å¦ç¿’データã«æœ€é©åŒ–ã•ã‚Œã™ãŽã¦æœ¬æ¥ã®ç›®çš„ãŒé”æˆã§ããªããªã‚‹ã€ŒéŽå¦ç¿’ã€ã¨åŒæ§˜ã®ç¾è±¡ã¯AIã ã‘ã§ãªã社会全体ã§èµ·ã“ã£ã¦ã„ã‚‹ã¨ã„ã†ä¸»å¼µæ©Ÿæ¢°å¦ç¿’ã«ãŠã‘ã‚‹éŽå¦ç¿’(éŽå‰°é©åˆï¼ã‚ªãƒ¼ãƒãƒ¼ãƒ•ã‚£ãƒƒãƒ†ã‚£ãƒ³ã‚°)ã¨ã¯ã€AIãŒå¦ç¿’データã®ã¿ã«æœ€é©åŒ–ã•ã‚Œã¦ã—ã¾ã„ã€æœªçŸ¥ã®ãƒ‡ãƒ¼ã‚¿ã«å¯¾ã™ã‚‹äºˆæ¸¬èƒ½åŠ›ãŒä½Žããªã£ã¦ã—ã¾ã†ã¨ã„ã†ç¾è±¡ã§ã™ã€‚ãã‚“ãªéŽå¦ç¿’ã¨åŒæ§˜ã®ç¾è±¡ãŒæ©Ÿæ¢°å¦ç¿’分野ã ã‘ã§ãªã社会全体ã®ã•ã¾ã–ã¾ãªå ´é¢ã§ã‚‚発生ã—ã¦ã„ã‚‹ã¨ã€Google傘下ã®äººå·¥çŸ¥èƒ½ç ”究所・Google Brainã®ç ”究者ã§ã‚ã‚Šè¿‘å¹´ã®ç”»åƒç”ŸæˆAIã«åºƒã用ã„られã¦ã„る「拡散モデルã€ã®è«–文執ç†è€…ã§ã‚‚ã‚ã‚‹Jascha Sohl-Dicksteinæ°ãŒä¸»å¼µã—ã¦ã„ã¾ã™ã€‚ Too much efficiency makes everything worse: overfitting and the strong version of Goodhart’s law | Jascha’s blog https://sohl-dickstein.github.io/2022/11/06/strong-Go
{{#tags}}- {{label}}
{{/tags}}