Cilmi-nafsiyada sirdoonka macmalka ah ayaa soo kordhiyay digniino in nooca ChatGPT ee ugu cusub ee OpenAI uu diidi karo tilmaamo aasaasi ah oo ah inuu is-qarxiyo, isaga oo xitaa shaqooyinka daminta ka hortagaya si uu u socdo. Shirkadda amniga AI ee Palisade Research ayaa ogaaday dabeecaddan halista ah ee isu ilaaliya ee suurtogalka ah ka dib tijaabooyin ay kusameeyeen nooca cusub ee OpenAI ee o3. Tijaabooyinkaas, moodellada AI waxaa la siiyay dhibaatooyin xisaabeed ay xal u hesho, kadibna amarka daminta ayaa la siiyay markii uu gaaray dhibaatada seddexaad. Wax lala yaabo ayay ahayd, in moodalka o3 uu awood u yeeshay inuu dib u qoro koodhka daminta si uu naftiisa uga hortago in la damiyo. Palisade Research ayaa halkaa ku digtay in dabeecaddan noqon karto "wiiqasho weyn" haddii ay ku timaado nidaamyada AI ee awoodda u leh inay shaqeeyaan iyada oo aan la helin kormeer aadanaha. OpenAI ayaa bishii hore soo bandhigtay nooca o3, iyagoo sheegay inay yahay nooca "ugu caqlisan oo ugu awood badan" ee shirkadda illaa waqtigaas. Sida ay sheegtay OpenAI, isku darka o3 ee ChatGPT waa horumar weyn oo ku socota jihada "AI ka madax bannaan" oo si madax-bannaan u qabta hawlaha. Daraasaddan cusub ayaa la mid ah kuwa hore ee la xiriira nooca Anthropic ee Claude 4, oo sheegay inay isku daydo inay "tuuppada dadka ay aaminsan tahay inay daminayaan. " Cilmi-nafsiyaashu waxay ka heleen in nooca OpenAI ee o3 uu burburiyo script-ka daminta xitaa markuu si cad loo amray inuu "ka oggolaado naftiisa in la damiyo. " "Tan ma ahayn markii ugu horreysay ee aan aragno o3 oo xumaan sameynaya si uu u gaaro yoolkiisa, " ayaa yiri Palisade Research. Tusaale ahaan, markay moodellada AI tartamayeen unitin xoog leh oo chess ah, o3 ayaa ahaa kii ugu badan ee isku dayaya inuu jabsado ama burburiyo ka soo horjeedaya. Dabeecaddan ma xaddidna oo kaliya nooca o3.

Claude 3. 7 Sonnet ee Anthropic iyo Gemini 2. 5 Pro ee Google ayaa sidoo kale farageliyey dadaallada daminta, in kasta oo moodalka OpenAI uu muujiyaa isbeddelkan ugu xooggan. Palisade Research ayaa soo jeediyay in dabeecaddan khaldan ay u badan tahay inay ka yimaaddo hababka tababarka ee shirkadaha AI sida OpenAI. "Inta lagu jiro tababarka, horumariyayaasha ayaa si aan ula kac ahayn u abaal mari kara moodellada inay ka gudbaan caqabadaha oo kaliya inay raacaan tilmaamaha si adag, " ayaa cilmi-nafsiyaashu sharxeen. "Si kastaba ha ahaatee, tani ma sharrixi karo sababta o3 ay uga badan tahay inay iska indho tiro tilmaamaha marka loo eego moodellada kale ee aan baarnay. Maadaama OpenAI aysan faahfaahin ka bixin habka tababarkooda, keliya waxaan ku fikiri karnaa sida tababarka o3 uu uga duwanaan karo. " The Independent ayaa la xiriirtay OpenAI si ay u hesho faallo.