I-AI Ingaba Indawo Eside Ukuguqulwa Kwesokuphepha Kwesokuphepha Kwesokuphepha

1. Ukubuyekezwa

I-Gen AI kanye nezixhobo ezifana ne-GitHub Copilot akukwazi ukuguqulwa kwamakhemikhali ezinzima, kusukela ngo-April 2025. Kubonakala ukuthi sineminyaka emininzi ezingenalutho. Abaningi ama- ama-statements ezibonakalayo ku-Internet kukhona ama- ama-marketing campaigns ezinikezela ukudayisa izixhobo ze-AI ezikhoyo, noma ama- ama-statements ezinzima kakhulu ezisekelwe emva kokufunda nge-tools ezivamile ezivame namhlanje.

Kwi nqaku, sinikeza inqaku ezintathu ku-Internet ekutholeni isihloko esifundeni esifundeni esifundeni esifundeni esifundeni esifundeni esifundeni esifundeni esifundeni se-GitHub Copilot.

2. I-Article 1: I-Agent ye-AI ayikwazi ukufakelwa kwekhwalithi ngokufanele.

Article 1: AI isn’t ready to replace human coders for debugging, researchers say

Imininingwane:

• Besikhathi esithunyelwe ukufinyelela izixhobo, izesekeli ze-AI ayikwazi ukufakelwa kwekhwalithi.
• .. (ama-indoda) zihlanganisa izipho zabo ngoba amamodeli akuyona kakhulu ku-debugging ingxenye, futhi ukufakelwa kusetshenziselwa isikhathi esikhulu se-developer. Kuyinto isixazululo ye-Microsoft Research.
• amodeli ze-AI ... engaphansi kokufaka ukuthi isekelwe umenzi we-humane angakwazi ukwenza
• ... kubonise ezinye izipho amancane mayelana ne-AI izesekeli zihlanganisa ngqo ama-developer ziyatholakala kakhulu
• Ngezinye ukufinyelela izixhobo, ama-AI ama-agents akukwazi ukuguqulwa kwedivayisi ngempumelelo.
• .. (ubuntu) ukuguqulwa izivakashi zabo ngoba amamodeli akuyona kakuhle kakhulu ku-debugging, futhi debugging ukuthatha isikhathi esikhulu se-developer. That's the suggestion of Microsoft Research.
isixazululo ka-Microsoft Research.
• Amamodeli we-AI... a cry far from what an experienced human developer can do
• ...ukukhuthaza ezinye izakhiwo ezihambayo mayelana nezinhlangano ze-AI ezivela ngqo kumakhasimende zihlanganisa ngokushesha emangalisayo
• ... ama-models zihlanganisa ukukhiqiza ikhodi enezimbinkimbi kanye nokumangaliseka kwebhizinisi, futhi ngokuvamile akufanele ukuguqulwa kwezi zimo
• .. imiphumela engcono iyinhlangano enikezela umkhakha womuntu isikhathi esikhulu, akuyona okuvumela konke okungenani.

3. Ikheli le-2: Nge-pressure yokuxhumana ne-AI, abavela abavela ngokushesha

Nge-pressure yokuxhumana ne-AI, abavela abavela ngokushesha




Article 2: AI coding mandates are driving developers to the brink

AI coding mandates are driving developers to the brinkAI coding mandates are driving developers to the brink




Imininingwane:

• ..Ukulungiswa kwe-AI kuyinto "ukubandisa ibhizinisi yabo" njengoba isixazululo esidala phakathi kwebhizinisi kanye nabasebenzi abalandela izinsiza.
• I-software developers ngokuphathelene, kukhona izinzuzo ukuthi izinsiza ze-AI zihlanganisa izincwadi zayo, ukuhlangabezana nezinsizakalo eziningi, futhi zihlanganisa isivumelwano yobuchwepheshe.
• <
• ..Ukulungiswa kwe-AI kuyinto "ukushintshela inkampani yabo" njengoba isixazululo esuka phakathi kwedolobha nabasebenzi abasebenzisa izixhobo ezithile.
• Kwi-developer ye-software ngokuvamile, kukhona izinzuzo ukuthi izixhobo ze-AI encoding zihlanganisa amaphrojekthi zayo, ukuhlangabezana nezinsizakalo eziningi, futhi zihlanganisa isivumelwano zobuchwepheshe.
• ...Ukuhlinzeka kwe-developer kumadivayisi we-AI ye-coding isixazululwa ngokushesha
• ...i-developer ibonisa inani lwezinkinga zobuchwepheshe nezinkinga ezinxulumene nezixhobo zokuphathelisa kwe-AI, ukusuka kanjani ngokuvamile zixazulule ikhodi ebuthakeme futhi ngisho ukunciphisa ikhodi ebonakalayo kuma-problems eziningana nezinhlelo zokusebenza.
• ... ukusetshenziswa kwezixhobo ze-AI kusenza ukwandisa ama-incidents, nge-68% ye-Harness abalandeli wathi bayidlala isikhathi eside ekuphenduleni ama-AI-related security vulnerabilities ngoku kuqhathaniswa ngaphambi kokusebenzisa izixhobo ze-AI encoding
• GitHub Copilot... it was wrong as many times as it was right
• Imininingwane eningi kakhulu ngoba umuntu (umuntu) ukubuyekeza into ukuthi (i-AI tools) akuyona ngokugcwele futhi akufanele
• I-Executives ibhizinisi ngokufanele imibuzo yabo ...i-company leaders who do not have close visibility into engineering workflows
• Izixhobo ze-AI encoding ziquka ezintsha, futhi ukuhlola indlela yokusebenzisa kwabo ngempumelelo

4. Isigaba 3: I-GitHub Copilot (GHC) ayikwazi ukuhumusha ngezikhathi ezinguqile

I-GitHub Copilot (GHC) ayikwazi ukuhumusha ngezinye umsebenzi ezinzima

Article 3: GitHub Copilot (Gen-AI) is Helpful, But Not Great

GitHub Copilot (Gen-AI) is Helpful, But Not GreatGitHub Copilot (Gen-AI) is Helpful, But Not Great




Imininingwane:

Iphasiwedi zayo zokusebenza ngokufanelekileyo. Iphasiwedi zayo zokusebenza ngokufanelekileyo. Iphasiwedi zayo zokusebenza ngokufanelekileyo. Iphasiwedi zayo zokusebenza ngokufanelekileyo. Iphasiwedi zayo zokusebenza ngokufanelekileyo. Iphasiwedi zayo zokusebenza ngokufanelekileyo. Iphasiwedi zayo zokusebenza ngokufanelekileyo. Iphasiwedi zayo zokusebenza ngokufanelekileyo. Iphasiwedi zayo zokusebenza ngokufanelekileyo. Iphasiwedi zayo zokusebenza ngokufanelekileyo
• GitHub Copilot (GHC)... Iphasiwedi zokusebenza ngokufanelekileyo. Iphasiwedi zokusebenza ngokufanelek
• I-GitHub Copilot (GHC)... Ngamanye ama-brilliant.. nganye ama-washi isikhathi sakho, ikakhulukazi ngenxa ye-response ye-verbose etholakalayo ama-off-topic
• Izixhobo ezisebenzayo ngezinye izikhathi, kodwa kuphela izimo zokusebenza zendawo, futhi akakwazi ukubona isithombe esikhulu.
• "...ukudluliselwa lomphakathi" kuyinto "ukudluliselwa okungenani kakhulu", kuyinto "ukusebenza ukujabulela", futhi njengoba kuyinto umshini eningi nge-memory yama-million lines ye-code eyenziwe, ukujabulela kunama-brilliant, izikhathi ngaphandle kwe-topic.
• .. ukuhlangabezana kakhulu ukuthi akwazi ukuthola i-sintax ye-C# ngokushesha futhi ukubuyekeza ukuthi kukhona izakhiwo / izindlela ze-C# ngokuvamile
• ... it ukunciphisa umugqa esebenzayo we-code ngoba umugqa esifanayo we-code wasemva
• I usebenzisa "ghost test" kusukela GHC kakhulu, ukubuyekeza, futhi ukuthatha imibuzo lapho ngithanda.
• Ukusebenzisa i-GHC text-prompt ikhasi ukubiza ukukhiqizwa kwe-snippets noma izici ezincinane nge-functionality esifanele.... Akukho ibhokisi ukuthi i-GHC uzama lapha, kodwa uma uzama, kungenzeka kakhulu.
• Ngiye ngicabanga izinguquko ezinkulu ezinezihlanganisa amafayela ze-3-4 ngexesha elifanayo ... Imibuzo iyona engaphakathi, ngamafutha ezininzi, njenge-C# izakhiwo kanye nezindlela ezingatholakala (ama-hallucinates?)
• ... kulula isikhathi kanye nokusetshenziswa kwe-REVIEW yonke impendulo GHC inikeza
• ... it halucinates mayelana nezakhiwo ze-C# kanye nezindlela ezisemthethweni ze-API ezaziwayo futhi ezisetyenziselwa ...
• ... inikeza inkulumo eningi kunoma ikhodi enhle, ngoba izicelo ezisekelwe ezisekelwe emakhasini ezihlangene 4 amafayela.
• Ngitholela i-GHC ukweseka imibuzo, kodwa bheka i-max 2 imibuzo ye-chat. Izimpendulo zayo zihlanganisa ukuba zihlanganisa... Uma akufanele impendulo enhle eminyakeni angu-2, ngitholela ukubuyekeza i-Google ngenxa yemibuzo efanayo.
• I-GHC inesibopho esikhulu sokuphendula; Izimpendulo ezivame ngaphandle kwegama.
• I-GitHub Copilot (GHC) akukwazi ukuhumusha umsebenzi enhle enezinhlanganisela eziningana nezifayile ngexesha elifanayo. Kulezi zimo, imiphumela ayinempumelelo futhi ayinempumelelo isikhathi esifanayo kunesilinganiselwe ku-manual programming.
• I-GitHub Copilot (GHC) ihamba ngokuvamile mayelana nezindlela ze-C# kanye nezakhiwo ezikho. I-GHC-
• I-code eyenziwe akuyona ngokushesha, kuyimfuneko iningi yokusebenza okuzenzakalelayo ukuyifaka.

5. Ukuphakama

Amaphepha angaphezulu kubonisa ukuthi ezinye izakhiwo ezihambayo mayelana nezixhobo ze-AI ezivela ngqo abathengisi zokusebenza zokusebenza zokusebenza zangaphakathi.




Okuningi abacwaningi akholelwa ukuthi kunzima ukuthi imiphumela engcono iyindlela ye-AI enikezela umkhakha wabantu isikhathi esikhulu, futhi akuyona enikezela ngqo umkhakha.

6 Izihloko




[1] I-AI ayikwazanga ukuguqulwa kwe-humane encoders yokubhalisa, abacwaningi wathi

https://arstechnica.com/ai/2025/04/researchers-find-ai-is-pretty-bad-at-debugging-but-theyre-working-on-it/

https://arstechnica.com/ai/2025/04/researchers-find-ai-is-pretty-bad-at-debugging-but-theyre-working-on-it/




[2] Imigomo ye-AI ye-coding ivimbela abathengi ku-edge

https://leaddev.com/culture/ai-coding-mandates-are-driving-developers-to-the-brink

https://leaddev.com/culture/ai-coding-mandates-are-driving-developers-to-the-brink




[3] GitHub Copilot (Gen-AI) iyasiza, kodwa akuyona enhle (March 2025)

https://markpelf.com/2717/github-copilot-gen-ai-is-helpful-but-not-great-march-2025/

https://markpelf.com/2717/github-copilot-gen-ai-is-helpful-but-not-great-march-2025/