I-ChatGPT inkinga evamile ye-internship yenyanga yonke esikhaleni sedatha. Inkinga ngandlela thize “yaxazululeka” kodwa anginaso isiqiniseko sokuthi isho lokho ebengicabanga ukuthi izokwenzeka. Kubasebenzi bedatha kanye ne-AI, lona manje umbuzo osebenzayo kakhulu. Amaqembu amaningi asebenzisa abaqeqeshwayo noma ama-spikes ocwaningo ukuhlola imibono: ingabe i-AI isilungile ngokwanele manje? Ingabe lawa maphrojekthi amayelana ne-artifact yokugcina kuphela?
Abafundi abaqeqeshwayo njengabahloli
Ukwakha imephu yomgwaqo yezobuchwepheshe ekuqaleni kwedatha yesigaba sokuqala akuhlukile kangako kumephu yomdlalo wevidiyo ojwayelekile:
Imephu yomgwaqo inkulu kakhulu hhayi kuphela kulokho ongakwenza, kodwa futhi nalokho ongakwenza bona. Ukube nje besingakwazi ukulunguza “emkhathini (womkhiqizo)” ngokuthumela umhloli ukuze asule imephu, sizobe sithola ukuqaphela okuthile ngalokho okulinde wena. uma usufikile (umhloli wamazwe angase afe, ngakho-ke isifaniso sihle sifinyelele iphuzu).
I-Bauplan (engayisungula ngokuhlanganyela ngo-2024) yenze ukukhetha okungajwayelekile (ngobukhulu bayo) kokuqhuba ama-internship ehlobo avela ezikhungweni eziphezulu (Columbia University, CMU, University of Wisconsin-Madison) ukuze alunguze umkhathizwe. Isebenze kahle kakhulu kuze kube manje. Ngaphandle kwefaneli yokuqasha engcono, ukuma komphakathi, kanye nethonya elithile emphakathini, ukuhlola kungene emkhiqizweni wethu, futhi kuzoba izimpahla zamasu njengoba inkampani ikhula.
Njengoba ngithumela okunikezwayo kokuqeqeshelwa umsebenzi ehlobo lika-2026, uhhafu wokuphakelayo kwami kwe-X ungitshela ukuthi ngenza konke okungalungile. Kunokuba kube yinkinga yokucatshangelwa, ezigabeni ezihlukene, osayizi kanye nezingqinamba, yonke idatha kanye namaqembu e-AI namuhla babhekene nombuzo ofanayo: ingabe ikhona manje indlela engcono yokwenza ama-spikes ocwaningo nama-ejenti? Uma kunjalo, yikuphi ukusetha okuhle, okuhloliwe kwe-AI okulula ukuzivumelanisa nakho?
Ngethemba lokuthi ulwazi lwethu nombono wethu kuzoba wusizo kubasebenzi abaningi bedatha, lokhu ukusetha kwethu kanye nezifundo esizitholile ocwaningweni lwangempela olwenziwe ngokumataniswa ne-ChatGPT.
Okokuqala, beza kuma-consultants, futhi angizange ngikhulume…
Ngesikhathi lapho i-AI isongela abasebenzi bolwazi, izikhundla zabancane zibonakala sengathi yizona eziwa kuqala. Kungani uMcKinsey kufanele aqashe umhlaziyi we-Ivy League lapho ukubhaliswa kwe-$ 200 kukhiqiza imibiko eyengeziwe, ngokushesha? Muva nje, okuphakelayo kwami kubukeka kukhombisa ukuthi i-AI kungenzeka iza ngemuva kwabacwaningi nayo, izifundiswa ezizama ukuzenza ngokuzenzakalela – “ucwaningo oluzimele ngokuphelele kusuka embonweni kuye ephepheni” – kanye noprofesa bephikisana ngokuthi bazophinde baqashe abasizi.
Kunezimpikiswano ezisobala zokumelana nomkhuba. Singakwazi ukuhlasela umphumelafuthi baphikisa ngokuthi ubuchwepheshe namanje buggy ngakho ukulingana “Ivy League” okuthenjisiwe akukhona nje. Singaphikisana ngokuthi inkontileka yomphakathi iphulwa: ngokuqinisekile, abacwaningi abasha bebelokhu be (ngandlela thile) “umthwalo”, kodwa lowo mthwalo wawuyindlela yokuwukhokhela phambili kanye nokutshalwa kwezimali esizukulwaneni esilandelayo. Singagqamisa futhi amandla umonakalo wesikhathi eside yokushintsha inqubo yokucabanga eqondwe kahle ngokugeleza komsebenzi okusha, okungahloliwe.
Yize kunesisindo ngemuva kwazo zonke lezi zimpikiswano, izimpikiswano ezifanayo zingahunyushwa ngokusobala ukuze kusungulwe izimoto noma lokho onakho. Kuhlala kunesikhathi nendawo yalezi zinkulumompikiswano, kodwa engikuthakaselayo namuhla yenziwe yasendaweni kakhulu futhi okomuntu siqu: kungazwakala kanjani uma I bengizolahla abafundi bami abafundela umsebenzi ukuze ngibhalisele u-$200?
Ngakho-ke (ngokungafani nalokhu kuhlolwa kwefiziksi engisanda kukuthola) Ngizamile ukuminyanisa inyanga yomuntu ofunda esikoleni ngempelasonto nge-ChatGPT.
Nakuba inkinga ngqo ingabalulekile kakhulu, ukuhlola i-internship kungase kube usizo ukuze uthole umuzwa we uhlobo wezinto abafundi abaqeqeshwayo abazenzayo e-Bauplan (zizwe ukhululekile ukweqa!). I-Bauplan iyinkundla yedatha ehlanganisa amagatsha: ama-agent nabantu bangavula amagatsha afana ne-Git kuwo amatafula. Ngenxa yalokho, ithebula elifanayo lingase libe nezinguqulo ezihlukene emagatsheni ahlukene. Esibonelweni sethu esikhuthazayo, Inkampani Acme Inc. ingumdayisi oku-inthanethi lapho uquqaba lwabasebenzeli bedatha lunikezwe umsebenzi wokwenza izibikezelo ezihlukene ngokuthengiswa kwakusasa:
Ngokufanelekile, umuntu angaqinisekisa umsebenzi, aqhathanise futhi aqhathanise okutholakele, bese ehlanganisa izibikezelo ithebula njenge i ukumelwa kwedatha ye-canonical. Kodwa kuthiwani uma othile ebuza umbuzo ngaphambi kokuba lokhu kwenzeke?
Amasistimu akhona angeke avele enqabe ukuphendula, ngisho noma lokhu kuzwakala kuwukumosha ngendlela eqondakalayo: ama-ejenti amabili asebenzisa ikhompuyutha imali engenayo yanyanga zonke angase angavumelani ngenani eliqondile, nokho bobabili bayavuma ukuthi imali engenayo ikhule >10% ikota ngaphezulu kwekota. Ngamanye amazwi, noma ingekho inguqulo yedatha okuvunyelwene ngayo yonkana, thina angakwazi ngisaphendula imibuzo eminingi ethokozisayo.
Umgomo wethu we-internship uwukwakha isibonelo sohlelo olunjalo. Kudinga ukufunda mayelana ne-branching, ukuthatha izibalo ezintsha, ukuklama isixazululo phezu kwe-Bauplan, nokwakha kokubili imojula yombhalo kuya ku-SQL (elula-ish) kanye nendlela yemibuzo yangokwezifiso (hard-ish).
Ukusethwa kwe-AI
I-Bauplaneers muva nje ibe nelungelo lokubona okuwukuphela kwakhe u-Wes McKinney enikeza idemo ebukhoma yokusetha kwakhe, ngakho nginqume ukukwamukela (ngamanye ama-tweaks amancane):
- I-ChatGPT 5.2 ukuhlela nokunquma ngamasu (okungukuthi indlela yokuklama ibhentshimakhi egqamisa umehluko phakathi kwezindlela zobunjiniyela);
- Ikhodi kaClaude ngaphakathi kwe-Visual Studio ukwenza iluphu yentuthuko yangempela;
- URoborev ukuze abuyekeze ukwenza okuthile ngendlela engafanele. Inikwa amandla yi-Codex, ukubuyekezwa kugqamisa izinkinga ezingaba khona futhi kusikisele ukuthuthukiswa;
- U-Roborev ubuyekeza ukuze alawule ubunkimbinkimbi bephrojekthi njalo ngezibopho eziyi-10 noma ngaphezulu: lezi zibuyekezo zithatha umbono wezokwakha futhi zisize ngokusika i-bloat.
Umcebo wangempela kwakungabangane esasibathola endleleni
Njengoba ngingakwazi ukubekezelela umbono wokungibhalela i-AI (ngoba, uma ngikhuluma iqiniso, angikwazi futhi ukubekezelela umbono wabafundi abakwenzayo), ngibhale okokugcina. mina. Njengoba ama-internship ngokuvamile ephela ngokwabelana ngemiphumela nomphakathi, ngigcine senginezinto ezanele zephepha le-ACM SAO, “Ukubuza Konke Yonke Indawo Kanye Kanye”.
Ngamamethrikhi athile, isixuku sika-X sasilungile: ngisho nokunikeza ukungafani kwekhwalithi, ngagada i-AI amahora angu-48 ukuze ngikwenze, ngithi, u-80% walokho obekungathatha amasonto. Kuyathakazelisa ukuthi ukugada ingane kungokwesimo esihlukile: i-AI ilangazelela ukujabulisa kangangokuthi ngokuvamile igcina “ikhohlisa” ukuze ithole imiphumela yezinga eliphezulu ngokusebenzisa izinqamuleli ezinamakhodi aqinile. Yize izinkinga eziningi zedatha ne-AI kulula kakhulu ukuziqinisekisa, kokuhlangenwe nakho kwethu kulula ukukudlala: lokhu kuyiqiniso ikakhulukazi noma kunini lapho ukutolika kokusetha kokuhlola kuhlukanisiwe, noma imethrikhi yokugcina ingaqondile: kufanele uhlole kathathu ukuthi ama-agent akho e-AI akhwela intaba noma ayazenzisa.
Ngakolunye uhlangothi, i-AI ayidingi ukufundiswa ngamamodeli weTarski noma ama-trust gluts, njengoba ukunamathisela amaphepha ambalwa kwanele ukushaya phansi. Imiphumela futhi “yabe ibambeka”: Nginohlelo lokusebenza lwewebhu olubukeka kahle ngaphandle kokuthi ngiphinde ngithathe i-D3.js (iminyaka engu-10 ngemva kokugcina kwami!), kanye neskripthi sedemo esilingisa amapayipi e-ejenti nemibuzo yebhizinisi phezu kwamagatsha. Uma ukholwa (njengoba ngenza) ukuthi ama-prototypes ngokuvamile ahlula i-PowerPoint (noma amaphepha), akungabazeki ukuthi isitaki se-AI esilethiwe okuthile.
Okunzima ukukubeka ngamazwi yilokho okungalethwanga, noma, ukukubeka ngokunembile, yini I ulahlekile ohlelweni. Kuyo yonke injabulo ngeshadi elimangalisayo kanye nebhentshimakhi emangalisayo, alikho elikhiqize okwengeziwe. ukuqonda. Angihlakaniphile ngokudlule kumnyakazo wocwaningo: Nginolwazi oluthe xaxa kunangaphambili (isb. indlela yokwazisa kangcono ukuhumusha okuhle kwe-SQL), kodwa amamodeli ami engqondo anesinqumo esifanayo nangesikhathi ngiqala. Ukusebenza nabaqeqeshwayo kungase kudle isikhathi futhi ngezinye izikhathi kuze kukhungathekise, kodwa njalo kuveza imicabango engcono, kubo. futhi mina ngokwami: ngokubachaza nokubafundisa, baphinde bangichazele futhi bangilule ngendlela ethile.
Uma manje ngithola imiphumela ngaphandle kokufunda okuningi, ngizizwa ngingakhululekile ikakhulukazi ngoba akucaci uma lokho kufanele kube nendaba. Angiqondile uma ibalulekile emhlabeni wonke, isikali sobuchopho obukhulu: kunjalo uma izingane zethu zingasafundi futhi ososayensi bethu bakhiphe ukucabanga engxoxweni, kubi lokho. Manje ngigxile nje ngesizotha lokhu: ingabe kunendaba kimi, enkampanini yami, kubatshalizimali bami?
Impendulo yasendaweni, yomuntu siqu – ngaphandle uma unomuzwa ophakeme kakhulu ngawe – ayicacile kahle. Ngiyazi ukubhala amakhodi futhi kungenzeka ngisakwazi ukufundisa ukucabanga okunengqondo kwezibalo, ngakho-ke akukho neyodwa kule phrojekthi ephula isisekelo esisha noma kunjalo: mhlawumbe, akukho. kuningi okumele sikufunde lapha (ngaphandle kokuba nokwenzeka kwakho konke, okuyinto engangikusola kwasekuqaleni), nokungakhululeki engikuzwayo kuyifa lengqondo yesikhathi esidlule. Nomamhlawumbe, awukho umsebenzi othobeke kakhulu ukuba ngibe inguqulo yami engcono kancane: ukwenza umsebenzi we-nitty-gritty wokuxhuma ama-API ethu eshadini, ngokwehluleka ukuhlanganisa i-DataFusion izikhathi ezingu-13, ukuya emuva naphambili mayelana nendlela yokukhetha imibuzo yebhentshimakhi ekholisayo lapho okungekho olunye uhlelo olungaveza – ingasaphathwa eyokuhlanganisa – indlela yethu yokubuza. Ngizizwa ngingakhululekile ngoba amaphrojekthi omhlaba wangempela wabantu bangempela, abangazikhukhumezi kakhulu banezinkinga ezinkulu kakhulu ezingezona ukucabanga kwezimiso zokuqala, noma imininingwane esobala yokuqaliswa.
Anginankinga namuhla (kusasa, sizobona…) ngombono olula wokuthi abantu kufanele bacabange futhi ama-LLM kufanele alungise i-syntax ye-matplotlib. Kodwa ngizabalaza nendawo enkulu empunga Iphakathinezwi elingaphakathi linyenyeza lokho ngokuphatha konke njengemininingwane yokuqaliswa, imicabango yami maduze ngeke isaba bukhali. Ingabe sesifana nalawo ma-VC “ahambisana nephethini” futhi alahlekelwe yiwo wonke ama-nuances? Ingabe iphuzu lobufakazi elifakazela inkolelo-mbono (noma ngabe ubufakazi bubukeka buhluke kangakanani), noma businika ukuqonda okuyinoveli?
Ikusasa lingalinda (kancane)
Ukubuka yami izinqumo (hhayi imizwa yami) ehlobo lika-2026 liyayidalula imiphumela yalokhu kuhlolwa. U-Bauplan uqashe abafundi ababili (abangabantu), ososayensi ababili abasebasha, abanamakhono, nabagqugquzelekile bekhompyutha abaphethe ukuhlola unqenqema lwemephu yomkhiqizo wethu ngokuphathelene nokwenza kahle kwe-AI (ukuziphendukela kwemvelo kwamakhono nge-GEPA) kanye nokukala i-git-for-data. Ngokombono ongokoqobo, ngenza isinqumo esifanayo engangingasenza ngaphambi kwalo msebenzi. Kodwa-ke, angikholwa ukuthi ngiphume ngingakashintshi futhi ngilimele kuyo: imizwa yami izoba nomthelela emicabangweni emisha ngesikhathi esithile bese iba nomthelela ezinqumweni zami.
Ngakolunye uhlangothi, njengomlandeli omkhulu we- INkosana Encaneakungilahleki ukuthi kwakuyisikhathi esichithwe kuleyo rose eyenza kwabaluleka: ukuchitha isikhathi nabafundi bami abaqeqeshwayo kuleli hlobo (ngiyakholwa) kuzobenza bona kanye nephrojekthi yethu ndawonye kubaluleke kakhulu. Ngakolunye uhlangothi, lokhu kuthwebula i-vibe yami kancane kulezi zinsuku. Kwadingeka ngimbe ku-Internet Archive ukuze ngithole okuthile engisanda kukhumbula ngakho kusukela ngo-2006 (izibalo zezibalo akuyona ukuphela kwento engiyikhumbula kusukela eminyakeni yami yama-20s, ngokusobala). Lokhu ukungena okungu-#1 ku-Blender's “50 Okubi kakhulu Izinto Ezizokwenzeka Emculweni”:
#01. IZINGANE NAMUHLA
Emuva osukwini lwethu, besingenawo ama-iPod amnandi namaringithoni nokulandiwe. Besingenakho ukunethezeka nokunethezeka kwama-scrotum-rings kanye namalogi akho e-World Wide Web. Ngenkathi sifuna ukuntshontsha i-albhamu entsha ye-URIAH HEEP, asikwazanga ukuvele sizitholele i-inthanethi, bekufanele sikwenze ngendlela yesidala—ngokuqwala intaba siye esitolo (ukhuphuke umqansi, zombili izindlela) nokushova u-12″ wama-vinyl ngaphansi kwamajezi ethu (okwakumele siziluke). Kungakho nina baculi be-whipper-snappers ningakuboni ukubaluleka kwangempela komculo. Noma u-Uriya Heep. Manje susa isihogo otshanini bethu!
Ingabe sisazokwazisa “inani langempela lezinto” uma manje “sesingazeba” ngokunethezeka kwamalaptop ethu?
Sanibonani, bafana bama-cowboy
Sibonga uLuca, uColin, u-Ethan ngokuphawula kwabo ngohlaka lwangaphambilini lwalesi sihloko.
Uma ufuna ukuba yi-Bauplan intern futhi wenze izinto ezinhle zedatha ne-AI (njengalokhu noma lokhu noma lokhu), ngisawamukela amakhandidethi angabantu: thintana!
