2026届高三英语培优 外刊阅读学案 学会推理的AI更能骗人了？.docxVIP

高三英语培优外刊阅读

班级：____________学号：____________姓名：____________

外刊精选｜OpenAI重磅发布新模型：学会推理的AI更能骗人了？

12月5日至20日，人工智能巨头OpenAI完成了连续12个工作日的12场直播，活动首日重磅推出了推理大模型o1的正式版本，最后一天又以升级版的最强推理模型o3收官，首尾呼应，无疑凸显了推理模型的重要性。o1是第一个具备真正通用推理能力的大模型，模型会在给人答复之前停下来思考，在推理时进行更多的计算。AI究竟是如何思考的？为什么有时它会欺骗和糊弄人类呢？

HowDoesA.I.Think?HeresOneTheory.

ByPeterCoy

IntheroughlytwoyearssincethepublicreleaseofChatGPT,artificialintelligencehasadvancedfarmorerapidlythanhumanityhaslearnedtouseitsgoodfeaturesandtosuppressitsbadones.Onthebadside,forexample,itturnsoutthatA.I.isreallygoodatmanipulatinganddeceivingitshumanmasters.

ComputerscientistsarecontinuallysurprisedbythecreativitydisplayedbynewgenerationsofA.I.Considerthatlyingisasignofintellectualdevelopment:Childrenlearntoliearoundage3,andtheygetbetteratitastheydevelop.Asaliar,artificialintelligenceiswaypastthetoddlerstage.

Onehypothesisforhowlargelanguagemodelssuchaso1thinkisthattheyusewhatlogicianscallabduction,orabductivereasoning.Abductionisntwellknown,butitscommonindailylife.Itsinferringthemostlikelyexplanationforagivenobservation.Unlikededuction,whichisastraightforwardprocedure,andinduction,whichcanbepurelystatistical,abductionrequirescreativity.

TheplanetNeptunewasdiscoveredthroughabductivereasoning,whentwoastronomersindependentlyhypothesizedthatitsexistencewasthemostlikelyexplanationforperturbationsintheorbitofitsinnerneighbor,Uranus.

Largelanguagemodelsgeneratesentencesonewordatatimebasedontheirestimatesofprobability.Itsonethingtohonorthemysteryofthehumanbrain,butquiteanothertoadmitthatartificialintelligence—bothcreativeandconniving—isslippingawayfromourunderstanding.

【词汇过关】

请写出下面文单词在文章中的中文意思。

1.advancev.___________________

2.suppressv.___________________

3.manipulate英[m??n?pj?le?t]美[m??n?pj?le?t]v._______________