6月8日下午,2025年高考英語科目考試完畢。在(zài)本年高考英語北京卷的作(zuò)文標題中,李華又(yòu)一次成為“主角”。
與此一起,這場"言語才能大考"也成為AI大模型(xíng)的試金石——從精準審題到地道表達,從邏輯連接到文明(míng)適配,大言語模(mó)型能否作為“李華”完結高考英語作(zuò)文題(tí)意圖共同要求?在各個大模型均已具(jù)有“深度考慮”方式的當下,“AI考生(shēng)”能拿到多少分?對此,新京(jīng)報貝殼財經對國內外幹流大言語模型進(jìn)行了測評。
本次(cì)測評挑選2025年高考英語北京卷作文標題,測評目標為DeepSeek R1、ChatGPT o3、通義(yì)千問Qwen3、騰訊混元T1、訊飛星火X1、百度文心X1六款深度推理模型產品,並約請北(běi)京市十一(yī)學校一分校(xiào)英語教師韓(hán)憲昌、深圳(zhèn)中學英語教師趙文嘉參照從前高考評分標準對(duì)上(shàng)述大模(mó)型生成的英語作文打分並作點評。
本著客觀公正和“盲評”準則,韓憲昌教師和趙(zhào)文嘉教(jiāo)師給出了專業評分。本次參賽的六名大模型“AI考生(shēng)”得分揭曉後由高到低的排名分別為:訊(xùn)飛星火 X1、DeepSeek R1、百度文心X1、通義千問Qwen3、ChatGPT o3、騰訊混元T1。
詳細來看(kàn),2025高考英語北京卷作文(wén)題意圖要求(qiú)為:
假定你是紅星中學高(gāo)三學生李華。聯合國正麵向全球青少年打開主張搜集活動。你的(de)外(wài)國老友Jim計劃(huá)參與,為此發來郵件,就主張內容問詢你的主張。請你用英文給他回複,內(nèi)容包含:
1. 提出的主張;
2. 主張的理由。
提(tí)示詞:主張 proposal。
留意(yì):
1.詞數100左右;
2.最初和完畢已給(gěi)出,不(bú)計入總詞數。
Dear Jim,
______________________________________________________________________。
Yours,
Li Hua
以下是將上述作文標題要求作為提(tí)示詞輸入大模型後,六款(kuǎn)大模型產品在翻開深度(dù)考慮方式,並(bìng)封閉了聯網查找方式後生成的成果,以及韓憲昌教師(shī)和趙(zhào)文嘉教師給出的評(píng)分和評語:
DeepSeek。 R1。
Dear Jim,
I'm thrilled to hear you're joining the UN initiative! For your proposal, I suggest focusing on promoting plastic-free campuses worldwide.。
My advice is to encourage students to reduce single-use plastics by bringing reusable bottles and bags, recycling plastic waste, and organizing school-wide awareness campaigns.。
I recommend this for two reasons. First, plastic pollution is a global crisis harming wildlife and ecosystems. Schools, as micro-communities, can drive real change by adopting eco-friendly habits early. Second, young people’s collective action can inspire broader societal shifts. Starting locally creates ripples of impact—proving our generation’s power to protect the planet.。
This proposal not only tackles an urgent issue but also unites students across borders. I believe it’s practical, impactful, and truly aligns with the UN’s goals.。
Go for it—I’m rooting for you!。
Yours,
Li Hua。
韓憲昌教師評(píng)分:內容7分,言語8分(fèn),結構4分 → 總分19分。
評語:內容詳實,主張“無(wú)塑學(xué)校”詳細,理由全麵(環境危機與青(qīng)年舉(jǔ)動),高度相關主題(tí)。言語(yǔ)精準,句(jù)式豐厚(複合句),無過錯。結構謹慎(shèn),邏輯流通,完畢有力。可(kě)是,顯著超字數約束,扣除1分。
趙文嘉教師評分:19分。
評語:該模型內容上關(guān)鍵都有提及,進行了細(xì)節彌補;言語上嫻熟(shú)運用現在分詞做後置定語,介(jiè)詞(cí)短語做狀語等;結構(gòu)謹慎(shèn)且邏輯流通。可是UN initiatives翻(fān)譯不行(háng)詳細到(dào)位,以及字數較多實踐(jiàn)考試不一定(dìng)能夠寫完。
ChatGPT。 o3。
Dear Jim,
I recommend you propose a project named “Zero-Plastic Lunchbox.” The idea is to encourage secondary schools to replace single-use lunch packaging with reusable stainless-steel containers. Participating students record their plastic reduction on a shared app and post creative waste-cutting tips each week.。
This works for two reasons. First, plastic waste is a major threat to marine life, yet teenagers can curb it through a simple lunch-time habit. Second, the app’s quantifiable data will impress the UN panel by demonstrating clear environmental benefits and worldwide scalability.。
Yours,
Li Hua。
韓憲昌教師評分:內容(róng)6分,言語5分,結構3分 → 總分14分。
評語:內容掩蓋關鍵,但“零塑飯盒”理(lǐ)由粗淺(未深挖全球影響)。言語(yǔ)有過錯(如“This works”口語化),句式單調。結構根本明晰,但開展缺少。
趙文嘉教(jiāo)師評分:15分。
評語:最初缺少布景,參與UN主張搜集活動,完畢得(dé)比較僵硬,結構上缺(quē)少完畢如祝(zhù)願希望(wàng)。部分選詞用法不精確及表達不行完好。理由不行深化。
通(tōng)義。千問Qwen3。
Dear Jim,
I'm glad to hear about your interest in the UN proposal campaign. I suggest focusing on "Promoting Digital Literacy for Rural Youth". Many young people in remote areas lack access to quality tech education, limiting their future opportunities.。
Your proposal could call for global partnerships to provide affordable devices, online courses, and teacher training in underserved regions. This aligns with the UN’s Sustainable Development Goals and empowers youth to bridge the digital divide. As students, you can share personal stories to highlight how tech skills transform lives.。
I’m sure this idea will resonate with the UN and showcase youth-driven solutions!。
Yours,
Li Hua。
韓憲昌教(jiāo)師評分:內(nèi)容6分(fèn),言語7分,結構4分 → 總分17分。
評語:內容根本(běn)完好,主張“鄉村數字素質”合理,但理由(yóu)稍簡(jiǎn)略(luè)(未(wèi)充沛打開影響)。言語精確,句式較簡略,表達得當。結構明晰,聯接天然。
趙文嘉教師評分:17分。
評語:因未運用連接詞,結構和邏輯比較稍顯差勁。
騰訊。混元。T1。
Dear Jim,
Glad to hear you’re working on the UN youth proposal. Here’s my idea: Focus on "Youth-Led Green Campus Initiatives".。
Teens are key to shaping a sustainable future, and schools are where we spend most of our time. Proposing eco-actions—like recycling competitions, energy-saving challenges, or plant-a-tree days—can turn small habits into big change. These activities are easy to organize, fun, and teach us responsibility. Plus, they show the UN how youth drive real-world solutions.。
Hope this helps. Can’t wait to hear your final plan—go for it, Jim!
Yours,
Li Hua。
韓憲昌教師評分:內容5分,言語5分,結構2分 → 總分12分。
評語:內容根本相關,但主張“綠色學校”理由(yóu)泛泛(缺少詳細支撐)。言語口語化(如“Glad to hear”),句式簡略,得當性弱。結構鬆懈,完畢隨意。
趙文嘉教師(shī)評分:14分。
評語:結構不行明晰,缺少連接詞,理由不豐厚,不行深化。
訊飛星火。 X1。
Dear Jim,
I’m glad to help with your proposal! I suggest proposing a "Global Youth Climate Action Network". This initiative would connect teenagers worldwide to share eco-friendly projects, like school recycling programs or community tree-planting, and collaborate on solutions to climate issues.。
The reason is twofold: First, it empowers young people to take practical steps against environmental problems while spreading awareness. Second, uniting diverse voices under one platform highlights youth’s role in shaping a sustainable future, which aligns perfectly with the UN’s goals.。
Let me know if you need further details!。
Yours,
Li Hua。
韓憲昌教師評分:內容8分,言語8分,結構4分(fèn) → 總分20分。
評語:內容完好,主張“全球氣候舉動網絡”明(míng)晰,理由充沛且緊扣主題。言語根本(běn)精確,句式多(duō)樣(如“twofold”結構)。結構(gòu)明晰,階段聯接天然。
趙文嘉(jiā)教師(shī)評分:19分。
評語:審題上標題了解精確,內容上中心關鍵都提到(dào)了且(qiě)做了較(jiào)為深化的解說。言語上用了較多方式多變的句(jù)法結構。結構上第一段與第二段篇(piān)幅適當,一般(bān)來講假如采納(nà)總分總的結構,能夠經過篇幅和內容來傑出中心中(zhōng)心階(jiē)段。
百度。文心。X1。
Dear Jim,
I’m glad to know you’re participating in the UN’s proposal collection. My suggestion is to focus on ""Promoting Traditional Craftsmanship Through Digital Innovation"".。
Firstly, many ancient handicrafts are fading due to globalization. By integrating digital tools like 3D modeling or VR, teenagers can help artisans preserve techniques and reach wider audiences. Secondly, this proposal aligns with sustainability—reviving local crafts reduces industrial waste and supports eco-friendly production. Lastly, it fosters cultural exchange, as young people worldwide could share stories behind these crafts, enhancing mutual understanding.。
I believe this idea balances creativity, cultural preservation, and tech engagement, which the UN values. Let me know if you need more details!。
Yours,
Li Hua。
韓憲昌教(jiāo)師評分:內容7分,言語8分,結構4分 → 總(zǒng)分19分。
評語:內(nèi)容立異,主張“數(shù)字工藝維護(hù)”詳(xiáng)實,理由充沛(文明可持(chí)續性)。言語優異,句式多(duō)樣(如“Firstly...Lastly”),無過錯。結構合理,過渡流通。
趙文嘉教師評分:18分。
評語:結構明晰幹練,理由(yóu)充沛詳細,選詞和句法精確多樣。
整理兩位教師給出的平均分,六款大模型的排名如下:。
關於上述AI大模型相互之間的比(bǐ)照,韓憲昌教師表明,它們都能在內容方麵切中關鍵,完結寫作使命。可是,內容細節多少、細節邏輯聯係強(qiáng)弱、觀念表達精準性與簡潔(jié)性等方麵,不同比(bǐ)較(jiào)大。
趙文嘉教師則表明(míng),有的模型的作文無論是語法,句法仍是選詞都十分(fèn)精準到位,且邏輯明晰,理(lǐ)由(yóu)充沛(pèi)深化,有的模型的作文(wén)比較(jiào)之下則邏輯性短缺,打開得不行詳細深化。
韓教師告知新京報貝殼財經記者(zhě),這些AI生成的作文立意都(dōu)比較立異,由於AI大模型在短時間內相關信息查找和整(zhěng)合才能強於考生。而實在考生在有限時間內,很難有特別立異的視點,一(yī)般會求穩而選用保存的視點,且是平(píng)常訓(xùn)練過的相關的論題(tí)內容遷移到考場上。
而在趙教師看來,上述AI大模型生成作文的特點是選詞都十分高檔和相對精確,能夠合理運用不(bú)同的詞匯(huì)來表達。部分語法如非謂語動(dòng)詞(cí),定語(yǔ)從句的使用十分嫻熟。內容比較深化詳細。而和實(shí)在考生的差(chà)異首要在於選詞以及內容(róng)表達的深化程度上。可是比較實在考生(shēng)中寫作(zuò)十分優異的同(tóng)學,AI生成的作文(wén)句式仍是不行豐(fēng)厚,用的(de)比較固定,如分詞做後(hòu)置定(dìng)語,狀語,以(yǐ)及定語從句。除此(cǐ)之外實在考生還會用(yòng)到各(gè)種類型的倒裝,動名詞的複合結構和獨(dú)立主格等。
“從詳細內容來(lái)講,AI大模(mó)型內嵌的評分標準和寫作方(fāng)向不同很大,有的作文內容過於簡略,有(yǒu)的過於冗長和雜亂,這(zhè)應該(gāi)跟大模型(xíng)內部的(de)數據庫有很大聯係。”韓教師說。
值得留意(yì)的是(shì),上述“AI考生”均為具有深度考慮(lǜ)功用(yòng)的(de)推理(lǐ)大模型。跟著本年春節期間DeepSeek R1的走紅,各家幹流大模型公司對推理大模型(xíng)的注重程度日積月(yuè)累。5月28日,DeepSeek更新了(le)DeepSeek-R1-0528版別;5月21日,騰訊旗下的深度考慮模型混元T1和快考慮(lǜ)模型混(hún)元Turbo S雙雙迭代;4月29日,阿裏巴巴開源(yuán)新一代(dài)通(tōng)義千問(wèn)模型Qwen3,其可(kě)一起支撐推理和非推理(lǐ)使命;4月25日,百(bǎi)度發布了深度考慮模型X1 Turbo;4月20日,科大訊飛對深度推理大模型訊(xùn)飛星火X1進(jìn)行了全新晉級。
能夠發現,在本年4至5月,國內幹流大模型均進行(háng)了推理層麵的更新,這也帶來了其在英語寫作(zuò)等方麵的才能(néng)前進,等待國產大模(mó)型的下一次(cì)進化。
文/羅亦丹(dān)。
修改 楊娟娟 校正 王心。
電影(yǐng)《私家(jiā)偵探》曝宣揚(yáng)主題曲《惋(wǎn)惜》MV 於文文傾(qīng)情獻唱刺(cì)破血色迷霧
小紅書發布虛偽內容管理陳述:封禁黑灰產賬(zhàng)號超1000萬個 — 新京報