谷歌专家Carlini：我是如何运用人工智能完成工作的

from flask import Flask, request, session, redirect, url_for, render_template
from sklearn.metrics import log_loss
from uuid import uuid4
import numpy as np


app = Flask(__name__)
app.config['SECRET_KEY'] = 'a very secret key'  # replace with your secret key


# Global variables
scores = {}  
entries = {}
questions = [("What is 1+1?", "2"), ("What is the capital of France?", "Paris")]
answers = [1, 0]  # Suppose for these questions GPT-4 got the first one right and the second one wrong




@app.route('/home', methods=['GET'])
def home():
    session['id'] = str(uuid4())
    return render_template('home.html')  # render a template that explains the game




@app.route('/question', methods=['GET'])
def question():
    qid = request.args.get('id', default=0, type=int)
    if 'id' in session:
        question, expected_answer = questions[qid]
        return render_template('question.html', question=question, expected_answer=expected_answer, id=qid)  
    else:
        return redirect(url_for('home'))




@app.route('/score', methods=['POST'])
def score():
    qid = request.form.get('id', type=int)
    guess = request.form.get('guess', type=float)
    user_id = session.get('id')


    if user_id and qid is not None and guess is not None:
        # Add to the list of all guesses for that question
        scores.setdefault(qid, []).append(guess)
        
        # Add to the person's guesses
        entries.setdefault(user_id, {})[qid] = guess


        # Compute the person's log loss
        user_log_loss = log_loss([answers[qid]], [guess], labels=[0, 1])


        # Compare with others
        other_scores = scores[qid]
        other_log_losses = log_loss([answers[qid]] * len(other_scores), other_scores, labels=[0, 1])


        comparison = 'better' if user_log_loss < other_log_losses else 'worse'


        return render_template('score.html', user_log_loss=user_log_loss, comparison=comparison)
    else:
        return redirect(url_for('home'))




if __name__ == "__main__":
    app.run(debug=True)