Iqembu eliholwa yi-Harvard University ne-Massachusetts Institute of Technology, elibandakanya i-OpenAI, i-Google DeepMind nezinye izinhlangano, lethule uhlelo lwe-MatrAIx kule nyanga.
Lolu hlelo lungakha ama-agent e-AI ayizigidigidi ezingu-8.3 ukuze lilingise ukuziphatha kwabantu emhlabeni wonke.
Le phrojekthi isebenzisa izilinganiso ezingu-1,290 ukwenza amamodeli abantu abanezizinda nezindlela zokuphila ezihlukahlukene. Lawa ma-agent e-AI angagcwalisa imibuzo njengabantu bangempela, axoxe nabasekeli bamakhasimende, aphequlule amawebhusayithi futhi asebenzise ama-app, kanti ukuvumelana kwemiphumela kufinyelele ku-91.5%.
Ithimba laqinisekisa lokhu ngocwaningo lokuqhathanisa olwalunezivivinyo ezingu-400, olwaluhlanganisa izimfanelo zokuziphatha ezingu-10 kanye nazo zonke izindawo ezine. Kuma-agent, izikhathi ezingu-366 kwezingu-400 zakhombisa izimfanelo ezifanele, noma zacindezela kahle leyo mfanelo lapho ingahambisani nesimo, okusho ukuthi zaphendula kahle ku-91.5% wezimo.
Le datha iyashintsha kuye ngendawo: ezimweni zemibuzo nezingxoxo, ukuvumelana kwenyuka kwaba phakathi kuka-92% no-96%, kanti ukunemba ezivivinyweni zokusebenzisa ama-app kwehla kwaba ngu-83%.
Iqoqo eliyisisekelo lama-persona ayisigidi, elinqunywe ngendlela eqondile futhi lahlungwa ngekhwalithi, selishicilelwe esidlangalaleni ku-Hugging Face ukuze lisetshenziswe ocwaningweni.
Isixhumanisi somthombo ovulekile:
https://github.com/MatrAIx-ai/MatrAIx-Persona-8B