Ngale Kwesizukulwane Sekhodi: I-AI Yokugeleza Okugcwele Kwesayensi Yedatha

ukuzwa umuzwa oqhubekayo we-AI FOMO. Nsuku zonke, ngibona abantu babelana ngamathiphu e-AI, ama-ejenti amasha namakhono abawakhile, nezinhlelo zokusebenza ezine-vibe-coded. Ngiya ngokuya ngiqaphela ukuthi ukuzivumelanisa ngokushesha ne-AI sekuba imfuneko yokuhlala unokuncintisana njengososayensi wedatha namuhla.

Kodwa angikhulumi nje kuphela ngokuxoxisana nge-ChatGPT, ukukhiqiza ikhodi nge-Cursor, noma ukupholisha umbiko no-Claude. Ukushintsha okukhulu yilokho I-AI manje ingabamba iqhaza ekuhambeni komsebenzi kwesayensi yedatha yokuphela-siya-ekupheleni.

Ukwenza umbono ubonakale, ngizame kuphrojekthi yangempela ngisebenzisa idatha yami ye-Apple Health.

Isibonelo Esilula – I-Apple Health Analysis

Umongo

Bengilokhu ngigqoke i-Apple Watch nsuku zonke kusukela ngo-2019 ukuze ngilandelele idatha yami yezempilo, njengokushaya kwenhliziyo, ukushiswa kwamandla, ikhwalithi yokulala, njll. Le datha iqukethe izimpawu zokuziphatha zeminyaka mayelana nempilo yami yansuku zonke, kodwa uhlelo lokusebenza lwe-Apple Health ngokuvamile luyibeka ngokubuka okulula kwethrendi.

Ngazama ukuhlaziya ukuthunyelwa kwe-Apple Health yeminyaka emibili eminyakeni eyisithupha edlule. Kodwa igcine isiphenduke enye yalawo maphrojekthi aseceleni ongakaze uyiqede… Umgomo wami kulokhu uwukukhipha imininingwane eyengeziwe kudatha eluhlaza ngokushesha ngosizo lwe-AI.

Okwakumele ngisebenze ngakho

Nazi izinsiza ezifanele enginazo:

1. Idatha yokuthekelisa ye-Apple Health eluhlaza: 1.85GB ku-XML, ilayishwe ku-Google Drayivu yami.
2. Isampuli yekhodi yokudlulisa ukuthunyelwa okungahluziwe kumadathasethi ahleliwe ku-repo yami ye-GitHub kusukela eminyakeni eyisithupha edlule. Kodwa ikhodi ingase iphelelwe yisikhathi.

Ukuhamba komsebenzi ngaphandle kwe-AI

Ukuhamba komsebenzi okujwayelekile ngaphandle kwe-AI kungabukeka kufana nalokhu engazama eminyakeni eyisithupha edlule: Hlola ukwakheka kwe-XML, bhala i-Python ukuze uyihlukanise kumadathasethi wendawo ahlelekile, yenza i-EDA nge-Pandas ne-Numpy, futhi ufingqe imininingwane.

Nginesiqiniseko sokuthi wonke usosayensi wedatha ujwayelene nale nqubo – akuyona isayensi yerokhethi, kodwa kuthatha isikhathi ukwakha. Ukuze uthole umbiko wemininingwane ophucuziwe, bekungathatha okungenani usuku olugcwele. Kungakho leyo repo eneminyaka engu-6 isamakwe njenge-WIP…

Ukusebenza kokuphela kwe-AI

Ukuhamba kwami ​​okubuyekeziwe nge-AI yilokhu:

1. I-AI ithola idatha engahluziwe ku-Google Drayivu yami futhi iyilande.
2. I-AI ibhekisela kwikhodi yami endala ye-GitHub futhi ibhala umbhalo we-Python ukuze uhlukanise idatha eluhlaza.
3. I-AI ilayisha amasethi edatha ahlukanisiwe ku-Google BigQuery. Yebo, ukuhlaziya kungenziwa futhi endaweni ngaphandle kwe-BigQuery, kodwa ngimise ngale ndlela ukuze ifane kangcono nendawo yokusebenza yangempela.
4. I-AI isebenzisa imibuzo ye-SQL ngokumelene ne-BigQuery ukuze iqhube ukuhlaziya futhi ihlanganise umbiko wokuhlaziya.

Empeleni, i-AI isingatha cishe zonke izinyathelo ukusuka kubunjiniyela bedatha kuya ekuhlaziyeni, mina ngisebenza kakhulu njengombuyekezi nomenzi wezinqumo.

Umbiko okhiqizwe yi-AI

Manje, ake sibone ukuthi yini i-Codex ekwazile ukuyenza ngesiqondiso sami kanye nokunye emuva naphambili emizuzwini engama-30ngaphandle kwesikhathi sokusetha imvelo nokusetshenziswa kwamathuluzi.

Ngikhethe i-Codex ngoba ngokuyinhloko ngisebenzisa i-Claude Code emsebenzini, ngakho ngangifuna ukuhlola ithuluzi elihlukile. Ngisebenzise leli thuba ukusetha indawo yami ye-Codex kusukela ekuqaleni ukuze ngikwazi ukuhlola kangcono wonke umzamo odingekayo.

Uyabona ukuthi lo mbiko uhlelwe kahle futhi uphucuziwe ngokubukeka. Ifingqe imininingwane ebalulekile kumathrendi aminyaka yonke, ukungaguquguquki kokuzivocavoca, kanye nomthelela wohambo kumazinga omsebenzi. Iphinde yanikeza izincomo kanye nemikhawulo eshiwo kanye nokucatshangwayo. Okwangihlaba umxhwele kakhulu kwakungelona nje ijubane, kodwa ukuthi umphumela waqala ngokushesha kanjani ukubukeka njengokuhlaziya okubheke ababambiqhaza esikhundleni sencwajana yokubhalela.

Sicela uqaphele ukuthi umbiko ugcwaliswe ubumfihlo bedatha yami.

Empeleni Ngenze Kanjani

Manje njengoba sesiwubonile umsebenzi ohlaba umxhwele we-AI ongawenza ngemizuzu engama-30, ake ngiwuhlukanise futhi ngikubonise zonke izinyathelo engizithathile ukuze ngenze lokho kwenzeke. Ngisebenzise i-Codex kulokhu kuhlolwa. Njengekhodi ye-Claude, ingasebenza kuhlelo lokusebenza lwedeskithophu, i-IDE, noma i-CLI.

1. Setha i-MCP

Ukuze i-Codex ikwazi ukufinyelela amathuluzi, afaka i-Google Drayivu, i-GitHub, ne-Google BigQuery, isinyathelo esilandelayo kwaba ukusetha amaseva e-Model Context Protocol (MCP).

Indlela elula yokumisa i-MCP ukucela i-Codex ikwenzele yona. Isibonelo, lapho ngiyicela ukuthi isethe i-Google Drayivu MCP, ilungise amafayela ami endawo ngokushesha ngezinyathelo ezicacile ezilandelayo zokuthi ungadala kanjani iklayenti le-OAuth ku-Google Cloud Console.

Akuphumeleli ngaso sonke isikhathi emzamweni wokuqala, kodwa ukuphikelela kuyasiza. Ngenkathi ngiyicela ukuthi isethe i-BigQuery MCP, yehlulekile okungenani izikhathi ezingu-10 ngaphambi kokuthi uxhumo luphumelele. Kodwa isikhathi ngasinye, yayinginikeza imiyalelo ecacile yokuthi ungayihlola kanjani nokuthi yiluphi ulwazi oluwusizo ekuxazululeni inkinga.

2. Yenza uhlelo nge-Plan Mode

Ngemva kokumisa ama-MCP, ngathuthela kuphrojekthi yangempela. Kuphrojekthi eyinkimbinkimbi ebandakanya imithombo yedatha/amathuluzi/imibuzo eminingi, ngivame ukuqala nge-Plan Mode ukuze ngixazulule izinyathelo zokuqalisa. Kuyo yomibili Ikhodi ye-Claude ne-Codex, ungavumela i-Plan Mode nge /plan. Isebenza kanje: uchaza umsebenzi kanye nohlelo lwakho olunzima, imodeli ibuza imibuzo ecacisayo futhi iphakamise isu lokusebenzisa elinemininingwane eyengeziwe ukuze ulibuyekeze futhi ulicwengisise. Kuzithombe-skrini ezingezansi, ungathola ukuphindaphinda kwami ​​​​kokuqala ngakho.

3. Ukwenza nokuphindaphinda

Ngemuva kokuthi ngishaye okuthi “Yebo, sebenzisa lolu hlelo”, i-Codex yaqala ukuzenzela yona, ilandela izinyathelo. Isebenze imizuzu eyi-13 futhi yakhiqiza ukuhlaziya kokuqala ngezansi. Ihambe ngokushesha kumathuluzi ahlukene, kodwa yenza ukuhlaziya endaweni njengoba ihlangabezane nezinkinga eziningi nge-BigQuery MCP. Ngemva komunye umjikelezo wokuxazulula inkinga, ikwazile ukulayisha amasethi edatha futhi yaqalisa imibuzo ku-BigQuery ngendlela efanele.

Kodwa-ke, ukuphuma kwephasi lokuqala kwakusengajulile, ngakho ngayiqondisa ukuthi ijule ngemibuzo yokulandelela. Isibonelo, nginamathikithi endiza nezinhlelo zokuhamba kusukela ohambweni oludlule ku-Google Drayivu yami. Ngiyicele ukuthi iwathole futhi ihlaziye amaphethini omsebenzi wami phakathi nohambo. Ithole ngempumelelo lawo mafayela, yakhipha izinsuku zami zokuhamba, futhi yaqhuba ukuhlaziya.

Ngemva kokuphindaphinda okumbalwa, ikwazile ukukhiqiza umbiko obanzi kakhulu, njengoba ngabelane ekuqaleni, phakathi nemizuzu engama-30. Ungathola ikhodi yayo lapha. Lokho bekungesinye sezifundo ezibaluleke kakhulu kulo msebenzi: I-AI yahamba ngokushesha, kodwa ukujula kusavela ekuphindaphindeni kanye nemibuzo engcono.

Okuthathayo Kososayensi Bedatha

Yini I-AI Ishintsha

Ngenhla isibonelo esincane sendlela engasebenzisa ngayo i-Codex ne-MCPs sebenzisa ukuhlaziya okusuka ekugcineni ukuya kwesinye ngaphandle kokubhala ngesandla umugqa owodwa wekhodi. Yiziphi izinto ezithathwa ososayensi bedatha emsebenzini?

1. Cabanga ngale kosizo lokubhala amakhodi. Esikhundleni sokusebenzisa i-AI kuphela ukubhala amakhodi nokubhala, kufanelekile ukwandisa indima yayo kuwo wonke umjikelezo wempilo yesayensi yedatha. Lapha, ngisebenzise i-AI ukuze ngithole idatha eluhlaza ku-Google Drayivu futhi ngilayishe amasethi edatha ancozululiwe ku-BigQuery. Kunamanye amacala amaningi okusebenzisa i-AI ahlobene nokusetshenziswa kwamapayipi edatha kanye nokuthunyelwa kwamamodeli.
2. Umongo uba amandla okuphindaphinda. Ama-MCP yiwona enze lokhu kuhamba komsebenzi kwaba namandla kakhulu. I-Codex iskene i-Google Drayivu yami ukuze ithole izinsuku zami zokuhamba futhi yafunda ikhodi yami endala ye-GitHub ukuze ithole ikhodi yokuhlaziya eyisampula. Ngokufanayo, ungakwazi ukunika amandla okunye kugunyazwe inkampani Ama-MCP ukusiza i-AI yakho (kanye nawe) ukuqonda kangcono umongo. Ngokwesibonelo:
   – Xhuma ku-Slack MCP kanye ne-Gmail MCP ukuze useshe izingxoxo ezibalulekile ezidlule.
   – Sebenzisa i-MCP ye-Atlassian ukuze ufinyelele imibhalo yetafula mayelana ne-Confluence.
   – Setha i-Snowflake MCP ukuze uhlole i-schema yedatha futhi uphendule imibuzo.
3. Imithetho namakhono angasetshenziswa kabusha kubalulekile. Nakuba ngingakubonisanga ngokucacile kulesi sibonelo, kufanele yenza ngokwezifiso imithetho futhi udale amakhono okuqondisa i-AI yakho futhi wandise amakhono ayo. Lezi zihloko zifanele i-athikili yazo ngokuzayo 🙂

Izovela Kanjani Iqhaza Lososayensi Bedatha

Kodwa ingabe lokhu kusho ukuthi i-AI izongena esikhundleni sososayensi bedatha? Lesi sibonelo siphinde sinikeze ukukhanya kokuthi izindima zososayensi bedatha zizojikeleza kanjani esikhathini esizayo.

1. Ukwenza kancane mathupha, ukuxazulula izinkinga ezengeziwe. Esibonelweni esingenhla, ukuhlaziywa kokuqala kwe-Codex ekhiqizwayo kwakuyisisekelo kakhulu. Izinga lokuhlaziya okukhiqizwa yi-AI lincike kakhulu kwikhwalithi yokufaka inkinga yakho. Udinga ukuchaza umbuzo ngokucacile, uwuhlukanise ube yimisebenzi engenzeka, uhlonze indlela efanele, bese ucindezela ukuhlaziya ujule.
2. Ulwazi lwesizinda lubalulekile. Ulwazi lwesizinda lusadingeka kakhulu ukuze uhumushe imiphumela ngendlela efanele futhi unikeze izincomo. Isibonelo, i-AI ibone izinga lomsebenzi wami lehle kakhulu kusukela ngo-2020. Ayikwazanga ukuthola incazelo egculisayo, kodwa yathi: “Izimbangela ezingaba khona zihlanganisa izinguquko ezivamile, isimiso somsebenzi, amashifu endlela yokuphila, ukulimala, ugqozi, noma ukuqeqeshwa okungahlelekile, kodwa lokho okucatshangwayo, hhayi okutholiwe..” Kodwa isizathu sangempela ngemuva kwakho, njengoba ungase uqaphele, ubhubhane. Ngaqala ukusebenza ekhaya ekuqaleni kuka-2020, ngakho ngokwemvelo, ngashisa amakhalori ambalwa. Lesi isibonelo esilula sokuthi kungani ulwazi lwesizinda lusabalulekile – ngisho noma i-AI ingafinyelela kuwo wonke amadokhumenti adlule enkampanini yakho, akusho ukuthi izowaqonda wonke ama-nuances ebhizinisi, futhi lokho kuyinzuzo yakho yokuncintisana.
3. Lesi sibonelo sasiqondile uma kuqhathaniswa, kodwa kusenezigaba eziningi zomsebenzi lapho ngingeke ngithembele i-AI ukuthi isebenze ngokuzimela namuhla, ikakhulukazi. amaphrojekthi adinga ukwahlulela okuqinile kwezobuchwepheshe nezibalonjengokucatshangelwa kwesizathu.

Izixwayiso ezibalulekile

Okokugcina, kunezinto ezithile okufanele uzikhumbule ngenkathi usebenzisa i-AI:

1. Ukuphepha kwedatha. Ngiqinisekile ukuthi usuzwile lokhu izikhathi eziningi, kodwa ake ngikuphinde futhi. Ingozi yokuphepha kwedatha yokusebenzisa i-AI ingokoqobo. Ngephrojekthi eseceleni yomuntu siqu, ngingakwazi ukumisa izinto ngendlela engiyifunayo futhi ngizibeke engcupheni (ngokweqiniso, ukunikeza i-AI ukufinyelela okugcwele ku-Google Drayivu kuzwakala njengomnyakazo oyingozi, ngakho lokhu kungokwezinjongo zemifanekiso). Kodwa emsebenzini, hlala ulandela isiqondiso senkampani yakho ngokuthi imaphi amathuluzi aphephile ukuwasebenzisa nokuthi kanjani. Futhi qiniseka ukuthi ufunda wonke umyalo owodwa ngaphambi kokuchofoza “gunyaza”.
2. Phinda uhlole ikhodi. Ngephrojekthi yami elula, i-AI ingabhala i-SQL enembile ngaphandle kwezinkinga. Kepha ezilungiselelweni zebhizinisi eziyinkimbinkimbi, ngisabona i-AI yenza amaphutha kukhodi yayo ngezikhathi ezithile. Kwesinye isikhathi, ihlanganisa amathebula anezimbudumbudu ezihlukene, okubangela ukuphuma ngaphandle nokubala kabili. Kwezinye izikhathi, igeja izihlungi nezimo ezibucayi.
3. I-AI ilungile, kodwa ingase ifeze ukubuza kwakho ngemiphumela emibi engalindelekile… Ake ngikutshele indaba ehlekisayo ukuze ngiphethe lesi sihloko. Namuhla ekuseni, ngivule i-laptop yami ngabona isexwayiso sokuthi akukho sitoreji sediski esisele – ngine-512GB SSD MacBook Pro, futhi benginesiqiniseko sokuthi ngisebenzise cishe uhhafu wesitoreji. Njengoba bengidlala neCodex izolo ebusuku, kube ngumsolwa wami wokuqala. Ngakho ngabuza ngempela, “hey kukhona okwenzile? 'Idatha yesistimu' yami ikhule ngo-150GB ngobusuku obubodwa”. Yaphendula, “Cha, i-Codex ithatha i-xx MB kuphela”. Ngabe sengimba amafayela ami futhi ngabona i-142GB “bigquery-mcp-wrapper.log”… Cishe, i-Codex imise lolu logi ngenkathi ixazulula inkinga yokusethwa kwe-BigQuery MCP. Kamuva emsebenzini wangempela wokuhlaziya, yaqhuma yaba ifayela elikhulu. Ngakho yebo, lo mshini wokufisa omlingo uza ngezindleko.

Lokhu okuhlangenwe nakho kwangifingqa kahle i-tradeoff: I-AI ingakwazi ukucindezela ngokuphawulekayo ibanga phakathi kwedatha eluhlaza nokuhlaziywa okuwusizo, kodwa ukuthola okuningi kuyo kusadinga ukwahlulela, ukwengamela, kanye nokuzimisela ukulungisa iphutha lokuhamba komsebenzi ngokwako.