1. Gabatarwa

Shahararriyar ChatGPT da ba a taɓa ganin irinta ba tana nuna sauyi mai girma a yadda mutane ke mu'amala da fasaha don dalilai na ilimi. Wannan takarda tana binciken fasahar da ba a saba da ita ba ta injiniyanci na umarni a tsakanin ɗaliban sakandare na Ingilishi a matsayin Harshen Waje (EFL). Duk da yake Manyan Samfuran Harshe (LLMs) irin su ChatGPT suna ba da dama mai yawa don tallafawa ci gaban rubutu, amfaninsu ya dogara ne da ikon mai amfani na ƙirƙirar umarni masu daidaito, masu inganci. Wannan binciken ya ɗauki tsarin lokaci na gaske, na gwaji-da-kuskure na sababbin masu amfani, yana nazarin abubuwan da ke ciki, inganci, da kuma ci gaban umarninsu don kammala aikin rubutu da aka ƙayyade. Sakamakon ya bayyana hanyoyin halaye daban-daban, yana jaddada buƙatar gaggawa na tsarin ilimin injiniyanci na umarni a cikin manhajojin EFL don motsa ɗalibai daga gwaji mara inganci zuwa haɗin gwiwar dabarun tare da AI.

2. Bita na Adabi & Bayanan Baya

2.1 Tashin SOTA Chatbots

Mafi kyawun chatbots na AI masu samarwa (SOTA), waɗanda ChatGPT ya zama misali, suna wakiltar tsalle mai girma daga na baya waɗanda suka dogara da ƙa'ida. Waɗanda aka yi amfani da su ta hanyar samfuran harshe na jijiyoyin jiki da aka horar da su akan tarin rubutu masu yawa, suna samar da rubutu mai kama da na ɗan adam bisa ga hasashen yuwuwar, suna ba da damar mu'amala mai sassauƙa da sanin mahallin (Caldarini et al., 2022). "ChatGPT" ana amfani da shi sosai a matsayin kalma gama gari don wannan nau'in AI, yana kafa sabon ma'auni na aiki.

2.2 Injiniyanci na Umarni a matsayin Fasaha Mai Muhimmanci

Injiniyanci na umarni shine fasaha da kimiyya na ƙirƙirar shigarwa don jagorantar LLM zuwa ga sakamakon da ake so. Ba fasaha ce kawai ba amma wani nau'i ne na tunani na lissafi da sanin yare. Umarni masu inganci sau da yawa suna buƙatar bayyanawa, mahallin, ƙuntatawa, da misalai (ƙaramin koyo). Ga masu amfani marasa fasaha, wannan yana gabatar da babban matakin koyo, wanda galibi ana siffanta shi da zato na maimaitawa.

2.3 AI a cikin Ilimin EFL

Bincike kan AI a cikin koyon harshe ya mai da hankali kan kimanta rubutu ta atomatik (AWE) da tsarin koyarwa mai hankali. Yanayin samarwa na mu'amala na SOTA chatbots yana gabatar da sabon yanayi—wanda ke canza matsayin ɗalibi daga mai karɓar ra'ayi zuwa daraktan kayan aikin fahimi. Wannan yana buƙatar sabbin ilimin karatu, haɗa ƙwarewar rubutu na al'ada tare da dabarun mu'amala na AI.

3. Hanyoyin Bincike

3.1 Mahalarta & Tattara Bayanai

Binciken ya haɗa da ɗaliban sakandare EFL a Hong Kong waɗanda ba su taɓa amfani da SOTA chatbots ba. An ba mahalarta aikin kammala takamaiman aikin rubutu (misali, maƙala mai gardama ko sakin layi mai siffantawa) ta amfani da ChatGPT. Bayanai na farko sun haɗa da rikodin allo na iPad, waɗanda suka ɗauki cikakken jerin umarni, amsoshin ChatGPT, da duk wani gyara da ɗalibai suka yi.

3.2 Tsarin Nazari

An yi amfani da hanyar nazarin hali na inganci. An fassara rikodin allo kuma an ƙididdige su tare da manyan fuskoki guda biyu: (1) Abubuwan da ke cikin Umarni (misali, ƙayyadaddun aiki, buƙatun salo, umarnin gyara) da (2) Tsarin Mu'amala (misali, adadin juzu'i, daidaitawa bisa ga sakamako). An tattara alamu don gano hanyoyin masu amfani daban-daban.

4. Sakamako: Hanyoyin Injiniyanci na Umarni Guda Hudu

Nazarin rikodin allo ya bayyana hanyoyi guda huɗu na asali, waɗanda ke wakiltar haɗuwa daban-daban na tsarin dabarun da ingancin umarni.

Rarraba Hanyoyi

Bisa ga alamu da aka lura a cikin ƙungiyar.

  • Mai Tsauraran Ra'ayi (The Minimalist): ~35%
  • Mai Gyara A Jere (The Iterative Refiner): ~30%
  • Mai Tsara Tsari (The Structured Planner): ~20%
  • Mai Bincike Ta Hanyar Tattaunawa (The Conversational Explorer): ~15%

4.1 Mai Tsauraran Ra'ayi (The Minimalist)

Waɗannan masu amfani suna shigar da umarni gajere sosai, sau da yawa jimla ɗaya wanda ke kwatanta umarnin aikin asali (misali, "Rubuta maƙala game da sauyin yanayi"). Suna nuna ƙarancin haƙuri don maimaitawa; idan sakamakon farko bai gamsu ba, suna iya yin watsi da kayan aikin ko ƙaddamar da sakamakon da bai dace ba. Wannan hanyar tana nuna kuskuren kayan aiki-a matsayin annabci.

4.2 Mai Gyara A Jere (The Iterative Refiner)

Wannan rukuni yana farawa da umarni mai sauƙi amma yana shiga cikin tsarin gyara na layi. Bisa ga sakamakon AI, suna ba da umarni na biyo baya kamar "ƙara tsawonsa," "yi amfani da kalmomi masu sauƙi," ko "ƙara ƙarin misalai." Mu'amalar tana da amsawa kuma tana ƙaruwa, tana nuna fahimtar da ke tasowa game da amsawar AI ga umarni amma ba ta da tsari mai girma.

4.3 Mai Tsara Tsari (The Structured Planner)

Ƙananan ɗalibai sun tunkari aikin tare da tsari da aka riga aka tsara. Umarninsu na farko sun kasance masu cikakken bayani, suna ƙayyadad da tsari, sautin murya, mahimman abubuwa, kuma wani lokacin suna ba da jigo (misali, "Rubuta maƙala mai sakin layi 5 don yin gardama don makamashin sabuntawa. Sakin layi na 1: Gabatarwa. Sakin layi na 2: Amfanin tattalin arziki... Yi amfani da sautin murya na yau da kullun."). Wannan hanyar tana samar da sakamako masu inganci tare da ƙarancin juzu'i, yana nuna ci gaban rarraba aiki da tsara tunani.

4.4 Mai Bincike Ta Hanyar Tattaunawa (The Conversational Explorer)

Waɗannan masu amfani suna ɗaukar ChatGPT a matsayin abokin tattaunawa. Maimakon kawai ba da umarni, suna yin tambayoyin meta ("Yaya zan iya inganta bayanina na zance?") ko neman bayani ("Me ya sa ka zaɓi wannan kalma?"). Wannan hanyar tana haɗa taimakon rubutu tare da koyo game da rubutu, ko da yake tana iya karkata kuma bazai iya kammala ainihin aikin da inganci ba.

5. Tattaunawa & Abubuwan Da Aka Samu

5.1 Matsawa Bayan Gwaji-da-Kuskure

Yawaitar hanyoyin Mai Tsauraran Ra'ayi da Mai Gyara A Jere suna nuna babban gibi. Idan aka bar su da kansu, yawancin ɗalibai ba sa haɓaka dabarun injiniyanci na umarni da suka ci gaba da kansa. Tsarinsu ba shi da inganci kuma sau da yawa ya kasa amfani da cikakken iyawar AI, wanda zai iya ƙarfafa halayen koyo mara aiki.

5.2 Haɗa Koyarwa

Binciken yana jayayya don bayyanannen ilimin injiniyanci na umarni a cikin ajin rubutu na EFL. Wannan ya kamata ya haɗa da:

  • Koyarwa Kai Tsaye: Koyar da sassan umarni (matsayi, aiki, mahallin, ƙuntatawa, misalai).
  • Tsare-tsare Masu Tsari: Gabatar da samfura kamar RTF (Matsayi, Aiki, Tsari) ko CRISPE (Ƙarfi, Matsayi, Fahimta, Bayani, Halin Mutum, Gwaji).
  • Zargi da Nazari: Kimanta sakamakon da AI ya samar don fahimtar alaƙar dalili-da-sakamako tsakanin umarni da samfurin.
  • La'akari da Da'a: Tattauna marubuci, satar aikin wasu, da kimanta abubuwan AI mai mahimmanci.

Manufar ita ce haɓaka ɗalibai waɗanda suke daraktoci masu dabaru maimakon masu amfani marasa aiki na rubutun da AI ya samar.

6. Nazarin Fasaha & Tsari

Babban Fahimta, Gudun Hankali, Ƙarfafawa & Kurakurai, Abubuwan Da Za'a Iya Aiwatarwa

Babban Fahimta: Wannan takarda tana ba da gaskiya mai mahimmanci, wacce sau da yawa ake rasa: ƙaddamar da kayan aikin AI kamar ChatGPT ba ta ƙaddamar da iyawa ta atomatik ba. Mu'amalar tana da sauƙi mai yaudara, amma nauyin fahimi na ingantaccen mu'amala yana da yawa. Ainihin matsalar a cikin "ajin da AI ya ƙarfafa" ba shiga fasaha ba ne; shine rashin ilimin karatu na mu'amala. Binciken yana canza hankali sosai daga sakamakon AI zuwa shigarwar ɗan adam, yana fallasa tsantsar, tsarin koyo mara ado.

Gudun Hankali: Gardamar tana da tsari kuma tana da ƙarfi. Ta fara ne ta hanyar kafa matsalar (SOTA chatbots suna buƙatar umarni mai fasaha), gabatar da gibin ilimi (ta yaya sababbin masu amfani ke yin haka?), gabatar da cikakken shaida na zahiri (hanyoyi huɗu), kuma ta ƙare da kira mai ƙarfi zuwa aiki (ilimi dole ne ya daidaita). Amfani da nazarin hali ya kafa ka'idar a cikin gaskiyar da ba ta da kyau.

Ƙarfafawa & Kurakurai: Babban ƙarfi shine ingancin muhalli. Yin amfani da rikodin allo na masu amfani na farko a cikin mahallin aiki na gaske yana ba da bayanai na gaske waɗanda binciken dakin gwaji sau da yawa ya rasa. Rubutun hanyoyi huɗu yana da fahimta kuma yana ba da tsari mai ƙarfi ga malamai don gano halayen ɗalibi. Babban aibi, wanda marubutan suka yarda da shi, shine ma'auni. Wannan nazarin hali ne mai zurfi, ba bincike mai faɗi ba. Hanyoyin suna misalta, ba ƙididdiga ba. Bugu da ƙari, binciken ya mai da hankali kan tsari, ba kimanta ingancin samfurin rubutu na ƙarshe a duk faɗin hanyoyi ba—wani mataki mai mahimmanci na gaba.

Abubuwan Da Za'a Iya Aiwatarwa: Ga malamai da masu tsara manhajoji, wannan takarda kira ce ta farkawa. Tana ba da umarni bayyananne: Injiniyanci na umarni shine ainihin ilimin karatu na ƙarni na 21 kuma dole ne a koya shi, ba a kama shi ba. Makarantu yakamata su haɓaka ƙananan darussan da suka haɗa tsare-tsare kamar Samfurin Matsayin Umarni, wanda ke motsawa daga umarni na asali ($P_{cmd}$) zuwa umarni masu tunani mai maimaitawa ($P_{reason}$). Misali, koya wa ɗalibai dabarar umarni mai inganci: $P_{optimal} = R + T + C + E$, inda $R$ shine Matsayi, $T$ shine Aiki, $C$ shine Ƙuntatawa, kuma $E$ shine Misalai. Kamfanonin EdTech yakamata su gina waɗannan ginshiƙan koyarwa kai tsaye a cikin mu'amalarsu, suna ba da samfuran ginin umarni da jagora da ra'ayi, suna matsawa bayan akwatin rubutu mara komai.

Cikakkun Bayanan Fasaha & Tsarin Lissafi

Daga mahangar koyon injini, umarnin mai amfani $p$ yana aiki azaman mahallin sharadi don samfurin harshe $M$. Samfurin yana samar da jerin sakamako $o$ bisa ga rarraba yuwuwar $P(o | p, \theta)$, inda $\theta$ ke wakiltar sigogin samfurin. Umarni mai inganci yana rage ɓacin rai na wannan rarraba sakamako, yana kai shi zuwa ga manufar mai amfani $t$. Kalubalen ɗalibi shine rage bambanci tsakanin rarraba yuwuwar sakamako da manufarsu, wanda aka tsara shi azaman rage $D_{KL}(P(o|p, \theta) \,||\, P(o|t))$, inda $D_{KL}$ shine bambancin Kullback–Leibler. Sababbin masu amfani, ta hanyar gwaji-da-kuskure, suna yin ingantaccen ingantaccen ɗan adam a cikin madauki na $p$ don cimma wannan.

Misalin Hali na Tsarin Nazari

Yanayi: Dalibi dole ne ya rubuta wasiƙar lallashi zuwa ga shugaban makaranta game da fara shirin sake yin amfani da kaya.

Hanyar Mai Tsauraran Ra'ayi (Mara Inganci):
Umarni 1: "Rubuta wasiƙa game da sake yin amfani da kaya."
Sakamako: Wasiƙa gama gari, mara daɗi.
Aikin Dalibi: Ya ƙaddamar da sakamako tare da ƙananan gyare-gyare.

Hanyar Mai Tsara Tsari (Mai Inganci - Ta Amfani da Tsarin RTF):
Umarni 1: "Yi aiki a matsayin ɗalibi na aji 10 mai damuwa. Rubuta wasiƙar lallashi ta yau da kullun zuwa ga babban shugaban makarantar sakandare. Manufar ita ce shawo kan su aiwatar da cikakken shirin sake yin amfani da robobi da takarda a cikin ɗakin cin abinci da azuzuwan. Yi amfani da sautin murya mai girmamawa amma mai gaggawa. Haɗa da gardama guda uku: 1) Tasirin muhalli, 2) Damar shiga ɗalibai/jagoranci, 3) Yuwuwar ceton kuɗi ko tallafi. Tsara wasiƙar tare da kwanan wata, gaisuwa, sakin layi na jiki na kowane gardama, da sa hannu na rufewa."
Sakamako: Wasiƙa mai kyau, mai tsari, mai manufa, da lallashi.
Aikin Dalibi: Ya sake duba sakamakon, yana iya neman gyara: "Ƙarfafa gardama ta uku game da ceton kuɗi ta hanyar ƙara ƙididdiga."

Wannan bambance-bambancen yana nuna yadda amfani da tsari mai sauƙi (Matsayi: ɗalibi, Aiki: rubuta wasiƙa, Tsari: na yau da kullun tare da takamaiman gardama) ke inganta inganci da ingancin haɗin gwiwar AI.

Sakamakon Gwaji & Bayanin Chati

Mahimman sakamakon binciken suna da inganci, an kama su a cikin bayanin hanyoyin. Ƙarin ƙididdiga na hasashe zai iya haifar da chati kamar: "Hoto 1: Ingantaccen Mu'amala vs. Ingancin Sakamako ta Hanyar." Axis na x zai wakilci adadin juzu'in umarni (sabanin inganci), kuma axis na y zai wakilci makin ingancin rubutun ƙarshe (misali, an tantance shi ta hanyar rubutu). Muna tsammanin:
- Mai Tsauraran Ra'ayi ya taru a cikin babban inganci (ƙananan juzu'i) amma ƙananan inganci.
- Mai Gyara A Jere ya nuna matsakaici-zuwa-babban juzu'i tare da inganci mai canzawa.
- Mai Tsara Tsari ya mamaye babban inganci, babban inganci (ƙananan juzu'i, babban maki).
- Mai Bincike Ta Hanyar Tattaunawa ya kasance a cikin ƙananan inganci (babban juzu'i) tare da inganci mai canzawa, mai yuwuwa babba idan binciken ya mai da hankali. Wannan hangen nesa zai yi gardama mai ƙarfi cewa hanyar Mai Tsara Tsari tana wakiltar manufar mafi kyau don koyarwa.

7. Aikace-aikace na Gaba & Jagorori

Abubuwan da aka samu na wannan binciken sun wuce ajin EFL:

  • Malamai na Daidaitawar Umarni: Haɓaka malamai masu amfani da AI waɗanda ke nazarin tarihin umarnin ɗalibi, gano hanyarsu, kuma suna ba da ra'ayi na lokaci na gaske, mai ginshiƙi don jagorance su zuwa ga dabarun da suka fi inganci (misali, "Gwada ƙayyadad da masu sauraronku a cikin umarni na gaba").
  • Ilimin Karatu na Tsakanin Fannoni: Haɗa injiniyanci na umarni cikin ilimin STEM don samar da lamba, tambayoyin nazarin bayanai, da bayanin kimiyya, kamar yadda cibiyoyi kamar shirin RAISE na MIT suka ba da shawarar.
  • Shirye-shiryen Ma'aikata: Kamar yadda aka lura a cikin rahotanni daga Dandalin Tattalin Arzikin Duniya, injiniyanci na umarni yana zama fasaha mai daraja a cikin sana'o'i cikin sauri. Ilimin sakandare dole ne ya shirya ɗalibai don wannan gaskiyar.
  • Nazarin Tsawon Lokaci: Bin diddigin yadda ƙwarewar injiniyanci na umarni ke tasowa akan lokaci tare da koyarwa, da kuma yadda suke da alaƙa da haɓaka a cikin rubutu na al'ada da ƙwarewar tunani mai mahimmanci.
  • Umarni Masu Nau'i Daban-daban: Bincike na gaba dole ne ya bincika injiniyanci na umarni don AI masu nau'i daban-daban (misali, DALL-E, Sora), inda umarni suka haɗa da ƙuntatawa na gani, na ɗan lokaci, da na salo—wani yanki na ilimin karatu mai rikitarwa.

8. Nassoshi

  1. Caldarini, G., Jaf, S., & McGarry, K. (2022). A Literature Survey of Recent Advances in Chatbots. Information, 13(1), 41.
  2. Woo, D. J., Guo, K., & Susanto, H. (2023). Cases of EFL Secondary Students’ Prompt Engineering Pathways to Complete a Writing Task with ChatGPT. [Manuscript in preparation].
  3. Zhao, W. X., et al. (2023). A Survey of Large Language Models. arXiv preprint arXiv:2303.18223.
  4. Moor, J. (2006). The Dartmouth College Artificial Intelligence Conference: The Next Fifty Years. AI Magazine, 27(4), 87–91.
  5. MIT RAISE. (2023). Day of AI Curriculum. Massachusetts Institute of Technology. Retrieved from [https://www.dayofai.org/]
  6. World Economic Forum. (2023). Future of Jobs Report 2023.
  7. Reynolds, L., & McDonell, K. (2021). Prompt Programming for Large Language Models: Beyond the Few-Shot Paradigm. Extended Abstracts of the 2021 CHI Conference on Human Factors in Computing Systems.