Kwatanta Ayyukan ChatGPT, Bing Chat, da Bard akan Bayanan Ingilishi na Jarrabawar Kammala Sakandare ta Vietnam (VNHSGE)

1. Gabatarwa

Wannan takarda tana gabatar da kwatancen ayyuka na manyan tsarin harshe (LLMs) guda uku da suka shahara—ChatGPT na OpenAI (GPT-3.5), Bing Chat na Microsoft, da Bard na Google—akan bayanan Ingilishi na Jarrabawar Kammala Sakandare ta Vietnam (VNHSGE). Binciken yana da nufin kimanta iyawarsu a cikin takamaiman yanayin ilimin Ingilishi na sakandare a Vietnam, musamman yayin da ChatGPT ba a samunsa a hukumance a Vietnam. Binciken ya magance tambayoyi guda uku masu mahimmanci game da aikin tsarin, kwatanta da ɗaliban ɗan adam, da yuwuwar aikace-aikacen LLMs a cikin wannan yanayin na ilimi.

2. Ayyukan Da Suka Gabata

Takardar ta sanya kanta a cikin faɗin mahallin haɗa AI cikin ilimi, tana nuna yuwuwar canji na LLMs kamar tsarin BERT da GPT.

3. Hanyoyin Bincike

Babban hanyar binciken ta ƙunshi gudanar da bayanan Ingilishi na VNHSGE ga LLMs guda uku. Wataƙila bayanan sun ƙunshi tambayoyin gwaji da aka daidaita waɗanda ke kimanta ƙwarewar harshen Ingilishi a matakin sakandare. Ana auna aikin ta hanyar daidaiton amsoshin tsarin idan aka kwatanta da maɓallin amsa na hukuma.

4. Sakamakon Gwaji

Mahimman Binciken:

• Matsayin Aiki: Microsoft Bing Chat (92.4%) ya fi duka Google Bard (86%) da OpenAI ChatGPT (79.2%).
• Ma'anar Aiki: An gabatar da Bing Chat da Bard a matsayin madadin da za a iya amfani da su ga ChatGPT don ilimin Ingilishi a Vietnam, inda aka hana samun ChatGPT.
• Kwatanta da Dan Adam: Dukkan LLMs guda uku sun zarce matsakaicin aikin ɗaliban sakandare na Vietnam akan gwajin ƙwarewar Ingilishi ɗaya, wanda ke nuna yuwuwarsu a matsayin manyan albarkatun ilimi ko kayan taimako na koyarwa.

Bayanin Zane: Zanen sanduna zai yi tasiri wajen ganin wannan tsarin aiki, tare da axis-y yana wakiltar daidaito (%) kuma axis-x yana jera LLMs guda uku. Sandar Bing Chat za ta fi tsayi, sannan Bard, sannan ChatGPT. Za a iya nuna wani layi na bambanci don nuna matsakaicin makin ɗaliban Vietnam don kwatanta kai tsaye.

5. Tattaunawa

Sakamakon ya nuna babbar yuwuwar LLMs da ake samu a kasuwa a matsayin kayan aiki don ilimin harshen Ingilishi. Babban aikin Bing Chat na iya kasancewa saboda haɗewarsa da injin bincike, yana ba da damar samun ƙarin bayani na yanzu ko na takamaiman mahalli. Gaskiyar cewa duk tsarin sun zarci ɗaliban ɗan adam tana nuna canjin tsari, inda AI za ta iya zama ba kawai mataimakiya ba amma a matsayin ma'auni mai ƙwarewa, mai yuwuwar keɓance koyarwa da ba da ra'ayi nan take, daidai.

6. Bincike Na Asali & Sharhin Kwararru

Babban Fahimta: Wannan takarda ba ma'auni kawai ba ce; alama ce ta kasuwa. A wani yanki (Vietnam) inda babban tsarin (ChatGPT) ya ke da ƙofa, binciken yana gano da tabbatar da madadin aiki (Bing Chat, Bard) da gaske, yana bayyana hanya mai aiki, ta farko-aikace-aikace don karɓar AI a cikin ilimi. Binciken cewa duk LLMs sun zarci matsakaicin aikin ɗalibi ba kawai batu ne na ilimi ba—ƙarfi ne mai rushewa, yana nuna cewa rawar AI na iya haɓaka daga kayan aiki na ƙari zuwa babban wakili ko ma'auni na koyarwa.

Kwararar Hankali & Ƙarfafawa: Hanyar bincike tana da sauƙi kuma tana da tasiri: yi amfani da jarrabawar ƙasa da aka sani, mai mahimmanci a matsayin ma'aunin kimantawa. Wannan yana ba da sahihanci nan take, mai dacewa ga malamai da masu tsara manufofi. Mayar da hankali kan samun dama ("abin da ake samu da gaske") fiye da fifikon ka'idar babban ƙarfi ne, yana sa binciken ya zama mai aiki nan take. Ya yi daidai da abubuwan da cibiyoyi kamar Cibiyar Stanford don AI Mai Tsarin Dan Adam suka lura, waɗanda ke jaddada kimanta AI a cikin mahallin duniya na gaske, masu iyaka.

Kurakurai & Gibin Mai Muhimmanci: Binciken yana a saman. Ya ba da rahoton maki amma yana ba da kaɗan game da yanayin kurakurai. Shin tsarin ya kasa akan nahawu, fahimtar karatu, ko bambancin al'adu? Wannan kimantawar baƙar fata tana kwatanta iyaka a cikin fannin kanta. Bugu da ƙari, kwatanta da makin "matsakaici" na ɗalibi yana da zurfi a ƙididdiga. Ƙarin bincike mai ƙarfi, kama da ka'idar amsa abu da ake amfani da ita a cikin ilimin halin ɗan adam, zai iya tsara ƙwarewar tsarin zuwa takamaiman matakan ƙwarewa akan gwajin. Takardar kuma ta kewaye gaba ɗaya batun mai mahimmanci na yadda ake haɗa waɗannan kayan aikin. Samun AI mai babban maki kawai ba ya fassara zuwa ingantaccen ilimin koyarwa ba, ƙalubalen da aka rubuta sosai a cikin International Journal of Artificial Intelligence in Education.

Fahimta Mai Aiki: Ga malamai a cikin kasuwanni masu iyaka iri ɗaya, wannan takarda littafin wasa ne: 1) Yi ma'auni a cikin gida: Kada ku dogara da tashin hankali na duniya; gwada kayan aikin da ake samu akan takamaiman manhajar ku. 2) Duba bayan jagora: Tsarin gasa na iya ba da isasshen ko mafi kyawun aiki a cikin mahalli. 3) Mayar da hankali kan "yadda": Lokaci na gaba na bincike dole ne ya canza daga idan LLMs suna aiki zuwa yadda ake tura su cikin alhaki—ƙirƙira umarni waɗanda ke ƙarfafa tunani mai mahimmanci fiye da dawo da amsa, ƙirƙira tsare-tsare don kimantawar da AI ta ƙarfafa, da magance daidaito a cikin samun dama. Nasara ta gaske ba za ta zama babban makin gwajin AI ba, amma ingantattun sakamakon koyo na ɗan adam.

7. Cikakkun Bayanai Na Fasaha & Tsarin Lissafi

Duk da yake takardar ba ta shiga cikin tsarin tsarin ba, ana iya fassara aikin ta hanyar duban yiwuwa da daidaiton aiki. Babban ma'aunin kimantawa shine daidaito ($Acc$), wanda aka ayyana a matsayin rabo na abubuwan da aka amsa daidai zuwa jimillar abubuwan ($N$).

$Acc = \frac{\text{Yawan Amsoshi Daidai}}{N} \times 100\%$

Don ƙarin fahimta mai zurfi, mutum zai iya ƙirƙira aikin LLM akan abin gwaji mai zaɓi da yawa a matsayin rarraba yiwuwa akan amsoshi masu yuwuwa. Bari yuwuwar zaɓin amsar daidai $c$ daga cikin zaɓuɓɓukan $O$ ta tsarin ya zama $P_M(c | q, \theta)$, inda $q$ shine tambaya kuma $\theta$ yana wakiltar sigogin tsarin da duk wani mahalli da aka dawo dasu (musamman masu dacewa don haɓaka binciken Bing Chat). Makin ƙarshe shine tarin waɗannan yuwuwar a duk abubuwan. Gibin aiki tsakanin tsarin yana nuna bambance-bambance masu mahimmanci a cikin wakilcin ciki $\theta$ ko hanyoyin haɓaka dawo da su $R(q)$ don samar da $P_M$.

$P_{\text{BingChat}}(c|q) \approx P(c|q, \theta_{\text{Bing}}, R_{\text{Web}}(q))$

$P_{\text{ChatGPT}}(c|q) \approx P(c|q, \theta_{\text{GPT-3.5}})$

8. Tsarin Bincike: Nazarin Shari'ar Ba tare da Lambar Kwamfuta ba

Yanayi: Shugaban sashen Ingilishi a Hanoi yana son kimanta kayan aikin AI don tallafawa ɗaliban Aji 12.

Aikace-aikacen Tsarin:

1. Ayyana Manufar Gida: Inganta aikin ɗalibi akan sassan nahawu da fahimtar karatu na VNHSGE.
2. Gano Kayan Aiki & Duba Samun dama: Lissafa kayan aikin da ake samu: Bing Chat (ana iya samun dama), Google Bard (ana iya samun dama), ChatGPT (yana buƙatar VPN, ba a tallafawa a hukumance ba). Ba da fifiko ga biyun na farko bisa ga binciken wannan takarda.
3. Ma'auni Mai Zurfi: Kar a yi amfani da takardun gwaji na baya kawai. Ƙirƙiri gwajin bincike mai mai da hankali:
   • Rukunin A: Tambayoyin nahawu 20 (lokaci, gabatarwa).
   • Rukunin B: Tambayoyin fahimtar karatu 20.
   • Gudanar da rukunoni A & B ga Bing Chat da Bard. Yi rikodin ba kawai daidaito ba, har ma da dalilin da aka bayar a cikin amsoshinsu.
4. Binciken Kuskure & Taswira: Rarraba kurakuran da kowane AI ya yi. Misali: "Bing Chat ya kasa akan tambayoyin yanayi na ƙaddara 3/5; Bard ya ba da taƙaitaccen dalili amma wani lokacin cikakken dalili don tambayoyin zato."
5. Ƙirar Haɗawa: Bisa ga binciken: Yi amfani da Bing Chat don bayanin aikin nahawu saboda babban daidaito. Yi amfani da amsoshin Bard a matsayin "samfurin amsoshi" don fahimtar karatu, amma ƙirƙiri takardar aikin ɗalibi wanda ke tambaya: "Kwatanta taƙaitaccen Bard da naku. Me ya rasa?" Wannan yana haɓaka kimantawa mai mahimmanci maimakon karɓar shiru.

Wannan tsarin yana motsawa bayan "wane AI ya fi kyau" zuwa "yadda za mu iya amfani da ƙarfin kowane AI da dabara a cikin iyakokin mu na koyarwa."

9. Aikace-aikace Na Gaba & Hanyoyin Bincike

Aikace-aikace Nan Take:

• Tsarin Koyarwa Na Musamman: Tura Bing Chat ko Bard a matsayin ginshiƙi ga malaman AI waɗanda ke ba da aiki da bayani akan buƙata, da aka keɓance ga manhajar VNHSGE.
• Samar da Kayan Aiki ta Atomatik: Yin amfani da waɗannan LLMs don ƙirƙirar tambayoyin aiki, samfurin rubuce-rubuce, da sauƙaƙan bayani na rikitattun rubutun da suka dace da manhajar ƙasa.
• Kayan Aikin Tallafawa Malami: Taimaka wa malamai wajen yin maki, ba da ra'ayi kan rubutun ɗalibi, da samar da ra'ayoyin shirin darasi.

Hanyoyin Bincike Masu Muhimmanci:

• Injiniyan Umarni don Ilimin Koyarwa: Bincike na tsari a cikin ƙirƙira umarni waɗanda ke tilasta LLMs su bayyana dalili, gane kuskuren ɗalibi, ko tsarin koyo maimakon kawai ba da amsoshi.
• Nazarin Tasirin Tsawon Lokaci: Shin yin amfani da malamin LLM yana inganta sakamakon koyo na ɗalibi da makin jarrabawa a cikin semester ko shekara? Ana buƙatar bincike mai sarrafawa.
• Kimantawa Mai Yawa: Jarrabawar gaba mai mahimmanci na iya haɗa sassan baki. Kimanta iyawar ganin magana da samar da magana na LLMs a cikin mahallin ilimi shine iyaka na gaba.
• Daidaito da Samun dama: Bincike cikin rage haɗarin faɗaɗa rarrabuwar dijital—tabbatar da fa'idodin sun isa ɗaliban da ke cikin makarantun da ba su da isassun albarkatu ba tare da ingantaccen intanet ko na'urori ba.
• Daidaitawar Al'adu & Mahalli: Gyara ko haɓaka hanyoyin dawo da su waɗanda ke ba da damar LLMs na duniya su fahimta da kuma komawa ga kayan ilimi na gida na Vietnam, tarihi, da al'adu.

10. Nassoshi

1. Dao, X. Q. (2023). Kwatanta Ayyukan Manyan Tsarin Harshe akan Bayanan Ingilishi na VNHSGE: OpenAI ChatGPT, Microsoft Bing Chat, da Google Bard. arXiv preprint arXiv:2307.02288v3.
2. OpenAI. (2023). ChatGPT: Daidaita Tsarin Harshe don Tattaunawa. OpenAI Blog.
3. Kasneci, E., et al. (2023). ChatGPT don alheri? Kan damammaki da ƙalubale na manyan tsarin harshe don ilimi. Koyo da Bambance-bambancen Mutum, 103, 102274.
4. Kung, T. H., et al. (2023). Aikin ChatGPT akan USMLE: Yuwuwar ilimin likitanci mai taimakon AI ta amfani da manyan tsarin harshe. PLOS Digital Health, 2(2), e0000198.
5. Cibiyar Stanford don Harkokin Wucin Gadi na Dan Adam (HAI). (2023). Rahoton Shekara-shekara na AI Index 2023. Jami'ar Stanford.
6. Ƙungiyar Duniya don Harkokin Wucin Gadi a Ilimi (IAIED). International Journal of Artificial Intelligence in Education.
7. Thorp, H. H. (2023). ChatGPT yana da daɗi, amma ba marubuci ba. Kimiyya, 379(6630), 313.