“论文记录了一个引人注目的案例：在处理一个涉及复杂数学表达式 √a - √(a + x) = x 的问题时，模型突然停下来说"Wait, wait. Wait. That's an aha moment I can flag here"（等等、等等、这是个值得标记的啊哈时刻），随后重新审视了整个解题过程。这种类似人类顿悟的行为完全是自发产生的，而不是预先设定的。”『一文读懂｜DeepSeek新模型大揭秘，为何它能震动全球AI圈』https://mp.weixin.qq.com/s?chksm=a9ec60ad9e9be9bb0a6488aa29e10739c34c1aa77c1ecb884422ad040db554d608b5d6270c07&exptype=unsubscribed_card_recommend_article_u2i_mainprocess_coarse_sort_tlfeeds&ranksessionid=1737859539_1&mid=2691555190&sn=ef787bf56ffddebb9e1a3203cd99f705&idx=1&__biz=Mjc1NjM3MjY2MA%3D%3D&scene=169&subscene=200&sessionid=1737859538&flutter_pos=10&clicktime=1737860313&enterid=1737860313&finder_biz_enter_id=5&ascene=56&fasttmpl_type=0&fasttmpl_fullversion=7572126-zh_CN-zip&fasttmpl_flag=0&realreporttime=1737860313151&devicetype=android-29&version=28003850&nettype=WIFI&abtest_cookie=AAACAA%3D%3D&lang=zh_CN&countrycode=CN&exportkey=n_ChQIAhIQeE6LUijRK7ahH%2FN6Dv9yfhLoAQIE97dBBAEAAAAAAH1qCAmiP3YAAAAOpnltbLcz9gKNyK89dVj02lcl%2FEj4M6ihwJ%2FL2mEaHwdm3LXegnRbgEGleVcwcgv8yRCfiwfVd95lsUsXtR5CPhDT%2FLVlYKHYEcrpfL2P%2FnjTz9BYYTx0KLbFM6OZB3XSneyyctl5vCgo6gJh4ecTXwVu7%2FwpanIHHS28zm9ee9V6b3suSmxL0qUjKs7wzgmYZbE3J%2FCdBpvIq9eHmcUjqoWMtS%2Fto78vfdxWLPHmITMsVevNwnDUQji%2B%2FmX7R24Ox2KR2BxEXJUQ%2B8%2BZZ1OQRII%3D&pass_ticket=bfsNmpSVv4XEo3bg2SBKCK1L1BWZ26sCR4fb9DrhRBrU1Sv1aC4sK0MnwWjp4AIS&wx_header=3