-
Notifications
You must be signed in to change notification settings - Fork 37
Expand file tree
/
Copy pathquiz.json
More file actions
78 lines (78 loc) · 2.72 KB
/
Copy pathquiz.json
File metadata and controls
78 lines (78 loc) · 2.72 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
{
"lesson": "29-moderation-systems-openai-perspective-llamaguard",
"title": "内容审核系统 —— OpenAI、Perspective、Llama Guard",
"questions": [
{
"stage": "pre",
"question": "OpenAI 的 omni-moderation-latest(2024)是什么?",
"options": [
"一个 RLHF 奖励模型",
"一个专注推理的前沿模型",
"一个 C2PA 水印端点",
"一个基于 GPT-4o 的审核分类器,可在一次调用中同时处理文本和图像,返回 13 类响应 schema(骚扰、仇恨、自残、性、暴力、违法,及其子类别),对大多数开发者免费"
],
"correct": 3,
"explanation": ""
},
{
"stage": "check",
"question": "以下哪一项不是 OpenAI Moderation 的主要顶层类别之一?",
"options": [
"violence(暴力)",
"election-interference(干预选举)",
"self-harm(自残)",
"harassment(骚扰)"
],
"correct": 1,
"explanation": ""
},
{
"stage": "check",
"question": "2026 年标准审核模式的三层是什么?",
"options": [
"Probe / detector / harness",
"Tokenizer / 模型 / detokenizer",
"望远镜式 / 潜望镜式 / 显微镜式",
"输入审核(生成前)、输出审核(生成后)、自定义审核(领域规则)"
],
"correct": 3,
"explanation": ""
},
{
"stage": "check",
"question": "Perspective API(Google Jigsaw)主要被设计用来对什么打分?",
"options": [
"水印强度",
"毒性(toxicity,带有严重毒性、侮辱、脏话、威胁、身份攻击等子维度),作为 LLM 时代之前的内容审核基线",
"推理质量",
"用户满意度奖励"
],
"correct": 1,
"explanation": ""
},
{
"stage": "post",
"question": "仅做输入审核的一种失败模式是什么?",
"options": [
"它无法在 GPU 上运行",
"它无法捕捉模型输出的失败或幻觉;编码攻击(第 12-14 课)可以完全绕过输入分类器",
"它总是拦截合法内容",
"它消除了对输出分类器的需求"
],
"correct": 1,
"explanation": ""
},
{
"stage": "post",
"question": "Azure Content Moderator 的弃用情况和迁移目标是什么?",
"options": [
"于 2024 年 2 月弃用,2027 年 3 月 15 日停服;迁移目标是 Azure AI Content Safety(基于 LLM,与 Azure OpenAI 集成)",
"被 Perspective API 取代",
"并入 Llama Guard 4",
"立即弃用且无替代品"
],
"correct": 0,
"explanation": ""
}
]
}