U-Anthropic ufuna ukwenza kube lula ukubona imibhalo ekhiqizwe yi-AI, kodwa abanye abathuthukisi bezinhlelo sebeqalile ukwakha amathuluzi akhethekile okususa lezi zimpawu.
Le nkampani ye-AI ithe ngesonto eledlule ukuthi imibhalo ekhiqizwa amanye amamodeli e-Claude izoba “nophawu lwamanzi olungabonakali”. Njengoba lolu phawu lwamanzi luyiphethini yezibalo eyakhiwa indlela u-Claude akhetha ngayo amagama, luzohlala embhalweni ngisho noma umsebenzisi ewukopisha awunamathisele.
Abanye bakholelwa ukuthi uphawu lwamanzi lwe-AI lungakhulisa obala kokuqukethwe futhi luyisinyathelo esamukelekayo. Nokho, le ndlela iphinde yabangela ukukhathazeka kubasebenzi bemboni yezobuchwepheshe: ngisho noma abasebenzisi besebenzisa u-Claude kuphela ukuhlola umbhalo, ukuwuhumusha noma ukuwufingqa, lolu phawu lungadlulela kokuqukethwe kokugcina. Phambilini, abanye abasebenzisi batshele i-Business Insider ukuthi ngenxa yalokho bakhansele okubhalisele u-Claude.
Lokhu kuphikiswa kwenze abanye baqale ukufuna izindlela zokudlula uphawu lwamanzi. E-United States, ukuseshwa kwe-“AI watermark remover” (ithuluzi lokususa uphawu lwamanzi lwe-AI) ku-Google Trends kukhuphuke ngo-60% uma kuqhathaniswa nenyanga edlule. Abanye abathuthukisi nosomabhizinisi sebeqalile nokuzakhela amathuluzi okususa izimpawu zamanzi ze-AI emibhalweni nasemafayilini.
U-Anthropic uthe ukwethulwa kophawu lwamanzi kuhloselwe ukufeza izibopho zawo ezihlobene noMthetho we-European Union Wezobuhlakani Bokwenziwa (AI Act).
UGuillaume Meyer, usomabhizinisi wezobuchwepheshe waseParis nomsunguli wethuluzi le-AI le-e-commerce i-Memo, washicilela iphrojekthi yomthombo ovulekile ebizwa ngokuthi “Watermarks Remover” ezinsukwini ezimbalwa ngemuva kokuba u-Anthropic ememezele lolu hlelo.
UMeyer uthe leli thuluzi lisusa izinhlamvu ezifihlekile nemethadatha, bese libhala kabusha umbhalo ngaphandle kokushintsha incazelo yokuqala, ngaleyo ndlela liphazamise iphethini yezibalo yokukhethwa kwamagama engase iphathe uphawu lwamanzi.
Uveze ukuthi inguqulo yokuqala yethuluzi yamthatha cishe amahora angu-5 ukuyakha. Njengamanje le phrojekthi isithole ama-Star angaphezu kuka-1.4 万 ku-GitHub, okuyinkomba yokuthandwa kwamaphrojekthi emphakathini wabathuthukisi, efana “nokuthanda” ezinkundleni zokuxhumana. Nokho, leli thuluzi alikwazi ukuqinisekisa ukuthi uphawu lwamanzi luzosuswa ngokuphelele.
UMeyer utshele i-Business Insider wathi: “Ngiyakweseka ngokuphelele ukulebula ukuthi okuqukethwe kungokabani, kodwa ngiyaphikisana nalobu buchwepheshe bezimpawu zamanzi; kunomehluko obaluleke kakhulu phakathi kwalezi zinto ezimbili.”
UMeyer ukholelwa ukuthi inkinga ukuthi ubuchwepheshe bezimpawu zamanzi “buthatha ubunikazi bombhali njengento emnyama noma emhlophe kuphela”. Uma i-AI isetshenziselwa ukukhiqiza umbhalo ngokuphelele, uhlelo lungashiya uphawu lwamanzi, kodwa ngisho noma umsebenzisi esebenzisa i-AI ukuhlela kancane kuphela, lolu phawu lungashiyeka.
U-Anthropic wona wathi kokuthunyelwe kwebhulogi ukuthi uphawu lwawo lwamanzi aluhloselwe ukunquma ukuthi ubani umbhali wokuqukethwe.
Mayelana nokwengeza uphawu lwamanzi emibhalweni ekhiqizwe yi-AI, uMeyer uthe: “Ngicabanga ukuthi lesi yisixazululo esingalungile senkinga yangempela.”
UMeyer akayedwa ozama ukuxazulula inkinga yezimpawu zamanzi ze-AI.
Kuleli sonto, uSabrina Ramonov, ohlinzeka osomabhizinisi ngezinsizakalo zemfundo nge-AI, uthe ku-X ukuthi usungule “Watermark Remover” yamahhala esebenza esipheqululini. Leli thuluzi lithi lingakwazi “ukususa izimpawu ze-AI ezifihlekile” emibhalweni, kuma-PDF, kumadokhumenti e-Word, emakhasini ewebhu, ezithombeni nasemafayilini edatha.
Ngesikhathi esifanayo, umthuthukisi wezinhlelo ohlala e-Tokyo, u-Ansh Aneja, uthe ngosuku u-Anthropic owamemezela ngalo lesi sici wakha ithuluzi elikhethekile lokususa uphawu lwamanzi luka-Claude. Kamuva washicilela nenguqulo yomthombo ovulekile esebenza endaweni ebizwa ngokuthi i-MarkScrub.
U-Aneja wathi ku-X ngesonto eledlule leli thuluzi lakhula lisuka kubasebenzisi abangu-0 laya kubasebenzisi abangu-8500 ngosuku olulodwa. Nokho, i-Business Insider ayikwazanga ukuqinisekisa le nombolo ngokuzimela.
La maphrojekthi asesezigabeni zokuqala, futhi i-Business Insider ayikwazanga ukuhlola ngokuzimela ukusebenza kwawo kwangempela. Ezinye izinkampani, okuhlanganisa i-Google ne-OpenAI, nazo zisebenzisa izimpawu zamanzi kokuqukethwe okufana nezithombe, kanti sekukhona amathuluzi amaningi athi angazisusa lezi zimpawu.
Nokho, uThibaud Gloaguen, umcwaningi e-Safety, Security, and Intelligent Systems Laboratory e-Swiss Federal Institute of Technology Zurich, uthe la mathuluzi aphinde aveze imikhawulo osekunesikhathi ikhona kubuchwepheshe bezimpawu zamanzi.
Utshele i-Business Insider wathi: “Kuyohlale kukhona indlela yokususa uphawu lwamanzi.” Uveze njengesibonelo ukuthi abasebenzisi bangawubhala kabusha wonke umbhalo, ngaleyo ndlela baphazamise iphethini yophawu lwamanzi eyayikhona.
Amathuluzi okususa izimpawu zamanzi ze-AI okwamanje asendaweni empunga ngokomthetho. I-European Union idinga izinkampani ze-AI ukuthi zengeze izimpawu ezihlala njalo kokuqukethwe okukhiqiziwe, kodwa ayicacisanga ukuthi abantu besithathu bangazisusa yini lezi zimpawu ngokuzithandela.
Umkhulumeli we-European Commission utshele i-Business Insider ukuthi imihlahlandlela efanele ye-European Union idinga ukuthi izinhlelo zokulebula kokuqukethwe zabahlinzeki bezinsizakalo ze-AI zikwazi ukumelana nezindlela ezivamile zokuguqulwa kanye nezenzo eziphikisana nezinhlelo zokulebula.
Lo mkhulumeli uphinde wathi izindinganiso ze-European Union zokuba sobala zikubeka ngokusobala ukususa, ukukhiqiza kabusha, ukukopisha nokuguqula njengosongo olungaba khona abahlinzeki bezinsizakalo ze-AI okufanele baluhlole.
Nokho, lezi zidingo ziqondiswe kubahlinzeki bezinsizakalo ze-AI. UDmitri Roussinov, umfundisi omkhulu eMnyangweni Wezesayensi Yekhompyutha Nolwazi eNyuvesi yaseStrathclyde, uthe uMthetho Wezobuhlakani Bokwenziwa awukuvimbeli ngokucacile ukuthi abantu besithathu bazame ukususa izimpawu zamanzi ze-AI.
Kodwa utshele i-Business Insider ukuthi uma ithuluzi lokususa uphawu lwamanzi lisebenzisa i-AI ukuze liphinde likhiqize umbhalo, umhlinzeki walelo thuluzi naye angase adinge, ngaphansi komthetho we-European Union, ukwengeza uphawu olufanele kokuqukethwe kwe-AI okukhiqizwe kabusha.
UKonrad Kollnig, umsizi kaprofesa e-Law and Technology Lab eNyuvesi yaseMaastricht, uthe okwamanje akubonakali sengathi uMthetho Wezobuhlakani Bokwenziwa we-European Union noma imigomo yesevisi ka-Anthropic ikwenqabela abantu ukwakha noma ukwabelana ngamathuluzi okususa izimpawu zamanzi ze-AI.
Nokho, abasebenzisi bangase babhekane nezinkinga uma besebenzisa la mathuluzi ukuze benze okuqukethwe okukhiqizwe yi-AI kubonakale sengathi kudalwe umuntu. Inqubomgomo yokusetshenziswa ka-Anthropic iyakwenqabela ukuthi abasebenzisi bazenze abantu ngokuthi bathi imiphumela ekhiqizwe yimodeli yenziwe ngumuntu.
Njengoba u-Anthropic ehlela ukwethula i-API yokuthola imibhalo ngesikhathi esifanayo nokukhishwa kwamamodeli esizukulwane esilandelayo, le nkinga ingase igqame nakakhulu. U-Anthropic akakawumemezeli isikhathi esiqondile sokwethulwa. Uphawu lwamanzi luzosebenza kumamodeli amasha akhishwe ngo-2 Agasti nangemva kwalolo suku, kanti u-Anthropic uhlela nokusebenzisa lesi sici kancane kancane kumamodeli amadala.
UKollnig uthe: “Bonke lobu buchwepheshe bezimpawu zamanzi banenkinga eyodwa enkulu efanayo: uma ithuluzi lokuthola uphawu lwamanzi seliveziwe emphakathini, noma ubani angalisebenzisa ukuhlola okuqukethwe okukhiqizwe yi-AI, okuhlanganisa imibhalo namavidiyo, bese eqhubeka nokwakha amathuluzi okususa izimpawu zamanzi.”