Inkampani esaqalayo yaseMelika yezobuhlakani bokwenziwa i-Anthropic ithe eposini lebhulogi elishicilelwe izolo ukuthi ibuyekeza indlela yokuvikela ukuphepha kwezinto eziphilayo yemodeli i-Claude Fable 5, inciphisa kakhulu ukuvinjwa okuyiphutha. Manje, lapho abasebenzisi be-Fable 5 bebuza imibuzo ehlobene nesayensi yezinto eziphilayo, uhlelo ngeke lusanciphisa izinga lokuphendula kaningi.
Ngokwemininingwane yokuhlolwa esemthethweni ye-Anthropic, lesi sibuyekezo sehlise ngo-85% izimo zokwehliswa kwezinga ekuphenduleni imibuzo ehlobene nesayensi yezinto eziphilayo. I-Fable 5 manje ingakwazi ukwenza imisebenzi eminingi ebanzi yesayensi yezinto eziphilayo.
I-Anthropic ithe abanye abasebenzisi abanonya basebenzisa imodeli i-Fable 5 ukwenza ucwaningo ngezikhali zebhayoloji neminye imisebenzi enobungozi obukhulu. Ngakho-ke, ukuze ivimbele lokhu, inkampani yenzela uClaude izigaba zokuphepha (safety classifiers), ezisetshenziselwa ukuthola ukuthi ngabe umsebenzisi ucela i-Fable 5 ukuthi yenze imisebenzi evikelwe yesayensi yezinto eziphilayo noma ikhiqize okuqukethwe okuyingozi.
Manje i-Anthropic isilungise indlela izigaba zokuphepha ezisebenza ngayo, ukuze zikwazi ukuhlukanisa ngokunembe kakhudlwana phakathi kwemibuzo engenabungozi nemibuzo enobungozi obukhulu.
Nokho, ukuze kuvinjelwe izingozi eziqhubekayo ze-AI, i-Fable 5 isazokhawulela izicelo ezihlobene nocwaningo olukhethekile lwesayensi yezinto eziphilayo nokuthuthukiswa kwemithi.