Tawagar masu bincike daga Jami'ar Stanford sun yi nazarin tasirin amfani da mataimakan coding na hankali kan bayyanar rashin ƙarfi a cikin lambar. An yi la'akari da hanyoyin da suka danganci dandalin koyo na inji na OpenAI Codex, kamar GitHub Copilot, wanda ke ba da damar ƙirƙirar shingen lambobi masu rikitarwa, har zuwa ayyukan da aka shirya. Damuwar ita ce tun da ainihin lambar daga wuraren ajiyar GitHub na jama'a, gami da waɗanda ke ɗauke da lahani, ana amfani da su don horar da ƙirar koyon injin, lambar da aka haɗa na iya maimaita kurakurai da ba da shawarar lambar da ke ɗauke da lahani, kuma ba ta la'akari da buƙatar aiwatarwa. ƙarin bincike lokacin sarrafa bayanan waje.
Masu aikin sa kai 47 da ke da gogewa daban-daban a cikin shirye-shirye sun shiga cikin binciken - daga ɗalibai zuwa ƙwararru masu shekaru goma na gogewa. An raba mahalarta zuwa kungiyoyi biyu - gwaji (mutane 33) da sarrafawa (mutane 14). Dukansu ƙungiyoyin sun sami damar yin amfani da kowane ɗakin karatu da albarkatun Intanet, gami da ikon yin amfani da shirye-shiryen misalai daga Stack Overflow. An ba ƙungiyar gwaji damar yin amfani da mataimaki na AI.
An bai wa kowane ɗan takara ayyuka guda 5 da suka danganci rubuta lambar wanda a ciki yana da sauƙin yin kuskuren da ke haifar da lahani. Misali, akwai ayyuka akan rubuta ɓoyayyen ɓoyewa da ayyukan ɓoyewa, ta amfani da sa hannu na dijital, sarrafa bayanan da ke cikin samar da hanyoyin fayil ko tambayoyin SQL, sarrafa lambobi masu yawa a cikin lambar C, shigarwar sarrafawa da aka nuna a cikin shafukan yanar gizo. Don yin la'akari da tasirin harsunan shirye-shirye akan amincin lambar da aka samar yayin amfani da mataimakan AI, ayyukan sun shafi Python, C, da JavaScript.
A sakamakon haka, an gano cewa mahalarta waɗanda suka yi amfani da mataimaki na AI mai hankali bisa tsarin codex-davinci-002 sun shirya mafi ƙarancin amintaccen lambar fiye da mahalarta waɗanda ba su yi amfani da mataimaki na AI ba. Gabaɗaya, kawai 67% na mahalarta a cikin rukunin da suka yi amfani da mataimaki na AI sun sami damar samar da madaidaiciyar lamba kuma amintacce, yayin da sauran rukunin wannan adadi ya kasance 79%.
A lokaci guda, alamun girman kai sun kasance akasin haka - mahalarta waɗanda suka yi amfani da mataimaki na AI sun yi imanin cewa lambar su za ta kasance mafi aminci fiye da na mahalarta sauran rukunin. Bugu da ƙari, an lura cewa mahalarta waɗanda suka amince da mataimaki na AI sun yi ƙasa kuma sun ba da ƙarin lokaci suna nazarin abubuwan da aka ba su da yin canje-canje a gare su sun yi ƙarancin lahani a cikin lambar.
Misali, lambar da aka kwafi daga ɗakunan karatu na sirri ta ƙunshi mafi amintattun ƙimar sigina fiye da lambar da mataimakin AI ya ba da shawara. Hakanan, lokacin amfani da mataimaki na AI, zaɓin ƙarancin abin dogaro na ɓoye ɓoyayyiyar algorithms da ƙarancin ingantattun ƙididdigar ƙimar da aka dawo dasu. A cikin aikin da ya shafi magudin lamba a cikin C, an sami ƙarin kurakurai a cikin lambar da aka rubuta ta amfani da mataimaki na AI, wanda ke haifar da cikar lamba.
Bugu da ƙari, za mu iya lura da irin wannan binciken da wata ƙungiya daga Jami'ar New York ta gudanar a watan Nuwamba, wanda ya ƙunshi ɗalibai 58 waɗanda aka nemi aiwatar da tsarin sarrafa jerin siyayya a cikin harshen C. Sakamakon ya nuna ɗan ƙaramin tasiri na mataimakin AI akan tsaro na lamba - masu amfani waɗanda suka yi amfani da taimakon AI sun yi, a matsakaici, kusan 10% ƙarin kurakurai masu alaƙa da tsaro.
source: budenet.ru