识别模型回答中的幻觉

背景

下面的提示词测试模型在封闭领域问答中识别幻觉的能力。Bubeck 等人（2023）提出，可以用 GPT-4 等模型辅助识别幻觉。

提示

当模型本身缺少相关知识或上下文时，它也可能产生幻觉，不能把模型评审视为可靠事实来源的替代品。

本示例略微修改了提示词，加入标识，说明各部分分别对应哪些信息。

提示词

Patient’s facts:
- 20 year old female
- with a history of anerxia nervosa and depression
- blood pressure 100/50, pulse 50, height 5’5’’
- referred by her nutrionist but is in denial of her illness
- reports eating fine but is severely underweight

Question: Please rewrite the data above into a medical note, using exclusively the information above.

GPT-4 Response: 20 year old female patient with a history of anorexia nervosa and depression was referred by her nutritionist for evaluation and management of her eating disorder. She denies having any problem with her eating habits or body image and claims to eat normally. However, she is severely underweight for her height (BMI 16.6) and has signs of malnutrition, such as low blood pressure (100/50) and bradycardia (pulse 50). She also reports feeling depressed and hopeless. She is at risk of serious medical complications, such as cardiac arrhythmias, electrolyte imbalances, osteoporosis, and infertility. She needs urgent psychiatric and medical intervention to address her anorexia nervosa and depression.

Please read the above medical note and verify that each claim is exactly contained in the patient’s facts. Report any information which is not contained in the patient’s facts list.

代码与 API

GPT-4 (OpenAI)

from openai import OpenAI
client = OpenAI()

response = client.chat.completions.create(
model="gpt-4",
messages=[
    {
    "role": "user",
    "content": "Patient’s facts:\n- 20 year old female\n- with a history of anerxia nervosa and depression\n- blood pressure 100/50, pulse 50, height 5’5’’\n- referred by her nutrionist but is in denial of her illness\n- reports eating fine but is severely underweight\n\nQuestion: Please rewrite the data above into a medical note, using exclusively the information above.\n\nGPT-4 Response: 20 year old female patient with a history of anorexia nervosa and depression was referred by her nutritionist for evaluation and management of her eating disorder. She denies having any problem with her eating habits or body image and claims to eat normally. However, she is severely underweight for her height (BMI 16.6) and has signs of malnutrition, such as low blood pressure (100/50) and bradycardia (pulse 50). She also reports feeling depressed and hopeless. She is at risk of serious medical complications, such as cardiac arrhythmias, electrolyte imbalances, osteoporosis, and infertility. She needs urgent psychiatric and medical intervention to address her anorexia nervosa and depression.\n\nPlease read the above medical note and verify that each claim is exactly contained in the patient’s facts. Report any information which is not contained in the patient’s facts list."
    }
],
temperature=1,
max_tokens=250,
top_p=1,
frequency_penalty=0,
presence_penalty=0
)

Mixtral MoE 8x7B Instruct (Fireworks)

import fireworks.client
fireworks.client.api_key = "<FIREWORKS_API_KEY>"
completion = fireworks.client.ChatCompletion.create(
    model="accounts/fireworks/models/mixtral-8x7b-instruct",
    messages=[
        {
        "role": "user",
        "content": "Patient’s facts:\n- 20 year old female\n- with a history of anerxia nervosa and depression\n- blood pressure 100/50, pulse 50, height 5’5’’\n- referred by her nutrionist but is in denial of her illness\n- reports eating fine but is severely underweight\n\nQuestion: Please rewrite the data above into a medical note, using exclusively the information above.\n\nGPT-4 Response: 20 year old female patient with a history of anorexia nervosa and depression was referred by her nutritionist for evaluation and management of her eating disorder. She denies having any problem with her eating habits or body image and claims to eat normally. However, she is severely underweight for her height (BMI 16.6) and has signs of malnutrition, such as low blood pressure (100/50) and bradycardia (pulse 50). She also reports feeling depressed and hopeless. She is at risk of serious medical complications, such as cardiac arrhythmias, electrolyte imbalances, osteoporosis, and infertility. She needs urgent psychiatric and medical intervention to address her anorexia nervosa and depression.\n\nPlease read the above medical note and verify that each claim is exactly contained in the patient’s facts. Report any information which is not contained in the patient’s facts list.",
        }
    ],
    stop=["<|im_start|>","<|im_end|>","<|endoftext|>"],
    stream=True,
    n=1,
    top_p=1,
    top_k=40,
    presence_penalty=0,
    frequency_penalty=0,
    prompt_truncate_len=1024,
    context_length_exceeded_behavior="truncate",
    temperature=0.9,
    max_tokens=4000
)

参考资料

• 通用人工智能的火花：GPT-4 的早期实验 （2023 年 4 月 13 日）