【时文阅读】DeepSeek（阅读理解+语法填空+书面表达）.docx

【时文阅读】3.DeepSeek（阅读理解+语法填空+书面表达）

ChineseAIModelDeepSeek-R1StunstheWorld

January20,2025,aChinesetechcompanynamedDeepSeekmadeanewAIcalledDeepSeek-R1.Thissmartprogramcansolvemathproblems,writecode,andanswerquestionsliketopmodelssuchasOpenAI’sGPT-4o,butitcostsmuchlesstobuild.Theteamspentonly$5.58million(1/70ofOpenAI’scost)andused2,048computerchipsfortwomonths.Thisshowsthatgoodideascanbeatbigmoney.

DeepSeek-R1learnsbytryingmanytimesandgettingbetter,likehowstudentspracticemaths.OtherAIsneedlotsofpreparedexamples,butR1improvesbyitself.Forexample,inaglobalmathtest,itgot79.8%right,slightlyhigherthanOpenAI’smodel.AsimpleversionofR1canevenrunonlow-costcomputers,helpingschoolsandsmallcompanies.

DeepSeek-R1isfreetouseandchange.Onitsfirstdayonline,over19,000programmersworldwidestartedprojectstostudyit.AGermanscientistsaidexperimentswithR1cost$10insteadof$300withotherAIs.EvenMeta(afamousU.S.techcompany)topscientistsaid,“ThiskindoffreeAIcouldchangetheworld.”

U.S.techcompaniesareworried.Metaplanstospend$65billionmoreonAIresearch.ExpertssayChinaiscatchingupfast.Areportin《Nature》magazinecalledR1a“bigstepforward”,provingthatsmartideasbeatexpensivetools.

AsaDeepSeekengineersaid,“WewantAItobeafriend,notsomethingonlyrichpeoplecanuse.”WithR1,Chinashowstheworldthatgreatthingscomefromnewideasandhardwork,notjustmoney.

中国AI模型DeepSeek-R1震惊世界

2025年1?20?，中国科技公司深度求索（DeepSeek）研发出名为DeepSeek-R1的新型??智能。这款智能程序能解决数学题、编写代码、回答问题，性能媲美OpenAI的GPT-4o等顶尖模型，但开发成本低得多。团队仅花费558万美元（约为OpenAI成本的1/70），使?2048块计算机芯?耗时两?完成训练。这证明：好创意可以战胜巨额资

?。

DeepSeek-R1通过“反复尝试、逐步改进”的?式学习，就像学?练习数学题?样。其他AI需要?量标注范例，但R1能?主提升。例如，在?项全球数学测试中，它以

79.8%的正确率略胜OpenAI模型。R1的简化版本甚?能在低成本计算机上运?，助?学校和?型企业。

DeepSeek-R1免费开放使?与修改。上线??，全球超过1.9万名程序员启动相关研究项?。?位德国科学家表?，?R1做实验仅需10美元，?其他AI要花费300美元。

连美