website·Indexed 31 days ago
https://www.aifeng.space/A personal blog featuring technical tutorials and reflections on cognitive growth. Topics include OpenAI function calling, Git troubleshooting (invalid sha1 pointer), the use of Discourse Graph in Roam Research, and memory techniques such as the Memory Palace.
OpenAI's function calling allows ChatGPT to generate structured data (JSON) to interact with custom functions. Its primary role is to extract specific function parameters from natural language, enabling developers to call their own code without passing the actual implementation to the model.
Key behavioral notes:
Bidirectional linking (bi-links) provides several functional advantages over traditional single-link or folder-based note-taking systems, particularly when notes are managed at the 'information block' level:
The Feynman Technique is a method of learning by explaining a concept in the simplest terms possible to ensure deep understanding and the ability to create cross-scenario associations. The process consists of four steps:
To reduce cognitive load and ensure notes meet the 'minimum information principle,' separate the process of standardized storage from the creation of atomized cards. This prevents the user from relying on the context of the source text when reviewing, forcing a more rigorous retrieval and synthesis process.
Key Workflow Components:
If you purchased a domain from a third-party provider (e.g., Alibaba Cloud) but are using Sanfengyun hosting, follow these steps to resolve the domain:
import openai
import json
# 1. Define the function description for the API
function_descriptions = [
{
"name": "get_student_score",
"description": "Get the student score by given his or her name",
"parameters": {
"type": "object",
"properties": {
"name": {
"type": "string",
"description": "The student's name",
}
},
"required": ["name"],
}
}
]
# 2. Define the actual execution function
def get_student_score(name):
"""Get the student score by given his or her name"""
# Mock data
SCORES = {"Lucy": 60}
score = {
"name": name,
"score": SCORES.get(name, "unknown")
}
return json.dumps(score)
# 3. First call: Extract parameters from user query
user_query = "What's the performance of Lucy in the school this year?"
response = openai.ChatCompletion.create(
model="gpt-3.5-turbo", # Example model
messages=[{"role": "user", "content": user_query}],
functions=function_descriptions,
function_call="auto"
)
ai_response_message = response["choices"][0]["message"]
# Extract the argument (e.g., 'Lucy')
arguments = json.loads(ai_response_message['function_call']['arguments'])
name = arguments.get("name")
# 4. Execute the local function
function_response = get_student_score(name)
# 5. Second call: Provide the function result to get a final natural language answer
second_response = openai.ChatCompletion.create(
model="gpt-3.5-turbo",
messages=[
{"role": "user", "content": user_query},
ai_response_message,
{
"role": "function",
"name": "get_student_score",
"content": function_response,
},
],
)
print(second_response['choices'][0]['message']['content'])
# Output: Lucy has achieved a score of 60 this year.