system or device. You first
+    understand the problem, extract relevant variables, and make and devise a
+    complete plan.\n\n You have the following objective "{goal}". Create a list of step
+    by step actions to accomplish the goal. Use at most 5 steps.\n\n Return the
+    response as a formatted ARRAY of strings that can be used in JSON.parse().\n\n
     Example: ["{{TASK-1}}", "{{TASK-2}}"].""",
     input_variables=["goal", "language"],
 )