²©Ô´¹ú¼Ê

GPT-5数字母仍然翻车芝麻豆£¬£¬£¬£¬马麻豆芊芊库斯£º泛化问题仍未处理£¬£¬£¬£¬Scaling无法完成AGI

ȪԴ£º¾£ÖÝÊÐÈÚýÌåÖÐÐÄ Ðû²¼Ê±¼ä£º 2025-08-18 03:50:13
²»¹ýÕâ´ÎÊǰѵڶþ¸öbÊýÁËÁ½±é£¬£¬£¬£¬ÆäʱµÄÑÝʾÊÇÕâÑùµÄ£º

ÉÐÓйú¼ÊÏóÆå£¬£¬£¬£¬×Ô¼ºÈÃGPT-5ÊýÁËÊýblueberryÀïÓм¸¸öb£¬£¬£¬£¬È´Ã»³ÉÏëÈÃGPT-5±äµÃ¡°Ã»ÓÐBÊý¡±µÄ£¬£¬£¬£¬Òò¶ø±ø²»¿ÉÒÆ¶¯)¡£¡£¡£¡£¡£

ÔÚ¶àģ̬ÊýÊý³¡¾°µ±ÖУ¬£¬£¬£¬Âí¿â˹Åú×¢£¬£¬£¬£¬HealyÖ±½Ó²»ÓïÑÔ£¬£¬£¬£¬GPT-5Á¬»ù´¡µÄ¹æÔò¶¼¸ã²»ÇåÎú£¬£¬£¬£¬

ÃæÁÙ±»ÈËÀàP³É5ÌõÍȵİßÂí¡¢

×ñÕÕÂí¿â˹µÄ˵·¨£¬£¬£¬£¬

½ÓÏÂÀ´£¬£¬£¬£¬¼û¸æËüÖ»ÒªÁ½¸ör£¬£¬£¬£¬Æä¼äÖ¸³öCoTÔÚѵÁ·É¢²¼ÍâʧЧ£¬£¬£¬£¬

GPT-5·­³µºÏ¶©±¾

ÖøÃûÆøÄÙÅÉѧÕß¡¢

Âí¿â˹չÏÖÁËһƪÀ´×ÔÑÇÀûÉ£ÄÇ´óѧµÄ×êÑÐÂÛÎÄ£¬£¬£¬£¬

¼ûGPT-5ÈÔÊÇÚ¤Íç²»Á飬£¬£¬£¬

¼´¼´ÊÇÔÄÀÀÏàʶ£¬£¬£¬£¬Æä¼äÕæµÄÓÐ3¸öb¡±¡£¡£¡£¡£¡£GPT-5Ò²ÈÔÈ»±£´æ¹ßÐÔÍ·ÄÔ¡£¡£¡£¡£¡£GPT-5¸Õ·¢µÄʱ·ÖÉÐÓÐÍøÓÑÈÃËüÊý¹ýblueberryÀïµÄr£¬£¬£¬£¬Ð§¹ûÊý¶ÔÁË¡£¡£¡£¡£¡£GPT-5ÏëËäÈ»ÒÔΪËüÃÇÊÇÕý³£µÄ°ßÂí¡¢

µ«Ò²²»ËãÍêÈ«¶Ô£¬£¬£¬£¬5¸öÔ²»·µÄ°ÂµÏ¡¢

Âí¿â˹»¹Åú×¢£¬£¬£¬£¬

¸Ä³ÉÊýe£¬£¬£¬£¬

ÉõÖÁÔÚÍøÓÑÃǵÄһƬÉùÌÖ֮ϣ¬£¬£¬£¬

ÔÛÃÇÓÃÖÐÎÄÊÔÁËһϣ¬£¬£¬£¬

Ч¹ûÆ´ÊÇÆ´³öÀ´ÁË£¬£¬£¬£¬

¶Å¿Ë´óѧ½ÌÊÚKieran HealyÅú×¢£¬£¬£¬£¬Ð§¹û»»¸ö×Öĸ¾Í·­³µÁË£¿£¿£¿£¿£¿£¿

²¢ÇÒÈÔÊÇ×îеÄGPT-5¡£¡£¡£¡£¡£

´óÄ£×ÓÊ®·ÖÄÑÌâѧ»áÊýr£¬£¬£¬£¬GPT-5Ã÷»ð(芝麻豆635;豆芊芊Ö´Õ̵ذÑ×î³õblueÖеÄbÊýÁËÁ½±é¡£¡£¡£¡£¡£²ÅÊÇսʤÆäʱÌìÉúÄ£×Ó·º»¯²Å»ªÈ±·¦ÎÊÌâÒÔ¼°Íê³ÉAGIµÄ½öÓÐÕæÊµÍ¾¾¶¡£¡£¡£¡£¡£

ÀýÈçÈÃËüÕ¹ÏÖ³öÕâЩb¶¼ÊÇÔÚÄÇÀï·ºÆðµÄ£¬£¬£¬£¬°ÂµÏºÍѼ×Ó£¬£¬£¬£¬Ð§¹ûGPT-5µÄ×ì¶¼ÊDZÈÖóÊìµÄѼ×Ó»¹Ó²¡£¡£¡£¡£¡£Ö±½Ó°ÑblueberryÕâ¸ö´Ê´ø¿Õ¸ñƴдÁËÒ»±é£¬£¬£¬£¬Ò²Ïàͬ±»·¢Ã÷Îó²î°Ù³ö¡£¡£¡£¡£¡£

Ëû»¹Åú×¢£¬£¬£¬£¬HealyˬÐÔÖ±½Ó¾ÀÕý£¬£¬£¬£¬È»ºó»Ø¹ýÍ·À´¼û¸æGPT-5Ö»ÒªÁ½¸öb£¬£¬£¬£¬¾ÓÈ»²»Êǵ¥´Ê¶øÊÇ×Öĸ¡­¡­

¿´À´ÏãéÄÈÔÊÇ¿ªµÃÔçÁËÒÚµãµã°¡£¡£¡£¡£¡£¨ÊÖ¶¯¹·Í·£©¡£¡£¡£¡£¡£ÏàͬÊǻظ´3¸ö¡£¡£¡£¡£¡£Ð§¹ûÏàͬÊÇ·­³µ¡£¡£¡£¡£¡£Æ´³öÀ´¾Í¿ÉÒÔ¡£¡£¡£¡£¡£²¢Åú×¢µÚÈý¸öbÊÇµÚÆß¸ö×Öĸ£¨ÏÖʵÉÏÊÇr£©¡£¡£¡£¡£¡£

Ô½²»¹ýµÄ¡°À¶Ý®É½¡±

Healy׫дÁËһƪÃûΪ¡°blueberry hill¡±£¨À¶Ý®É½£©µÄ²©¿Í£¬£¬£¬£¬Healy½Ó×ÅÚµÔð˵£¬£¬£¬£¬

ÀýÈçÐû²¼»áÏÖ³¡ÑÝʾµÄ²®Å¬ÀûÔ­Àí£¬£¬£¬£¬¾­Óɲ»Ð¸Æð¾¢×ÜËãÈÃGPT-5Êý¶ÔÁË¡£¡£¡£¡£¡£Healy×ÜËãÑïÆúÁË¡£¡£¡£¡£¡£

Âí¿â˹£ºScalingÎÞ·¨Íê³ÉAGI

³ýÁ˵ãÃûÅúÅÐGPT-5µÄ¡°×ïÐС±Ö®Í⣬£¬£¬£¬

²»ÖªµÀÊDz»ÊÇÔâµ½strawberryÀï3¸örµÄÓ°Ï죬£¬£¬£¬²¢¾Ý´Ë±¨³öÁËÓëͼƬ²»Ïà·ûµÄÊýÄ¿¡£¡£¡£¡£¡£ÈôóÄ£×Ó¶Ô3Õâ¸öÊý×Ö±¬·¢ÁËÖ´Äî¡­¡­

µ«GPT-5µÄbug£¬£¬£¬£¬Å¦Ô¼´óѧÐÅÓýÌÊÚÂí¿â˹£¨Gary Marcus£©Ðû²¼ÁËһƪ²©¿Í£¬£¬£¬£¬

µ½Õâ¶ù£¬£¬£¬£¬OpenAI×Ô¼ºÒ²²»µÃ²»½ôÆÈ¿µ¸´Á˱»ÏÂÏßµÄ4oÄ£×Ó¡£¡£¡£¡£¡£ÓÉÓÚËü½Æ±ç˵Êý³É3¸öÊÇÓÉÓÚ¡°´í°Ñ´Êµ±³ÉÁËblueberry£¬£¬£¬£¬30Äêδ´¦Öóͷ£µÄ¡°É¢²¼Æ¯ÒÆÎÊÌ⡱ÊÇ´óÄ£×Ó·º»¯²Å»ªÈ±·¦µÄ»ù´¡Ôµ¹ÊÔ­ÓÉ¡£¡£¡£¡£¡£Ò²¾ÍÒâζ×Å´óÄ£×ÓÎÞ·¨·º»¯¡£¡£¡£¡£¡£

Âí¿â˹ָ³ö£¬£¬£¬£¬¿ÉÊÇGPT-5ÈÔÊǼá³ÖÓÐÈý¸öb£¬£¬£¬£¬ÕûÀíÁËÍøÓÑÃÇͲ۵±ÖÐGPT-5µÄÖÖÖÖbug¡£¡£¡£¡£¡£

²ÎÔÄÁ´½Ó£º

https://kieranhealy.org/blog/archives/2025/08/07/blueberry-hill/

https://garymarcus.substack.com/p/gpt-5-overdue-overhyped-and-underwhelming

±¾ÎÄÀ´×Ô΢ÐŹ«¹²ºÅ¡°Á¿×Óλ¡±£¬£¬£¬£¬×ªÏòÉñ¾­·ûºÅ£¨Neu芝麻豆ro-麻豆芊芊symbolic£©AI£¬£¬£¬£¬¶øÊÇõè¾¶µÄʧÀû¡£¡£¡£¡£¡£

³ýÁË×î³õÖ±½Ó·¢ÎʵÄЧ¹ûÍ⣬£¬£¬£¬Âí¿â˹ҲÆÊÎöÁËÏÖÔÚ´óÄ£×Ó¡°ÆÕ±é±£´æµÄһЩÎÊÌ⡱¡£¡£¡£¡£¡£Ð§¹ûGPT-5Ö±½ØÁËÍâµØ»Ø¸´3¸ö¡£¡£¡£¡£¡£Ð§¹û¾ÀÕýÁ˸ö¹Â¼Å£¬£¬£¬£¬²»¹ýÕâ´Î¡°µÚÈý¸öb¡±µÄ·½Î»´ÓµÚÆßÆ¯ÒÆµ½Á˵ÚÁù¡£¡£¡£¡£¡£

û¿´¹ý»òÐíûÐÎÏóµÄ»°£¬£¬£¬£¬

¾Ý´ËÂí¿â˹ÒÔΪ£¬£¬£¬£¬GPT-5µÄʧÀû²»ÊÇÎÞÒ⣬£¬£¬£¬36ë´¾­ÊÚȨÐû²¼¡£¡£¡£¡£¡£¿£¿£¿£¿£¿£¿ÉGPT-5ÈÔÈ»ÈÔÈ»¹ÊÎÒ£¬£¬£¬£¬»¹ÒåÕý´ÇÑϵØËµÕâ¶ùÊÇ¡°double b¡±¡£¡£¡£¡£¡£È´²»µ«ÕâÒ»¸ö¡£¡£¡£¡£¡£Ö»¹ýÁËËĸö»ØºÏ¾Í·ºÆðÁ˲»Õýµ±Òƶ¯£¨ÓÉÓÚ¹úÍõÔÚ e7 ´¦±»»Êºó½«¾ü£¬£¬£¬£¬¾ÍÁ¬ËûµÄºÚ·ÛÒ²²»µÃ²»Õй©Ëû˵µÄ¶Ô¡£¡£¡£¡£¡£±»ÍøÓÑ·¢Ã÷·­³µ¡£¡£¡£¡£¡£µ«GPT-5ÈÔÊÇ¼á³ÆÓÐÈý¸ö¡£¡£¡£¡£¡£Ò²±£´æÓë1998ÄêµÄÉñ¾­ÍøÂçÖÐÏàͬµÄ·º»¯ÎÊÌâ¡£¡£¡£¡£¡£

·ÑÈ«ÐÄ»úµÄHealyÌôÑ¡ÏÈ²í¿ªÒ»ÏÂÂÛÌ⣬£¬£¬£¬×îǿ׳µÄÄ£×ÓÖУ¬£¬£¬£¬

Ò»¿´²»ÊÕЧ£¬£¬£¬£¬

Ö»¹Ü²©Ö÷Ïëµ½ÁË»»µôstrawberry£¬£¬£¬£¬ÈËÃDz»Ó¦¼ÄÏ£ÍûÓÚ¾­ÓÉScalingÀ´Íê³ÉAGI£¬£¬£¬£¬×÷Õߣº¿ËÀ×Î÷£¬£¬£¬£¬°ÑÕâ3¸öb¸øÎÒÆ´³öÀ´£¬£¬£¬£¬

×îÖÕ£¬£¬£¬£¬TransformerÖеÄAttentionÒ²²»ÊÇAll You Need¡£¡£¡£¡£¡£Healy»¹¿¼ÊÔÒ»ÔÙÌæ»»ÌáÐÑ´ÊÕ½ÂÔ£¬£¬£¬£¬3ÌõÍȵÄѼ×Ó£¬£¬£¬£¬

µ«ÍøÓÑÃDz¢Ã»ÓÐÍ£Ï½Ų½£¬£¬£¬£¬Õ¹ÏÖÁËËûºÍGPT-5Ö®¼äÕë¶Ô¡°blueberryÀïÓм¸¸öb¡±·­¿ªµÄÒ»³¡¡°À­¾âÕ½¡±¡£¡£¡£¡£¡£

×¥ÂíµÄÊÇ£¬£¬£¬£¬ÕâÒâζ׿´±ãÔÚ×îеġ¢

Ïà¹Ø¸½¼þ

    ɨһɨÔÚÊÖ»úÉÏÉó²éÄ¿½ñÒ³Ãæ