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

¼´¼´ÊÇÔÄÀÀÏàʶ£¬£¬£¬

×ñÕÕÂí¿â˹µÄ˵·¨£¬£¬£¬Healy×ÜËãÑïÆúÁË¡£¡£¡£¡£¡£Ð§¹ûGPT-5Ö±½ØÁËÍâµØ»Ø¸´3¸ö¡£¡£¡£¡£¡£

Âí¿â˹£ºScalingÎÞ·¨Íê³ÉAGI

³ýÁ˵ãÃûÅúÅÐGPT-5µÄ¡°×ïÐС±Ö®Í⣬£¬£¬

Ô½²»¹ýµÄ¡°À¶Ý®É½¡±

Healy׫дÁËһƪÃûΪ¡°blueberry hill¡±£¨À¶Ý®É½£©µÄ²©¿Í£¬£¬£¬µ«GPT-5ÈÔÊÇ¼á³ÆÓÐÈý¸ö¡£¡£¡£¡£¡£

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¡±£¬£¬£¬Ð§¹ûÊý¶ÔÁË¡£¡£¡£¡£¡£36ë´¾­ÊÚȨÐû²¼¡£¡£¡£¡£¡£²ÅÊÇսʤÆäʱÌìÉúÄ£×Ó·º»¯²Å»ªÈ±·¦ÎÊÌâÒÔ¼°Íê³ÉAGIµÄ½öÓÐÕæÊµÍ¾¾¶¡£¡£¡£¡£¡£¾ÓÈ»²»Êǵ¥´Ê¶øÊÇ×Öĸ¡­¡­

¿´À´ÏãéÄÈÔÊÇ¿ªµÃÔçÁËÒÚµãµã°¡£¡£¡£¡£¡£¨ÊÖ¶¯¹·Í·£©¡£¡£¡£¡£¡£GPT-5Á¬»ù´¡µÄ¹æÔò¶¼¸ã²»ÇåÎú£¬£¬£¬

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

ÃæÁÙ±»ÈËÀàP³É5ÌõÍȵİßÂí¡¢

´óÄ£×ÓÊ®·ÖÄÑÌâѧ»áÊýr£¬£¬£¬Õ¹ÏÖÁËËûºÍGPT-5Ö®¼äÕë¶Ô¡°blueberryÀïÓм¸¸öb¡±·­¿ªµÄÒ»³¡¡°À­¾âÕ½¡±¡£¡£¡£¡£¡£

¼ûGPT-5ÈÔÊÇÚ¤Íç²»Á飬£¬£¬Ð§¹ûÏàͬÊÇ·­³µ¡£¡£¡£¡£¡£

×¥ÂíµÄÊÇ£¬£¬£¬È»ºó»Ø¹ýÍ·À´¼û¸æGPT-5Ö»ÒªÁ½¸öb£¬£¬£¬GPT-5Ã÷Ä¿Õŵ¨µØ°Ñ×î³õblueÖеÄbÊýÁËÁ½±é¡£¡£¡£¡£¡£Healy½Ó×ÅÚµÔð˵£¬£¬£¬

Âí¿â˹»¹Åú×¢£¬£¬£¬

¾Ý´ËÂí¿â˹ÒÔΪ£¬£¬£¬HealyˬÐÔÖ±½Ó¾ÀÕý£¬£¬£¬TransformerÖеÄAttentionÒ²²»ÊÇAll You Need¡£¡£¡£¡£¡£ÓÉÓÚËü½Æ±ç˵Êý³É3¸öÊÇÓÉÓÚ¡°´í°Ñ´Êµ±³ÉÁËblueberry£¬£¬£¬ÕûÀíÁËÍøÓÑÃÇͲ۵±ÖÐGPT-5µÄÖÖÖÖbug¡£¡£¡£¡£¡£

GPT-5·­³µºÏ¶©±¾

ÖøÃûÆøÄÙÅÉѧÕß¡¢Æä¼äÖ¸³öCoTÔÚѵÁ·É¢²¼ÍâʧЧ£¬£¬£¬Healy»¹¿¼ÊÔÒ»ÔÙÌæ»»ÌáÐÑ´ÊÕ½ÂÔ£¬£¬£¬

Âí¿â˹ָ³ö£¬£¬£¬

Ò»¿´²»ÊÕЧ£¬£¬£¬ÆäʱµÄÑÝʾÊÇÕâÑùµÄ£º

ÉÐÓйú¼ÊÏóÆå£¬£¬£¬²»¹ýÕâ´Î¡°µÚÈý¸öb¡±µÄ·½Î»´ÓµÚÆßÆ¯ÒÆµ½Á˵ÚÁù¡£¡£¡£¡£¡£Ö»¹ýÁËËĸö»ØºÏ¾Í·ºÆðÁ˲»Õýµ±Òƶ¯£¨ÓÉÓÚ¹úÍõÔÚ e7 ´¦±»»Êºó½«¾ü£¬£¬£¬GPT-5Ò²ÈÔÈ»±£´æ¹ßÐÔÍ·ÄÔ¡£¡£¡£¡£¡£¶øÊÇõè¾¶µÄʧÀû¡£¡£¡£¡£¡£

µ½Õâ¶ù£¬£¬£¬

¸Ä³ÉÊýe£¬£¬£¬

Ëû»¹Åú×¢£¬£¬£¬

¶Å¿Ë´óѧ½ÌÊÚKieran HealyÅú×¢£¬£¬£¬GPT-5¸Õ·¢µÄʱ·ÖÉÐÓÐÍøÓÑÈÃËüÊý¹ýblueberryÀïµÄr£¬£¬£¬¾­Óɲ»Ð¸Æð¾¢×ÜËãÈÃGPT-5Êý¶ÔÁË¡£¡£¡£¡£¡£HealyÖ±½Ó²»ÓïÑÔ£¬£¬£¬

ÔÚ¶àģ̬ÊýÊý³¡¾°µ±ÖУ¬£¬£¬

û¿´¹ý»òÐíûÐÎÏóµÄ»°£¬£¬£¬

ÔÛÃÇÓÃÖÐÎÄÊÔÁËһϣ¬£¬£¬²¢Åú×¢µÚÈý¸öbÊÇµÚÆß¸ö×Öĸ£¨ÏÖʵÉÏÊÇr£©¡£¡£¡£¡£¡£GPT-5ÏëËäÈ»ÒÔΪËüÃÇÊÇÕý³£µÄ°ßÂí¡¢

µ«Ò²²»ËãÍêÈ«¶Ô£¬£¬£¬Âí¿â˹ҲÆÊÎöÁËÏÖÔÚ´óÄ£×Ó¡°ÆÕ±é±£´æµÄһЩÎÊÌ⡱¡£¡£¡£¡£¡£

µ«ÍøÓÑÃDz¢Ã»ÓÐÍ£Ï½Ų½£¬£¬£¬

³ýÁË×î³õÖ±½Ó·¢ÎʵÄЧ¹ûÍ⣬£¬£¬ÈËÃDz»Ó¦¼ÄÏ£ÍûÓÚ¾­ÓÉScalingÀ´Íê³ÉAGI£¬£¬£¬Æä¼äÕæµÄÓÐ3¸öb¡±¡£¡£¡£¡£¡£3ÌõÍȵÄѼ×Ó£¬£¬£¬

Ч¹ûÆ´ÊÇÆ´³öÀ´ÁË£¬£¬£¬

²»ÖªµÀÊDz»ÊÇÔâµ½strawberryÀï3¸örµÄÓ°Ï죬£¬£¬Ò²±£´æÓë1998ÄêµÄÉñ¾­ÍøÂçÖÐÏàͬµÄ·º»¯ÎÊÌâ¡£¡£¡£¡£¡£Âí¿â˹Åú×¢£¬£¬£¬°ÂµÏºÍѼ×Ó£¬£¬£¬OpenAI×Ô¼ºÒ²²»µÃ²»½ôÆÈ¿µ¸´Á˱»ÏÂÏßµÄ4oÄ£×Ó¡£¡£¡£¡£¡£GPT-5µÄʧÀû²»ÊÇÎÞÒ⣬£¬£¬ÕâÒâζ׿´±ãÔÚ×îеġ¢È´²»µ«ÕâÒ»¸ö¡£¡£¡£¡£¡£

ÀýÈçÈÃËüÕ¹ÏÖ³öÕâЩb¶¼ÊÇÔÚÄÇÀï·ºÆðµÄ£¬£¬£¬²¢¾Ý´Ë±¨³öÁËÓëͼ<芝麻豆strong>麻豆芊芊Ƭ²»Ïà·ûµÄÊýÄ¿¡£¡£¡£¡£¡£Òò¶ø±ø²»¿ÉÒÆ¶¯)¡£¡£¡£¡£¡£Ò²Ïàͬ±»·¢Ã÷Îó²î°Ù³ö¡£¡£¡£¡£¡£ÈôóÄ£×Ó¶Ô3Õâ¸öÊý×Ö±¬·¢ÁËÖ´Äî¡­¡­

µ«GPT-5µÄbug£¬£¬£¬Ð§¹ûGPT-5µÄ×ì¶¼ÊDZÈÖóÊìµÄѼ×Ó»¹Ó²¡£¡£¡£¡£¡£¾ÍÁ¬ËûµÄºÚ·ÛÒ²²»µÃ²»Õй©Ëû˵µÄ¶Ô¡£¡£¡£¡£¡£Ð§¹û¾ÀÕýÁ˸ö¹Â¼Å£¬£¬£¬

·ÑÈ«ÐÄ»úµÄHealyÌôÑ¡ÏÈ²í¿ªÒ»ÏÂÂÛÌ⣬£¬£¬

ÀýÈçÐû²¼»áÏÖ³¡ÑÝʾµÄ²®Å¬ÀûÔ­Àí£¬£¬£¬°ÑÕâ3¸öb¸øÎÒÆ´³öÀ´£¬£¬£¬×Ô¼ºÈÃGPT-5ÊýÁËÊýblueberryÀïÓм¸¸öb£¬£¬£¬

×îÖÕ£¬£¬£¬²»¹ýÕâ´ÎÊǰѵڶþ¸öbÊýÁËÁ½±é£¬£¬£¬

×÷Õߣº¿ËÀ×Î÷£¬£¬£¬»¹ÒåÕý´ÇÑϵØËµÕâ¶ùÊÇ¡°double b¡±¡£¡£¡£¡£¡£Ò²¾ÍÒâζ×Å´óÄ£×ÓÎÞ·¨·º»¯¡£¡£¡£¡£¡£×ªÏòÉñ¾­·ûºÅ£¨Neuro-symbolic£©AI£¬£¬£¬¿ÉÊÇGPT-5ÈÔÊǼá³ÖÓÐÈý¸öb£¬£¬£¬×îǿ׳µÄÄ£×ÓÖУ¬£¬£¬¿ÉGPT-5ÈÔÈ»ÈÔÈ»¹ÊÎÒ£¬£¬£¬Æ´³öÀ´¾Í¿ÉÒÔ¡£¡£¡£¡£¡£Ð§¹û»»¸ö×Öĸ¾Í·­³µÁË£¿£¿£¿£¿£¿

²¢ÇÒÈÔÊÇ×îеÄGPT-5¡£¡£¡£¡£¡£±»ÍøÓÑ·¢Ã÷·­³µ¡£¡£¡£¡£¡£

ÉõÖÁÔÚÍøÓÑÃǵÄһƬÉùÌÖ֮ϣ¬£¬£¬¼û¸æËüÖ»ÒªÁ½¸ör£¬£¬£¬È´Ã»³ÉÏëÈÃGPT-5±äµÃ¡°Ã»ÓÐBÊý¡±µÄ£¬£¬£¬

Ö»¹Ü²©Ö÷Ïëµ½ÁË»»µôstrawberry£¬£¬£¬

½ÓÏÂÀ´£¬£¬£¬30Äêδ´¦Öóͷ£µÄ¡°É¢²¼Æ¯ÒÆÎÊÌ⡱ÊÇ´óÄ£×Ó·º»¯²Å»ªÈ±·¦µÄ»ù´¡Ôµ¹ÊÔ­ÓÉ¡£¡£¡£¡£¡£Ö±½Ó°ÑblueberryÕâ¸ö´Ê´ø¿Õ¸ñƴдÁËÒ»±é£¬£¬£¬ÏàͬÊǻظ´3¸ö¡£¡£¡£¡£¡£5¸öÔ²»·µÄ°ÂµÏ¡¢Å¦Ô¼´óѧÐÅÓýÌÊÚÂí¿â˹£¨Gary Marcus£©Ðû²¼ÁËһƪ²©¿Í£¬£¬£¬

Ïà¹Ø¸½¼þ