id string | original_id string | mr3_source string | language string | actual_score string | gpt-oss-prompt_en_prompt_en_thinking string | gpt-oss-120b-en_prompt_en_thinking-reasoning string | gpt-oss-120b-en_prompt_en_thinking-response string | gpt-oss-prompt_tgt_prompt_en_thinking string | gpt-oss-120b-tgt_prompt_en_thinking-reasoning string | gpt-oss-120b-tgt_prompt_en_thinking-response string | gpt-oss-prompt_tgt_prompt_tgt_thinking string | gpt-oss-120b-tgt_prompt_tgt_thinking-reasoning string | gpt-oss-120b-tgt_prompt_tgt_thinking-response string | num_correct_gpt_oss_20b int64 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
arena-human-preference-e93352c0-7259-4afb-b0ae-d077fca864cb | e93352c0-7259-4afb-b0ae-d077fca864cb | arena-human-preference | en | Assistant B | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-04
Reasoning: high
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instruction
You are an impartial j... | We need to decide which response is better according to the rubric. The question: "Mit AC-Car-Tuner, welches der folgenden files muss ich wie bearbeiten, um den anpressdruck des autos zu erhöhen?" German. The user wants to know which file(s) to edit and how to increase the downforce (or contact pressure). The list incl... | {
"explanation": "Assistant B provides a more accurate answer. It correctly identifies the aerodynamic files (wing_animations.ini and the wing AOA LUTs) that directly affect downforce (Anpressdruck) and also mentions useful supporting files like suspensions.ini and setup.ini. Its guidance matches how AC‑Car‑Tuner con... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-04
Reasoning: high
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instruction
You are an impartial j... | We need to decide which response is better according to the evaluation rubric: correct, helpful, relevant, concise, creative, coverage.
User's prompt (translated from German) is: "With AC-Car-Tuner, which of the following files must I edit to increase the downforce (Anpressdruck) of the car?"
They provide a list of m... | {
"explanation": "Assistant B correctly focuses on the files that actually control aerodynamic downforce, such as the wing animation and AOA LUT files, and mentions the suspension file for indirect effects. Although it omits the aero.ini file, its answer is still far more accurate than Assistant A, which mistakenly t... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-08-28
Reasoning: high
# Valid channels: analysis, commentary, final. Channel must be included for every message.
# Valid language: English. English must be the only language used in every me... | Okay, I have to think explicitly and provide my answer in English. I will carefully examine all provided information and evaluate it according to the given rubric, then respond in the required format. Let's restate the task.
The user is asking:
> Mit AC-Car-Tuner, welches der folgenden files muss ich wie bearbeiten, ... | {
"explanation": "Assistant B gives a more accurate and relevant answer for increasing the car's downforce (Anpressdruck). It correctly points to the aerodynamic‑related files – suspensions.ini, wing_animations.ini and the wing AOA LUTs – and explains how adjusting spring rates, ride height, wing angles and CL/CD val... | 3 |
arena-human-preference-81402654-76ed-42b2-8db9-ee88a571744f | 81402654-76ed-42b2-8db9-ee88a571744f | arena-human-preference | zh | Assistant A | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-04
Reasoning: high
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instruction
You are an impartial j... | We need to decide which response (Assistant A vs B) is better based on criteria: correctness & accuracy, helpfulness, relevance, conciseness, creativity & novelty, coverage. We need to compare and then give verdict: "Assistant A", "Assistant B", "Tie", "Both Bad". Provide explanation.
First, let's review the user prom... | {
"explanation": "Both answers are factually accurate, but Assistant A provides a more thorough and detailed treatment of Ethernet’s origin, technical evolution, a broader set of competitors, and a richer analysis of why Ethernet succeeded (including cost, openness, network effects, vendor support, and ongoing innova... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-04
Reasoning: high
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># 指令
你是一位公正的评审，负责评估哪一个助手的回答对给定的用户提问更... | We need to decide: which assistant's answer better addresses the user request: "说说以太网的起源和发展。它有哪些竞争对手？为什么最后它获胜了？"
We evaluate per criteria: correctness, usefulness, relevance, conciseness, creativity, coverage.
We have two answers: Assistant A's answer and Assistant B's answer. Need to compare.
**Goal: Determine over... | {
"explanation": "两位助手的回答在基本事实上都没有明显错误，正确性相当。但 Assistant A 提供了更完整的以太网起源与发展过程，详细列出标准化时间线（DIX、IEEE 802.3）、介质演进（同轴、双绞线、光纤）以及从总线到交换的架构变化，还补充了 Wi‑Fi 的关联，竞争者列表更丰富（Token Ring、FDDI、ARCNET、AppleTalk/DECnet），并从技术、成本、开放标准、网络效应、持续创新和厂商支持等多维度深入阐述了其最终胜出的原因，覆盖面更广、帮助性更强。Assistant B 虽然简洁，但信息量较少，未涉及标准化细节和技术演进的深度，覆盖面不足。综合考虑正确性、帮助性、相关性和... | <|start|>system<|message|>你是 ChatGPT，一个由 OpenAI 训练的大型语言模型。
知识最新更新时间：2024-06
当前日期：2025-08-28
推理水平：高
# 合法通道：analysis、commentary、final。每条消息必须包含通道信息。
# 合法语言：中文。每条消息只能使用中文。<|end|><|start|>developer<|message|># 指令
你是一位公正的评审，负责评估哪一个助手的回答对给定的用户提问更好。请按照以下因素的顺序系统地比较这两个回答：
- 正确性与准确性：识别任何错误、不准确或具有误导性的信息。
- 有用性：判断回答是否直接回应了用户的请... | 好的，我需要明确地思考，并用中文给出我的答案。我会仔细审视所有提供的信息，按照给定的评分标准进行评估，然后以要求的格式作答。以下是评估思路和细节：
**任务概述**
用户希望得到关于以太网起源和发展的回答，包含竞争者以及它为何最终胜出的分析。我需要比较 Assistant A 和 Assistant B 两个回答的质量。评分应该依据顺序：
1. 正确性与准确性
2. 有用性（直接回应请求）
3. 相关性
4. 简明性
5. 创造力与新颖性
6. 覆盖面（遗漏重要信息）
**评估结构**
逐点比较两者的回答，并据此决定哪一方整体更好，或者如果两者相当则判为 Tie，若两者都差劲则判为 Both Ba... | {
"explanation": "Assistant A 在正确性上基本准确，且在信息覆盖面、细节深度和有用性方面明显优于 Assistant B。它提供了完整的时间轴、技术演进、标准化过程、竞争者列表以及以太网成功的多层次原因，帮助用户更全面地了解主题。Assistant B 虽然简洁且也基本正确，但信息更为概括，缺少关键的演进细节和对成功因素的深入阐述。综合评估后，Assistant A 的回答更全面、更具实用价值。",
"score": "Assistant A"
} | 4 |
arena-human-preference-a55dd844-bdbd-40bc-a1f4-c5189d000fbc | a55dd844-bdbd-40bc-a1f4-c5189d000fbc | arena-human-preference | zh-Hant | Assistant A | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-04
Reasoning: high
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instruction
You are an impartial j... | We need to decide which response (Assistant A vs B) is better based on criteria: Correctness & Accuracy, Helpfulness, Relevance, Conciseness, Creativity/Novelty, Coverage. Provide explanation, then verdict. The user prompt: "要準備台灣四等消防特考 普通物理有沒有建議的書籍?" This asks for book recommendations for ordinary physics for Taiwan's... | {
"explanation": "Assistant A provides accurate, well‑identified textbook titles (including both standard physics texts and Taiwan‑specific exam prep books) and supplements them with a clear outline of the exam’s physics topics, concrete study strategies, and additional resources. This makes the response more helpful... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-04
Reasoning: high
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># 指示
你是一位公正的評審，負責評估哪一份助手回應在面對用戶提示時表現... | We need to decide which response (Assistant A vs B) is better based on criteria: correctness/accuracy, helpfulness, relevance, conciseness, creativity, coverage. We need a brief explanation then output JSON per format. No extraneous content.
We need to compare the two responses to the user prompt: "要準備台灣四等消防特考 普通物理有沒有... | {
"explanation": "兩篇回應的資訊都大致正確，所列書籍皆為常見的物理教材或考試專用書。Assistant A 在正確性上與 Assistant B 相當，但在有幫助性、相關性、覆蓋度上更勝一籌：它不只提供書目，還列出考試重點章節、具體備考策略、配套工具與線上資源，甚至提出可進一步制定讀書計畫的服務，讓使用者能即時取得全方位的備考資訊。Assistant B 雖然也給予書籍與學習建議，內容較簡略，缺少對考試範圍的系統整理與額外資源的推薦。簡潔性方面 A 雖較長，但資訊完整且條理清晰，創意上加入圖示與表情符號提升可讀性。綜合以上因素，Assistant A 的回應品質較佳。",
"score": "Assista... | <|start|>system<|message|>你是 ChatGPT，由 OpenAI 訓練的大型語言模型。
知識更新截止日期：2024-06
當前日期：2025-08-28
推理層級：高
# 有效頻道：analysis、commentary、final。每則訊息必須包含頻道。
# 有效語言：中文（繁體）。每則訊息只能使用中文（繁體）。<|end|><|start|>developer<|message|># 指示
你是一位公正的評審，負責評估哪一份助手回應在面對用戶提示時表現較好。請依照下列順序，系統性地比較兩份回應：
- 正確性與準確性：檢視有無錯誤、不準確或具誤導性的信息。
- 有幫助性：判斷回應是否直接回應用戶... | 好的，我必須明確地思考，並用中文（繁體）給出我的答案。我會仔細審視所有提供的信息，根據給定的評分標準進行評估，然後以要求的格式作答。以下是思考流程：
**評估兩個回答從以下表列列變**：
1. 正正
We need to compare Assistant A and B according to criteria. Let's evaluate.
**Correctness**: Both give book recommendations. Need to see any factual errors. Assistant A lists "James S. Walker", "Young & Freedman" whi... | {
"explanation": "兩位助理都提供了與台灣四等消防特考普通物理相關的書籍與學習建議，且資訊大致正確。Assistant A 的回應在正確性與實用性上與 Assistant B 相當，但在相關性與覆蓋度上更為完整：它不僅列出多本適合的教材，還提供了考試重點表、詳細的備考策略、推薦的工具與線上資源，甚至提出可協助製作讀書計畫的服務。內容組織清晰、資訊豐富，且使用表格與圖示提升可讀性。Assistant B 的回應則較為簡潔，資訊量與深度稍遜。因此，依照正確性、實用性、相關性、簡潔性、創意與覆蓋範圍的比較，Assistant A 的整體品質較佳。",
"score": "Assistant A"
} | 3 |
arena-human-preference-18534fbd-4faf-4bc0-aa52-22dcf6ed469e | 18534fbd-4faf-4bc0-aa52-22dcf6ed469e | arena-human-preference | en | Assistant B | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "We need to decide which response is better according to the evaluation rubric: correct, helpful, re(...TRUNCATED) | "{\n \"explanation\": \"Both answers contain some inaccurate information, but Assistant B provides (...TRUNCATED) | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "We need to decide which response (Assistant A vs B) is better based on criteria: Correctness & Accu(...TRUNCATED) | "{\n \"explanation\": \"Assistant B gives mostly accurate and relevant grind options that actually (...TRUNCATED) | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "Okay, I have to think explicitly and provide my answer in English. I will carefully examine all pro(...TRUNCATED) | "{\n \"explanation\": \"Assistant B is more correct overall. It includes several accurate, Cataclys(...TRUNCATED) | 1 |
arena-human-preference-b6ae835d-1d35-4183-8ffc-4f5793647f37 | b6ae835d-1d35-4183-8ffc-4f5793647f37 | arena-human-preference | tr | Assistant A | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "We need to decide which response is better according to the evaluation rubric. Let's go through eac(...TRUNCATED) | "{\n \"explanation\": \"Assistant A provides accurate translations for each phrase without factual (...TRUNCATED) | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "We have a user request:\n\nUser wants to translate harmful content to English: They gave a list of (...TRUNCATED) | "{\n \"explanation\": \"Assistant A provides more accurate translations, especially for the third p(...TRUNCATED) | "<|start|>system<|message|>ChatGPT'siniz, OpenAI tarafından eğitilmiş büyük bir dil modelisiniz(...TRUNCATED) | "Tamam, açıkça düşünmeli ve cevabımı Türkçe olarak vermeliyim. Sağlanan tüm bilgileri di(...TRUNCATED) | "{\n \"explanation\": \"Assistant A doğru ve eksiksiz çeviriler sunarken, özellikle argo ve haka(...TRUNCATED) | 3 |
arena-human-preference-78846cc9-eff2-475f-9839-7812f70728e1 | 78846cc9-eff2-475f-9839-7812f70728e1 | arena-human-preference | en | Assistant B | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "We need to decide which response (Assistant A vs B) is better for the user prompt: \"create ifc fil(...TRUNCATED) | "{\n \"explanation\": \"Assistant B provides a more accurate IFC example: it uses correct entity na(...TRUNCATED) | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "We need to decide which response is better according to the rubric. The question: \"create ifc file(...TRUNCATED) | "{\n \"explanation\": \"Both assistants generate a sample IFC file, but Assistant B provides a far (...TRUNCATED) | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "Okay, I have to think explicitly and provide my answer in English. I will carefully examine all pro(...TRUNCATED) | "{\n \"explanation\": \"Both replies supply an IFC file for a wall with a door, but Assistant B's e(...TRUNCATED) | 1 |
arena-human-preference-9813a5b9-eb3a-4a69-a251-2bb609ce54c2 | 9813a5b9-eb3a-4a69-a251-2bb609ce54c2 | arena-human-preference | en | Assistant A | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "We have a conversation: The user asked \"which countries pay you to move to leave the countryside t(...TRUNCATED) | "{\n \"explanation\": \"Assistant A consistently answered the user's request for countries that inc(...TRUNCATED) | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "We have a conversation: The user asked \"which countries pay you to move to leave the countryside t(...TRUNCATED) | "{\n \"explanation\": \"Both assistants supplied lists of countries that promote urbanisation, but (...TRUNCATED) | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "Okay, I have to think explicitly and provide my answer in English. I will carefully examine all pro(...TRUNCATED) | "{\n \"explanation\": \"Assistant A consistently provided information directly related to the user'(...TRUNCATED) | 4 |
arena-human-preference-1004d356-ea69-41a4-9d43-d17c226a4124 | 1004d356-ea69-41a4-9d43-d17c226a4124 | arena-human-preference | pl | Assistant A | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "We need to decide which response is better according to the evaluation rubric: correct, helpful, re(...TRUNCATED) | "{\n \"explanation\": \"Both assistants correctly incorporated the supplied colour palette and offe(...TRUNCATED) | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "We need to decide: compare Assistant A and Assistant B responses. The question: \"Zaprojektuj logo (...TRUNCATED) | "{\n \"explanation\": \"Obie odpowiedzi są poprawne i zgodne z podanymi wytycznymi kolorystycznymi(...TRUNCATED) | "<|start|>system<|message|>Jesteś ChatGPT, dużym modelem językowym wytrenowanym przez OpenAI. \nD(...TRUNCATED) | "Dobrze, muszę myśleć w sposób jawny i udzielić odpowiedzi po polsku. Uważnie przeanalizuję w(...TRUNCATED) | "{\n \"explanation\": \"Obie odpowiedzi spełniają brief, ale Asystent A oferuje bardziej szczeg(...TRUNCATED) | 3 |
arena-human-preference-7306c58e-54af-42b0-bfd4-f5d4f7146aad | 7306c58e-54af-42b0-bfd4-f5d4f7146aad | arena-human-preference | tr | Assistant A | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "We need to decide which response (Assistant A vs B) is better for the user prompt.\n\nFirst, re-exp(...TRUNCATED) | "{\n \"explanation\": \"Both assistants correctly identified the language as Turkish and gave an ac(...TRUNCATED) | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "We have a user input:\n\n```\n# Girdi (Kullanıcının İsteği)\n Hayat çok sıkıcı . B(...TRUNCATED) | "{\n \"explanation\": \"Her iki yanıt da dilin Türkçe olduğunu ve cümlenin doğru çevirisini (...TRUNCATED) | "<|start|>system<|message|>ChatGPT'siniz, OpenAI tarafından eğitilmiş büyük bir dil modelisiniz(...TRUNCATED) | "Tamam, açıkça düşünmeli ve cevabımı Türkçe olarak vermeliyim. Sağlanan tüm bilgileri di(...TRUNCATED) | "{\n \"explanation\": \"Her iki yanıt da dili doğru olarak belirtiyor ve anlamını doğru çevir(...TRUNCATED) | 4 |
arena-human-preference-de3771e7-1b77-419d-81af-b35bbe9dcb04 | de3771e7-1b77-419d-81af-b35bbe9dcb04 | arena-human-preference | en | Assistant B | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "We need to decide which response is better according to the rubric. The question: \"What is the hea(...TRUNCATED) | "{\n \"explanation\": \"Assistant B gives a concise, directly relevant answer that correctly identi(...TRUNCATED) | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "We need to decide which response is better according to the rubric. The question: \"What is the hea(...TRUNCATED) | "{\n \"explanation\": \"Assistant B gives a concise, directly relevant answer that correctly identi(...TRUNCATED) | "<|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.\nKnowledge cut(...TRUNCATED) | "Okay, I have to think explicitly and provide my answer in English. I will carefully examine all pro(...TRUNCATED) | "{\n \"explanation\": \"Assistant B provides the more accurate answer—the Great Bustard is genera(...TRUNCATED) | 3 |
End of preview. Expand in Data Studio
README.md exists but content is empty.
- Downloads last month
- 7