ä¸å½ã®AIã¹ã¿ã¼ãã¢ãããDeepSeekãã¯2025å¹´1æ20æ¥ãæ°å¦çæ¨è«ãã³ã¼ãã£ã³ã°è½åã§OpenAIã®ææ°ã¢ãã«ão1ãã¨åçæ§è½ãæã¤å¤§è¦æ¨¡è¨èªã¢ãã«ãDeepSeek-R1ããå ¬éããã 使ç¨ã»è¤è£½ã»æ¹å¤ã»åé å¸ãèªç±ã«è¨±å¯ããå¯å®¹ãªMITã©ã¤ã»ã³ã¹ä¸ã§ã®ãªã¼ãã³ã½ã¼ã¹æä¾ããã³å¾æ¥æ¯95ã97%ã®ä½ã³ã¹ãAPIä¾¡æ ¼ãç¹å¾´ã§ãAIæ¥çã«æ¿éãèµ·ãã¦ããã å¼·åå¦ç¿ãéè¦ãã³ã¼ã«ãã¹ã¿ã¼ãåé¡ã«ãå¯¾å¿ è«æã«ããã¨ãDeepSeek-R1ã®ç¹çãã¹ãç¹ã¯ãå¼·åå¦ç¿ï¼RLï¼Reinforcement Learningï¼ãé§ä½¿ããå¾æ¥ã®æ師ããå¦ç¿ï¼SFTï¼Supervised Fine-Tuningï¼ã«é ¼ãããèªå¾çã«æèé£éï¼CoTï¼Chain-of-Thoughtï¼ãå¦ç¿ããç¹ã ããã®ã¢ããã¼ãã«ãããã¢ãã«ã¯è¤éãªåé¡ã解決ããããã®æèã®é£éãæ¢ç´¢ããèªå·±æ¤è¨¼ãæ¯ãè¿ãã
{{#tags}}- {{label}}
{{/tags}}