home All News open_in_new Full Article

Chatbotovi su vrlo uspješni u uvjeravanju, ali sve više haluciniraju

Upravo je objavljeno izvješće tvrtke OpenAI o testiranju njihovih velikih jezičnih modela. Pokazalo je da su modeli o3 i o4-mini, objavljeni u travnju, imali znatno veće stope halucinacija od prethodnog modela o1, koji je izašao krajem 2024. godine.


today 3 w. ago attach_file Other

attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other
attach_file Other


ID: 1899042766
Add Watch Country

arrow_drop_down