Ii-AI detectors, ngamanye amaxesha zikhankanywa njenge-AI yokubhala okanye i-AI detectors, zisebenza ngenjongo yokuchonga ukuba isicatshulwa siye saqulunqwa ngokuyinxenye okanye ngokupheleleyo ngezixhobo zobukrelekrele bokwenziwa. NcokolaGPT.
Ezi zichongi ziluncedo ekuchongeni iimeko apho isiqwenga esibhaliweyo sinokudalwa yi-AI. Ukusetyenziswa kunenzuzo ngezi ndlela zilandelayo:
- Ukuqinisekisa umsebenzi wabafundi. Ootitshala banokuyisebenzisa ukuqinisekisa ubunyani bezabelo zokuqala zabafundi kunye neeprojekthi zokubhala.
- Ukuchasa uphononongo lwemveliso yobuxoki. Iimodareyitha zinokuwusebenzisa ukuchonga kunye nokujongana nophononongo lwemveliso yomgunyathi ejolise ekulawuleni imbono yabathengi.
- Ukujongana nomxholo wespam. Inceda ekubhaqeni nasekususeni iindlela ezahlukeneyo zomxholo we-spammy ezinokugqwetha umgangatho kunye nokuthembeka kwamaqonga e-intanethi.
Ezi zixhobo zisentsha kwaye ziyavavanywa, ngoko ke asiqinisekanga ngokupheleleyo ukuba zithembeke kangakanani ngoku. Kula macandelo alandelayo, siphengulula ekusebenzeni kwawo, sijonge ukuba banokuthenjwa kangakanani, kwaye sijonge uluhlu lwezicelo ezisebenzayo abazinikezelayo.
| Amaziko emfundo, kubandakanywa neeyunivesithi, akwinkqubo yokuqulunqa izikhundla zawo malunga nokusetyenziswa ngokufanelekileyo kwe-ChatGPT kunye nezixhobo ezifanayo. Kubalulekile ukubeka phambili izikhokelo zeziko lakho ngaphezu kwalo naliphi na icebiso odibana nalo kwi-intanethi. |
Zisebenza njani izixhobo ze-AI?
Ii-detectors ze-AI zihlala zisebenzisa iimodeli zolwimi ezifana nezo zikwizixhobo zokubhala ze-AI ezizama ukuzifumana. Ngokusisiseko, imodeli yolwimi ijonga igalelo kwaye ibuze, "Ngaba oku kujongeka njengento endinokuba ndenze yona?" Ukuba ithi ewe, imodeli iqikelela ukuba isicatshulwa mhlawumbi senziwe yi-AI.
Ngokukodwa, ezi modeli zikhangela iimpawu ezimbini kwisicatshulwa: "ukubhideka" kunye "nokugqabhuka." Xa le miba mibini isezantsi, kukho amathuba aphezulu okuba umbhalo uveliswe yi-AI.
Nangona kunjalo, abonisa ntoni la magama angaqhelekanga?
Ukubhideka
Ukubhideka kume njengemetric ebalulekileyo esetyenziselwe ukuvavanya ubugcisa bemizekelo yolwimi. Ibhekisa kwindlela imodeli ekwazi ngayo ukuqikelela igama elilandelayo ngokulandelelana kwamagama.
Iimodeli zolwimi ze-AI zisebenzela ekudaleni izicatshulwa ezinobunzima obuphantsi, obukhokelela ekwandeni kokuhambelana, ukuhamba kakuhle, kunye nokuqikelela. Ngokwahlukileyo koko, ukubhala kwabantu kuhlala kubonakalisa ukubhideka okuphezulu ngenxa yokusetyenziswa kolwimi oluthelekelelwayo ngakumbi, nangona lukhatshwa ziimpazamo zokuchwetheza rhoqo.
Iimodeli zolwimi zisebenza ngokuqikelela ukuba leliphi igama eliza kulandela ngokwendalo kwisivakalisi kunye nokulifaka. Unokubona umzekelo ongezantsi.
| Umzekelo wokuqhubekeka | Ukubhideka |
| Andikwazanga ukuyigqiba iprojekthi okokugqibela busuku. | Phantsi: Mhlawumbi okona kuqhubeka kunokwenzeka |
| Andikwazanga ukuyigqiba iprojekthi okokugqibela ixesha andiseli ikofu ngokuhlwa. | Phantsi ukuya phakathi: Kuncinci, kodwa yenza ingqiqo yegrama kunye nengqiqo |
| Andikwazanga ukugqiba iprojekthi kwisiqingatha sonyaka ophelileyo amaxesha amaninzi ngenxa yendlela endandingenamdla ngayo ngelo xesha. | Phakathi: Isivakalisi siyahambelana kodwa sakhiwe ngendlela engaqhelekanga kwaye sinomoya omde |
| Andikwazanga ukuyigqiba iprojekthi okokugqibela Ndiyavuya ukwazi. | Ephakamileyo: Igrama ayichanekanga kwaye ayikho ngqiqweni |
Ukudideka okuphantsi kuthathwa njengobungqina bokuba isicatshulwa senziwe nge-AI.
Ukugqabhuka
“Ukugqabhuka” yindlela yokubona ukuba izivakalisi zahluke njani kwindlela ezidityaniswa ngayo kunye nokuba zinde kangakanani. Kufana nokubhideka kodwa kwizivakalisi ezipheleleyo endaweni yamagama nje.
Xa isicatshulwa ubukhulu becala sinezivakalisi ezifanayo ngendlela esenziwe ngayo kunye nokuba zinde kangakanani, ziba nokugqabhuka okuphantsi. Oku kuthetha ukuba ifundeka ngokutyibilikayo. Kodwa ukuba isicatshulwa sinezivakalisi ezahluke kakhulu enye kwenye kwindlela ezakhiwe ngayo kunye nokuba zinde kangakanani, ziba nokugqabhuka okuphezulu. Oku kwenza ukuba isicatshulwa sizive singazinzanga kwaye sihluke kakhulu.
Isicatshulwa esenziwe yi-AI sithande ukuguquguquka kancinane kwiipateni zezivakalisi xa kuthelekiswa nesicatshulwa esibhalwe ngumntu. Njengoko iimodeli zolwimi ziqikelela igama ekusenokwenzeka ukuba lilandelayo, bakholisa ukwenza izivakalisi ezimalunga namagama ali-10 ukuya kwangama-20 ubude kwaye balandele iipateni eziqhelekileyo. Kungenxa yoko le nto ukubhala kwe-AI ngamanye amaxesha kunokubonakala kumnandi.
Ukugqabhuka okuphantsi ibonisa ukuba okubhaliweyo kunokwenzeka ukuba kuveliswe yi-AI.
Olunye ukhetho lokuqwalasela: iiWatermarks
I-OpenAI, umyili we-ChatGPT, kuthiwa uphuhlisa indlela ebizwa ngokuba “yi-watermarking.” Le nkqubo ibandakanya ukongeza uphawu olungabonakaliyo kwisicatshulwa esiveliswe sisixhobo, esinokuthi kamva sichongwe ngenye inkqubo yokuqinisekisa imvelaphi ye-AI yombhalo.
Nangona kunjalo, le nkqubo isaphuhliswa, kwaye iinkcukacha ezichanekileyo zendlela eza kusebenza ngayo azikachazwa. Ngapha koko, akucaci ukuba kukho naziphi na iiwatermark ezicetyisiweyo ziya kuhlala zinjalo xa uhlela lwenziwa kumbhalo owenziweyo.
| Ngelixa umbono wokusebenzisa le ngcamango ukufumanisa i-AI kwixa elizayo ibonakala inethemba, kubalulekile ukuqaphela ukuba iinkcukacha ezicacileyo kunye noqinisekiso malunga nokuyisebenzisa zisalindile. |
Yintoni ukuthembeka kwe-AI detectors?
- Abachongi be-AI ngokuqhelekileyo basebenza ngokufanelekileyo, ngakumbi ngeetekisi ezinde, kodwa banokuba neengxaki ukuba isicatshulwa esiyilwe yi-AI senziwe ngenjongo yokuba singalindelwanga okanye sitshintshwe emva kokuba senziwe.
- Ababoni be-AI banokuthi ngempazamo bacinge ukuba isicatshulwa esibhalwe ngabantu senziwe ngokwenene yi-AI, ngakumbi ukuba iyahlangabezana neemeko zokudideka okuphantsi kunye nokugqabhuka.
- Uphando malunga ne-AI detectors lubonisa ukuba akukho sixhobo sinokunika ukuchaneka okupheleleyo; ukuchaneka okuphezulu kwakuyi-84% kwisixhobo se-premium okanye i-68% kwisixhobo esihle kakhulu samahhala.
- Ezi zixhobo zibonelela ngengqiqo ebalulekileyo malunga nokuba kunokwenzeka ukuba isicatshulwa senziwa yi-AI, kodwa sincoma ukuba singathembeli kubo kuphela njengobungqina. Ngenkqubela phambili eqhubekayo yeemodeli zolwimi, izixhobo ezibaqaphelayo kuya kufuneka zisebenze nzima ukuze zihlale zihambelana.
- Ababoneleli abathembekileyo ngakumbi bavuma ukuba izixhobo zabo azinakusebenza njengobungqina obubambekayo bombhalo oveliswe yi-AI.
- Iidyunivesithi, okwangoku, azinathemba kakhulu kwezi zixhobo.
| Ukuzama ukufihla ukubhala okuveliswe yi-AI kunokwenza ukuba isicatshulwa sibonakale singaqhelekanga kakhulu okanye singafanelekanga ukuba sisetyenziswe. Umzekelo, ukwazisa ngabom iimpazamo zopelo okanye ukusebenzisa amagama angekho ngqiqweni kwisicatshulwa kunokunciphisa amathuba okuba ichongwe sisichongi se-AI. Nangona kunjalo, isicatshulwa esizaliswe zezi mpazamo kunye nokhetho olungaqhelekanga mhlawumbi aluyi kubonwa njengokubhala kakuhle kwezemfundo. |
Ziziphi iinjongo ezisetyenziselwa ii-AI detectors?
Ii-detectors ze-AI zenzelwe abantu abafuna ukuqinisekisa ukuba isicatshulwa sinokuba senziwe ngobukrelekrele bokwenziwa. Abantu abanokuyisebenzisa ngaba:
- Ootitshala nootitshala. Ukuqinisekisa ubunyani bomsebenzi wabafundi kunye nokuthintela ukukopa.
- Abafundi bajonga izabelo zabo. Ukujonga ukuqinisekisa ukuba umxholo wabo uyingqayizivele kwaye awubonakali ngokungalindelekanga njengombhalo owenziwe yi-AI.
- Abapapashi kunye nabahleli baphonononga izingeniso. Ufuna ukuqinisekisa ukuba bapapasha kuphela umxholo obhalwe ngabantu.
- Abaphandi. ufuna ukufumanisa nawaphi na amaphepha okanye amanqaku ophando anokuba yi-AI.
- Iibhloga kunye nababhali: Unqwenela ukupapasha umxholo oveliswe yi-AI kodwa unexhala lokuba unokuba sezantsi kwiinjini zokukhangela ukuba zamkelwa njengokubhalwa kwe-AI.
- Iingcali ekumodareyitheni umxholo. Ukuchonga ugaxekile owenziwe yi-AI, uphononongo olungeyonyani, okanye umxholo ongafanelekanga.
- Amashishini aqinisekisa umxholo wokuqala wentengiso. Ukuqinisekisa ukuba imathiriyeli yokwazisa ayiphazamiyo ngetekisi eveliswe yi-AI, igcina intembeko yophawu.
| Ngenxa yokukhathazeka malunga nokuthembeka kwabo, abasebenzisi abaninzi bayathandabuza ukuxhomekeka ngokupheleleyo kwi-AI detectors okwangoku. Nangona kunjalo, aba bachongi sele bethandwa ngakumbi njengophawu lokuba isicatshulwa sinokuba senziwe nge-AI, ngakumbi xa umsebenzisi sele ethandabuza. |
Ukufunyanwa kombhalo we-AI-Generated text
Ngaphandle kokusebenzisa i-AI detectors, unokufunda ukuchonga iimpawu ezizodwa zokubhala kwe-AI uwedwa. Akusoloko kulula ukwenza oku ngokuthembekileyo-ukubhala komntu ngamanye amaxesha kunokuvakala njengerobhothi, kwaye ukubhala kwe-AI kuya kuba ngumntu okholisayo-kodwa ngokuziqhelanisa, unokuphuhlisa ingqiqo yayo.
Imithetho ethile elandelwa ngabachongi be-AI, njengokubhida okuphantsi kunye nokugqabhuka, kunokubonakala kunzima. Nangona kunjalo, ungazama ukufumana ezi mpawu ngokwakho ngokujonga isicatshulwa kwiimpawu ezithile:
- Oku kufundeka ngemonoton, kungekho mahluko amancinci kulwakhiwo lwesivakalisi okanye ubude
- Ukusebenzisa amagama alindelekileyo kwaye angaqhelekanga kakhulu, kwaye anezinto ezimbalwa kakhulu ezingalindelekanga
Unokusebenzisa iindlela ezingenazo ii-AI detectors, ngokujonga:
| tindlela | Inkcazo |
| Ukuba nembeko ngokugqithisileyo | Ii-Chatbots ezifana ne-ChatGPT zenziwa ukuba zibe ngabancedisi abaluncedo, ngoko bahlala besebenzisa ulwimi oluhloniphekileyo nolusesikweni olusenokungavakali nje ngokungakhathali. |
| Ukungangqinelani kwelizwi | Ukuba uqhelene nendlela umntu adla ngokubhala ngayo (njengomfundi), unokuqaphela xa into ayibhalileyo yahluke kakhulu kwisitayile sabo sesiqhelo. |
| Ulwimi lokuthetha | Nika ingqalelo ekubeni akukho zimvo zininzi zinamandla nezintsha, kwaye uqaphele kwakhona ukuba kukho umkhwa wokusebenzisa amabinzana abonisa ukungaqiniseki kakhulu: “Kubalulekile ukuqaphela ukuba …” “X ithathwa ngokubanzi njenge…” “X ijongwa… ” “Abanye abantu banokuphikisa ukuba …”. |
| Amabango angekhoyo okanye akhankanywe gwenxa | Xa kuziwa ekubhaleni kwezemfundo, kubalulekile ukukhankanya apho ulufumene khona ulwazi lwakho. Nangona kunjalo, izixhobo zokubhala ze-AI zihlala zingawulandeli lo mgaqo okanye zenze iimpazamo (njengokukhankanya imithombo engekhoyo okanye engafanelekanga). |
| Iimpazamo ezisengqiqweni | Nangona ukubhala kwe-AI kuya kuba ngcono ekuvakalani kwendalo, ngamanye amaxesha iingcamango ezikuyo azihambelani kakuhle. Nika ingqalelo kwiindawo apho isicatshulwa sithetha izinto ezingahambelaniyo, ezivakala zingenakwenzeka, okanye ziveze izimvo ezingadibaniyo kakuhle. |
| Ngokubanzi, ukuzama izixhobo ezahlukeneyo zokubhala ze-AI, ukubukela iintlobo zezicatshulwa abanokuvelisa, kunye nokuqhelana nendlela ababhala ngayo kunokukunceda ube ngcono ekuboneni isicatshulwa esinokuba senziwe yi-AI. |
Abachongi bemifanekiso ye-AI kunye neevidiyo
Imifanekiso ye-AI kunye neejenereyitha zevidiyo, ngokukodwa ezidumileyo ezifana ne-DALL-E kunye ne-Synthesia, zinokwenza iziboniso ezinokwenyani kunye nezitshintshileyo. Oku kwenza ukuba kubaluleke kakhulu ukuchonga "i-deepfakes" okanye imifanekiso kunye neevidiyo ezenziwe nge-AI ukuthintela ukusasazeka kolwazi lobuxoki.
Okwangoku, imiqondiso emininzi inokutyhila imifanekiso kunye neevidiyo ezenziwe nge-AI, ezinje:
- Izandla ezineminwe emininzi kakhulu
- Iintshukumo ezingaqhelekanga
- Umbhalo ongekho ngqiqweni emfanekisweni
- Iimpawu zobuso ezingeyonyani
Nangona kunjalo, ukubona ezi mpawu kunokuba nzima njengoko i-AI isibangcono.
Kukho izixhobo eziyilelwe ukukhangela ezi ziboniso zenziwe nge-AI, kubandakanywa:
- Deepware
- Intel's FakeCatcher
- I-Illuminarty
Akukacaci ukuba ezi zixhobo zisebenza kwaye zithembeke kangakanani, ngoko ke uvavanyo oluninzi luyafuneka.
Ukuvela rhoqo komfanekiso we-AI kunye nokuveliswa kwevidiyo kunye nokufumanisa kudala isidingo esiqhubekayo sokuphuhlisa iindlela eziqinileyo nezichanekileyo zokufumanisa ukujongana nemingcipheko enokubakho ehambelana ne-deepfakes kunye ne-AI-generated visuals.
isiphelo
| Ii detectors ze-AI zinceda ekuchongeni izicatshulwa eziveliswe zizixhobo ezifana neChatGPT. Ubukhulu becala bajonga "ukubhideka" kunye "nokugqabhuka" ukubona umxholo owenziwe yi-AI. Ukuchaneka kwazo kuhlala kuyinkxalabo, kunye nezona zibalaseleyo zibonisa iimpazamo. Njengoko iteknoloji ye-AI iqhubela phambili, ukwahlula abantu kumxholo oveliswe yi-AI, kubandakanywa nemifanekiso kunye neevidiyo, kuba nzima, kugxininisa isidingo sokuhlala ulumkile kwi-intanethi. |
Imibuzo edla ngokubuzwa
| 1. Yintoni umahluko phakathi Abachongi be-AI kwaye IiCheckers zokukopa? A: Zombini i-AI detectors kunye nabahloli bokukopa bafumana ukusetyenziswa kwiidyunivesithi ukuthintela ukunganyaniseki kwezemfundo, kodwa bayahluka kwiindlela zabo kunye neenjongo: • Ii-detectors ze-AI zijolise ekuchongeni isicatshulwa esifana nesiphumo esivela kwizixhobo zokubhala ze-AI. Oku kubandakanya ukuhlalutya iimpawu zesicatshulwa ezifana nokubhideka kunye nokugqabhuka, kunokuba zithelekiswe kuluhlu lwedatha. • Abakhangeli bokukopa bajonge ukukhangela umbhalo okhutshelweyo kweminye imithombo. Bafezekisa oku ngokuthelekisa isicatshulwa kunye nedatha ebanzi yomxholo opapashwe ngaphambili kunye neethisisi zabafundi, ukuchonga ukufana-ngaphandle kokuxhomekeka ekuhlalutyeni iimpawu ezithile zesicatshulwa. 2. Ndingayisebenzisa njani i-ChatGPT? A: Ukusebenzisa iChatGPT, yenza ngokulula iakhawunti yasimahla: • Landela eli khonkco kwiwebhusayithi ye-ChatGPT. • Khetha "Bhalisa" kwaye unikeze ulwazi olufunekayo (okanye usebenzise i-akhawunti yakho kaGoogle). Ukubhalisa kunye nokusebenzisa isixhobo simahla. • Chwetheza umyalezo kwibhokisi yencoko ukuze uqalise! Inguqulo ye-iOS ye-ChatGPT app ifikeleleka ngoku, kwaye kukho izicwangciso ze-Android app kwipayipi. I-app isebenza ngokufanayo kwiwebhusayithi, kwaye ungasebenzisa i-akhawunti efanayo ukungena kuwo omabini amaqonga. 3. iChatGPT iya kuhlala isimahla kude kube nini? A: Ubukho bexesha elizayo be-ChatGPT simahla kuhlala kungaqinisekanga, kungekho xesha lithile lichaziweyo. Esi sixhobo saqala saziswa ngoNovemba ka-2022 “njengophononongo lophando” oluza kuvavanywa ngabasebenzisi abaninzi ngaphandle kweendleko. Igama elithi "preview" libonisa iintlawulo ezinokubakho kwixesha elizayo, kodwa akukho siqinisekiso sisemthethweni sokuphelisa ukufikelela simahla. Inketho eyongeziweyo, i-ChatGPT Plus, ixabisa i-20 yeedola / ngenyanga kwaye ibandakanya iimpawu eziphambili ezifana ne-GPT-4. Akukacaci ukuba le nguqulo yeprimiyamu iya kuthatha indawo yaleyo yasimahla okanye iya kuqhubeka na. Izinto ezifana neendleko zeseva zinokuchaphazela esi sigqibo. Ikhosi yexesha elizayo ayiqinisekanga. 4. Ngaba kulungile ukubandakanya i-ChatGPT kwizicatshulwa zam? A: Kwiimeko ezithile, kufanelekile ukubhekisa kwi-ChatGPT emsebenzini wakho, ngakumbi xa isebenza njengomthombo obalulekileyo wokufunda imifuziselo yolwimi lwe-AI. Iiyunivesithi ezithile zinokufuna ukucatshulwa okanye ukwamkelwa ukuba i-ChatGPT incede uphando lwakho okanye inkqubo yokubhala, njengokuphuhlisa imibuzo yophando; uyacetyiswa ukuba uqhagamshelane nezikhokelo zeziko lakho. Nangona kunjalo, ngenxa yokuthembeka okwahlukeneyo kwe-ChatGPT kunye nokunqongophala kokuthembeka njengomthombo, kungcono ukuba ungayicaphuli ngolwazi oluyinyani. KwiSimbo se-APA, ungayiphatha impendulo ye-ChatGPT njengonxibelelwano lobuqu kuba iimpendulo zayo azifumaneki kwabanye. Kwisicatshulwa, yicaphule ngolu hlobo lulandelayo: (IncokoGPT, unxibelelwano lomntu, ngoFebruwari 11, 2023). |
