Àte ŋu atrɔ asi le míaƒe ahɔhɔ̃mekawo ƒe dɔwɔwɔ ŋua?

Afɔku si le dzidzim ɖe edzi le Neural Network Reverse Engineering — Kple Nusi Wòfia Na Wò Dɔwɔƒe

Le ƒe 2024 me la, numekula siwo le yunivɛsiti gã aɖe ɖee fia be yewoate ŋu agbugbɔ gbegbɔgblɔ gã ƒe kpɔɖeŋu si nye ame ŋutɔ tɔ ƒe xɔtuɖaŋu ememetɔ awɔ to naneke wu eƒe API ŋuɖoɖowo kple akɔntabubu si ƒe home ade dɔlar 2,000 zazã o. Dodokpɔa na ƒutsotsoe siwo wɔ nuku ŋutɔ to AI dɔwɔƒea, gake nusiwo wòfia la de ŋgɔ sasasã wu Silicon Valley. Asitsaha ɖesiaɖe si zãa mɔ̃wo ƒe nusɔsrɔ̃ ƒe kpɔɖeŋuwo — tso ametafatafa didi ƒe ɖoɖowo dzi va ɖo asisiwo ƒe kafukafumɔ̃wo dzi — dze ŋgɔ nyabiase si me mekɔ o fifia: ɖe ame aɖe ate ŋu afi nunya si nèzã ɣleti geɖe tsɔ tu? Neural network reverse engineering meganye nukpɔsusu ƒe afɔku o. Enye amedzidzedze ƒe mɔ̃ si ŋudɔ woate ŋu awɔ, si ŋu woate ŋu aɖo bɔbɔe wu, si gɔme wòle be habɔbɔ ɖesiaɖe si mɔ̃ɖaŋununya ʋãna nase egɔme.

Alesi Neural Network Reverse Engineering Le Le nyateƒe me

Reverse engineering a neural network mehiã be woakpɔ mɔ ŋutɔŋutɔ na server si le ewɔm o. Zi geɖe la, amedzidzelawo zãa mɔnu aɖe si woyɔna be model extraction, afisi wobiaa model ƒe API ɖe ɖoɖo nu kple nusiwo wotsɔ de eme siwo wowɔ nyuie, emegbe wozãa nusiwo woɖe tso eme tsɔ naa hehe kɔpi si sɔ kloe. Numekuku aɖe si wowɔ le ƒe 2023 me si wota ɖe USENIX Security me ɖee fia be amedzidzelawo ateŋu agbugbɔ nyametsotso ƒe liƒo siwo le asitsa nɔnɔmetatawo ƒe hatsotsowo me kple nuteƒewɔwɔ si wu 95% to nyabiase siwo mede 100,000 o zazã me — ɖoɖo si mexɔa dɔlar alafa ʋɛ aɖewo le API fewo me o.

Le eɖeɖe godo la, model inversion attacks li, siwo wɔa dɔ le mɔ si to vovo na ema nu. Le esi teƒe be amedzidzelawo nawɔ kpɔɖeŋua ƒe kɔpi la, wogbugbɔa hehenana ŋuti nyatakakawo ŋutɔ ɖoa ​​anyi. Ne wona hehe wò ahɔhɔ̃mekawo ƒe kadodo le asisiwo ƒe nuŋlɔɖiwo, asixɔxɔ ƒe mɔnu siwo nye ame ŋutɔ tɔ, alo asitsatsa ƒe xexlẽdzesi ememetɔwo ŋu la, menye ɖeko tɔtrɔ ɖe megbe ƒe amedzidzedze dzidzedzetɔe fi wò kpɔɖeŋua ko o — eɖea nyatakaka vevi siwo woɖa ɖe eƒe kpekpemewo me la ɖe go. Ha etɔ̃lia, hameviwo ƒe nutsotso ƒe amedzidzedze, ɖea mɔ na futɔwo be woakpɔe ɖa be nyatakaka teƒe aɖe koŋ nye hehenana ƒe ɖoɖoa ƒe akpa aɖe hã, si na ameŋunyatakakawo ŋuti dzitsitsi vevie le sededewo abe GDPR kple CCPA ene te.

Nya si bɔ enye be "aɖaka yibɔ" ƒe susu — susu si nye be kpɔɖeŋu aɖe ɖoɖo ɖe API megbe nana wònɔa dedie — la gblẽ le gɔmedzedzea me. Nyagblɔɖi ɖesiaɖe si wò kpɔɖeŋua atrɔ la nye nyatakaka teƒe si amedzidzela ateŋu azã ɖe ŋuwò.

Nusita Wòle Be Asitsahawo Natsɔ Ðe le Wo Ŋu Wu Wu Ale Si Wolé Be Na Fifia

Habɔbɔ akpa gãtɔ tsɔa woƒe kɔmpiuta dzi dedienɔnɔ ƒe gazazãwo ɖoa network ƒe golɔgoewo, nuwuƒe takpɔkpɔ, kple nyatakakawo tsɔtsɔ ɣla ŋu. Gake nunya ƒe nunɔamesi siwo wotsɔ de ahɔhɔ̃mekawo ƒe kadodo si wona hehee me ate ŋu atsi tre ɖi na Numekuku kple Dɔwɔnawo ƒe ɣleti geɖe kple ŋgɔyiyi ƒe gazazã miliɔn geɖe. Ne hoʋlila alo fefewɔla vɔ̃ɖi aɖe ɖe wò kpɔɖeŋua ɖa la, wokpɔa asixɔxɔ si katã le wò numekukua ŋu gazazã aɖeke manɔmee. Le IBM ƒe ƒe 2024 ƒe Cost of a Data Breach nyatakaka nu la, sedzidada si ku ɖe AI ɖoɖowo ŋu le mama dedie nu gblẽ ga dɔlar miliɔn 5.2 na habɔbɔwo — 13% sɔ gbɔ wu sedzidada siwo me AI ƒe nunɔamesiwo mele o.

Afɔkua nu sẽ ŋutɔ na dɔwɔƒe suewo kple gãwo. Dɔwɔƒewo ate ŋu aƒle ML dedienɔnɔ ƒe ƒuƒoƒo siwo tsɔ wo ɖokui na kple xɔtuɖoɖo siwo wowɔ ɖe ɖoɖo nu. Gake SMB siwo le dzidzim ɖe edzi siwo tsɔa mɔ̃ɖaŋununya sɔsrɔ̃ dea woƒe dɔwɔnawo me — eɖanye na ŋgɔyiyi ƒe dzesidede, didi ƒe nyagblɔɖi, alo asisiwo ƒe kpekpeɖeŋu le wo ɖokui si o — zi geɖe la, wozãa kpɔɖeŋu siwo me dedienɔnɔ ƒe sesẽme suetɔ kekeake le. Woɖoa ŋu ɖe ame etɔ̃lia ƒe mɔ̃ siwo ate ŋu awɔ ametakpɔnu nyuiwo ŋudɔ alo womawɔe o ŋu.

Nukpɔsusu si me afɔku le wu le AI ƒe dedienɔnɔ me enye be nusiwo sesẽ sɔ kple ametakpɔkpɔ. Neural network si me parameter miliɔn 100 le mele dedie le dzɔdzɔme nu wu esi si miliɔn 1 o — nusi le vevie enye alesi nèkpɔa mɔɖeɖe ɖe eƒe inputs kple outputs ŋu dzi.

Ametakpɔnu Nyui Atɔ̃ Siwo Tsi Tre Ðe Kpɔɖeŋu Fififi Ŋu

Wò ahɔhɔ̃mekawo takpɔkpɔ mehiã be nàxɔ PhD le mɔ̃ɖaŋunusɔsrɔ̃ si tsi tre ɖe wo nɔewo ŋu me o, gake ebia be nàɖoe koŋ awɔ xɔtuɖaŋu ŋuti nyametsotsowo. Mɔnu siwo gbɔna la tsi tre ɖi na fifi nuwɔna nyuitɔ siwo habɔbɔwo abe NIST kple OWASP kafu hena ML ƒe kpɔɖeŋu siwo wozã la ƒe dedienɔnɔ.

ƒe nyawo
• Seɖoƒeɖoɖo kple biabia ƒe gazazã ƒe ɖoɖo: Se API yɔyɔ ƒe xexlẽme si zãla alo safui ɖeka ɖesiaɖe ateŋu awɔ le ɣeyiɣi ƒe fesre si wona me. Model extraction attacks bia biabia akpe ewo — aggressive rate limiting na be extraction gãwo megawɔa dɔ ne womefɔ ɣlidodowo o.
• Dɔwɔwɔ ƒe tɔtɔ: Tsɔ toɣliɖeɖe si dzi wokpɔna kpe ɖe kpɔɖeŋu ƒe nyagblɔɖiwo ŋu. Le esi teƒe be nàtrɔ kakaɖedzi ƒe dzesi siwo sɔ pɛpɛpɛ (e.g., 0.9237), ƒo xlã dometsotso siwo le gbadzaa wu (e.g., 0.92). Esia kpɔa zazã ta esime wòle biabia siwo amedzidzela hiã be wòagbugbɔ wò kpɔɖeŋua atu la dzim ɖe edzi ŋutɔ.
• Tsidzesi: De asidede agbalẽ te siwo womate ŋu ade dzesii o wò kpɔɖeŋu ƒe nuwɔna me — eve tɔxɛ siwo wotsɔ de eme-do goe siwo wɔa dɔ abe asibidɛ ƒe dzesi ene. Ne wò kpɔɖeŋu ƒe kɔpi si wofi do la, tsidzesiwo naa ʋɔnudrɔ̃ƒe ƒe kpeɖodzi be wofi fi.
• Adzamenyawo ƒe vovototodedeameme le hehexɔxɔ me: Do akɔntabubu ƒe toɣliɖeɖe de eme le hehexɔxɔa ŋutɔ me. Esia ate ŋu aɖo seɖoƒe na nyatakaka agbɔsɔsɔme si ku ɖe hehenana ƒe kpɔɖeŋu ɖekaɖeka ɖesiaɖe ŋu si dona to kpɔɖeŋua ƒe nyagblɔɖiwo me, si ʋlia ame ta tso inversion kple membership inference attacks siaa me.
• Ŋkuléle ɖe nu ŋu kple nusiwo mesɔ o didi: Kpɔ API zazã ƒe ɖoɖowo dzi hena dzesi siwo fia be wole numekuku wɔm ɖe ɖoɖo nu. Amedzidzedze siwo woɖena ɖa la wɔa biabia mama tɔxɛ siwo meɖi naneke abe zãla ƒe ʋuʋu le se nu ene o — nuxlɔ̃ame siwo wowɔna le wo ɖokui si ateŋu ade dzesi nuwɔna si ŋu ɖikeke le hafi amedzidzedze nakpɔ dzidzedze.

Afɔɖeɖe siawo dometɔ eve alo etɔ̃ gɔ̃ hã zazã naa amedzidzedze ƒe gazazã kple sesẽme dzina ɖe edzi zi gbɔ zi geɖe. Taɖodzinua menye dedienɔnɔ deblibo o — ele ewɔm be extraction nazu nusi me susu mele o le ganyawo gome ne wotsɔe sɔ kple model tutu tso gɔmedzedzea me.

Akpa si Dɔwɔnawo ƒe Dɔwɔƒewo Wɔna le AI Dedienɔnɔ Me

Akpa ɖeka si dzi woŋea aɖaba ƒuna le dzeɖoɖo siwo ku ɖe kpɔɖeŋu ƒe dedienɔnɔ ŋu me enye dɔwɔwɔ ƒe nɔnɔme si keke ta wu. Neural network meli le eɖokui si o — edoa ka kple nyatakakadzraɖoƒewo, CRM ɖoɖowo, fexexe ƒe mɔnuwo, dɔwɔlawo ƒe nuŋlɔɖiwo, kple asisiwo ƒe kadodomɔnu. Amedzidzela si mateŋu atrɔ asi le wò kpɔɖeŋua ŋu tẽ o ateŋu aɖo taɖodzinu na nyatakaka pɔmpi siwo le nuɖuɖu nam, API siwo le eƒe emetsonuwo ɖum, alo asitsaɖoɖo siwo dzraa eƒe nyagblɔɖiwo ɖo la boŋ.

Afi siae dɔwɔwɔ ƒe mɔnu ɖeka ƒe amesinɔnɔ va zua dedienɔnɔ ƒe viɖe vavãtɔ tsɔ wu be wòanye nusi sɔ ko. Ne asitsahabɔbɔwo tsɔ SaaS dɔwɔnu gbogbo aɖewo siwo ƒe kadodo metso o la ƒo ƒui la, teƒe ɖesiaɖe si wowɔa ɖeka le va zua afisi woate ŋu adze wo dzi le. Mewayz kpɔa esia gbɔ to asitsatsa ƒe modules 207 ƒoƒo ƒu — tso CRM kple fexexe dzi va ɖo HR kple numekuku dzi — ɖe mɔnu ɖeka si me wokpɔa mɔɖeɖe ɖe nu ŋu le teƒe ɖeka kple agbalẽdzikpɔkpɔ ƒe nuŋlɔɖiwo le. Le esi teƒe be woakpɔ dɔwɔnu vovovo wuiatɔ̃ ta kple mɔɖeɖe ƒe kpɔɖeŋu vovovo wuiatɔ̃ la, ƒuƒoƒowo kpɔa nusianu dzi tso dashboard ɖeka dzi.

Le habɔbɔ siwo le AI ŋutetewo zãm gome la, ƒoƒo ɖekae sia fia be nyatakakawo ƒe asiɖeɖe le ɖoɖowo dome ʋɛ aɖewo, API safui ʋɛ aɖewo siwo le tsia dzi le ɖoɖowɔɖi ƒe faɛlwo me, kple sedziwɔwɔ ƒe teƒe ɖeka na mɔɖeɖe ƒe ɖoɖowo. Ne wò asisiwo ƒe nyatakakawo, dɔwɔwɔ ƒe xexlẽdzesiwo, kple asitsatsa ƒe susuŋudɔwɔwɔ katã le nɔnɔme ɖeka si dzi woɖu me la, amedzidzedze ƒe anyigba na nyatakakawo ɖeɖeɖa — nusi nye kpɔɖeŋu ƒe tɔtrɔ ƒe amedzidzedzewo ƒe nu gbãtɔ — dzi ɖena kpɔtɔna ŋutɔ.

Xexeme Ŋutɔŋutɔ ƒe Nudzɔdzɔ Siwo Trɔ Dzeɖoɖoa

Le ƒe 2022 me la, fintech gɔmedzedze aɖe kpɔe be hoʋlila aɖe do gaxɔgbalẽvi si sɔ kloe ɖe ŋgɔ le gɔmedzedzea ŋutɔ ƒe gɔmedzedze megbe ɣleti enyi pɛ ko. Ememe numekuku ɖee fia be hoʋlila la nɔ nya biam le ɖoɖo nu tso gɔmedzedzea ƒe dzeside API ŋu ɣleti geɖe, eye wòzã ŋuɖoɖoawo tsɔ na hehe kpɔɖeŋu si wogbugbɔ wɔ. Gɔmedzedzea meɖo seɖoƒe aɖeke na ga home aɖeke o, etrɔa nusiwo ate ŋu adzɔ ƒe mama blibowo vɛ, eye meléa nyabiasewo ŋuti nuŋlɔɖi aɖeke me ɖe asi si ate ŋu ado alɔ senyawo gbɔ kpɔkpɔ o. Hoʋlila la medze ŋgɔ emetsonu aɖeke o.

Nyitsɔ laa, le ƒe 2024 ƒe nuwuwu la, dedienɔnɔ ŋuti numekulawo ɖe mɔnu aɖe si woyɔna be "side-channel model extraction" fia si zã ɣeyiɣi ƒe vovototo le API ƒe ŋuɖoɖowo me — ɣeyiɣi didi si dɔwɔƒea xɔ hafi trɔ emetsonuwo na nusiwo wotsɔ de eme vovovowo — tsɔ ƒo nya ta le kpɔɖeŋua ƒe ememe wɔwɔme ŋu evɔ womeku nyagblɔɖiawo ŋutɔ gɔ̃ hã me o. Amedzidzedzea wɔ dɔ ɖe kpɔɖeŋu siwo woɖo ɖe alilikpo me dɔwɔƒe gã etɔ̃awo katã dzi eye mehiã be woakpɔ mɔ tɔxɛ aɖeke wu API safui si wozãna ɖaa o.

Nudzɔdzɔ siawo te gbe ɖe nya vevi aɖe dzi: ŋɔdzia le tɔtrɔm kabakaba wu habɔbɔ akpa gãtɔ ƒe ametakpɔnuwo. Mɔnu siwo wobu be wonye numekuku deŋgɔwo ƒe etɔ̃ enye sia la li fifia abe dɔwɔnu siwo woate ŋu azã le mɔ gbadza nu ene le GitHub. Asitsaha siwo bua kpɔɖeŋu dedienɔnɔ be enye nusi ŋu woatsi dzi ɖo le etsɔme la le megbe xoxo.

AI Dekɔnu si Wotsɔ Dedienɔnɔ Gbãtɔ Tutu

Mɔ̃ɖaŋununya ɖeɖe mekpɔa kuxi sia gbɔ o. Ele be habɔbɔwo natu dekɔnu si me woawɔ nu ɖe ​​AI ƒe nunɔamesiwo ŋu vevie abe alesi wowɔa nu ɖe ​​dzɔtsoƒe ƒe nyatakakawo, asitsatsa ŋuti nya ɣaɣlawo, kple asisiwo ƒe nyatakakadzraɖoƒewo ŋu ene. Esia dzea egɔme tso nudzraɖoƒewo dzi — dɔwɔƒe geɖewo meléa xexlẽdzesi blibo gɔ̃ hã ɖe asi si fia be mɔ̃ siwo woɖo, afisi woate ŋu akpɔ wo le, kple amesiwo si API le o. Màte ŋu akpɔ nusi mènya be ele o ta o.

Nuwɔwɔ aduadu le dɔwɔwɔ vovovowo me le vevie ŋutɔ. Ele be nyatakakaŋutinunyalawo nase tsitretsiɖeŋu ƒe ŋɔdzidonamewo gɔme. Ele be dedienɔnɔ ƒe ƒuƒoƒowo nase alesi mɔ̃ siwo wotsɔ srɔ̃a nu ƒe mɔ̃wo wɔa dɔe gɔme. Ele be adzɔnuwo dzikpɔlawo nawɔ nyametsotso siwo ŋu wonya nu tsoe le nyatakaka siwo ƒe kpɔɖeŋu APIwo ɖena fiana ŋu. "Dzĩ ƒuƒoƒo" kamedede edziedzi — afisi ememe ƒuƒoƒowo dzea agbagba be yewoaɖe alo atrɔ wò ŋutɔ wò kpɔɖeŋuwo — ɖea afɔkuwo fiana hafi gotagome amedzidzelawo nawɔe. Dɔwɔƒewo abe Google kple Microsoft ene wɔa kamedede siawo ƒe etɔ̃ ɖesiaɖe; susu aɖeke meli si ta habɔbɔ suewo mateŋu axɔ tɔtrɔ bɔbɔewo o.

Nyatakakadzraɖoƒewo abe Mewayz siwo tsɔa dɔwɔwɔ ŋuti nyatakakawo vaa xɔ ɖeka te hã nana wònɔa bɔbɔe be woawɔ nyatakakawo dzikpɔkpɔ ƒe ɖoɖo siwo kpɔa ŋusẽ ɖe AI ƒe dedienɔnɔ dzi tẽ. Ne ète ŋu lé ŋku ɖe amesiwo ge ɖe asisiwo ƒe akpa kawo me, ɣekaɣie wowɔ numekuku ŋuti nyatakakawo, kple alesi nyatakakawo sisina le modules dome ŋu la, ètu ŋkuléleɖenuŋu ƒomevi si nana nyatakakawo ɖeɖe si ŋu womeɖe mɔ ɖo o kple kpɔɖeŋu fifi siaa sesẽna ŋutɔ be woawɔ wo womede dzesii o.

Nusi Akplɔe ɖo: Ðoɖowɔwɔ, Dzidzenuwo, kple Dzadzraɖo

Sedziwɔwɔ ƒe nɔnɔme le edzi yim. EU AI Se, si ge ɖe edziwɔwɔ me le afɔɖeɖewo me tso ƒe 2025 me la, lɔ ɖoɖo siwo ƒo xlã kpɔɖeŋu ƒe nuwɔwɔ le gaglãgbe kple dedienɔnɔ si abia tso habɔbɔwo si be woaɖee afia be yewoɖe afɔ siwo sɔ be yewoakpɔ AI ɖoɖowo ta tso asitɔtrɔ kple fififi me. Le United States la, NIST ƒe AI Afɔku Dzikpɔkpɔ ƒe Ðoɖowɔha (AI RMF) ƒo nu tso kpɔɖeŋuwo ɖeɖeɖa ŋu tẽ fifia abe ŋɔdzinu ƒe hatsotso ene. Asitsaha siwo xɔa ɖoɖo siawo do ŋgɔ la akpɔe be sedziwɔwɔ anɔ bɔbɔe wu — eye woanɔ nɔnɔme nyuitɔ me be woaʋli woƒe AI gadedewo ta.

Nya la le tẽ: neural network reverse engineering menye nususugblɔ ƒe ŋɔdzinu si wodzra ɖo ɖi na dukɔ-dziɖuɖu ƒe dɔwɔlawo o. Enye aɖaŋu si ŋu woate ŋu aɖo, si ŋu woŋlɔ nu tsoe nyuie, si hoʋlila alo nuwɔla vɔ̃ɖi ɖesiaɖe si ʋãe ate ŋu awɔ ɖe ɖoɖo siwo womeʋli ta nyuie o ŋu. Asitsaha siwo le ŋgɔ yim le AI ƒe ɣeyiɣia me la manye esiwo tua kpɔɖeŋu nyuitɔwo ko o — woawoe anye esiwo akpɔ wo ta. Dze egɔme kple mɔɖeɖe ɖe nu ŋu, nusiwo dona ƒe tɔtɔ, kple zazã dzi kpɔkpɔ. Tu ɖe dɔwɔwɔ ƒe gɔmeɖoanyi ɖeka si ɖea nyatakakawo ƒe kaka dzi kpɔtɔna dzi. Eye bu wò kpɔɖeŋu siwo wona hehee la abe nunɔamesi siwo ƒe asixɔxɔ de ŋgɔ ene, elabena wò hoʋlilawo awɔe godoo.

Nyabiase Siwo Wobiana Enuenu

Nukae nye ahɔhɔ̃mekawo ƒe dɔwɔwɔ ƒe tɔtrɔ?

Neural network reverse engineering nye mɔ̃ si srɔ̃a nu ƒe kpɔɖeŋu ƒe emetsonuwo, API ƒe ŋuɖoɖowo, alo nuwɔna ƒe nɔnɔmewo me dzodzro be woatsɔ agbugbɔ eƒe ememe xɔtuɖaŋu, kpekpemewo, alo hehenana ŋuti nyatakakawo atu. Amedzidzelawo ate ŋu azã mɔnuwo abe model extraction, membership inference, kple adversarial probing atsɔ afi algorithms siwo nye wo tɔ. Le asitsaha siwo ɖoa ŋu ɖe dɔwɔnu siwo AI ʋãna ŋu gome la, esia hea nunya ƒe nunɔamesiwo kple hoʋiʋli ƒe afɔku gãwo vɛ si bia dedienɔnɔ ƒe afɔɖeɖe siwo woawɔ do ŋgɔ.

Aleke asitsalawo ateŋu akpɔ woƒe AI ƒe kpɔɖeŋuwo ta be woagatrɔ asi le wo ŋu o?

Ametakpɔnu veviwo dometɔ aɖewoe nye API biabia siwo ɖoa seɖoƒe na agbɔsɔsɔ, toɣliɖeɖe si dzi wokpɔna tsɔtsɔ kpe ɖe kpɔɖeŋu ƒe emetsonuwo ŋu, ŋkuléle ɖe mɔɖeɖe ƒe ɖoɖo siwo ŋu ɖikeke le ŋu, kple adzamenyawo zazã le hehenana me. Mɔ̃wo abe Mewayz, si nye 207-module business OS, kpena ɖe dɔwɔƒewo ŋu be woaɖo dɔwɔwɔwo ɖe teƒe ɖeka eye woɖea nusiwo woɖena fiana dzi kpɔtɔna to AI dɔwɔwɔ ƒe ɖoɖo veviwo léle ɖe nɔnɔme si le dedie, si wɔ ɖeka me tsɔ wu be woakaka ɖe ame etɔ̃lia ƒe ƒoƒo ɖekae siwo ŋu woate ŋu adze afɔku le me.

Ðe asitsaha suewo le afɔku me be woafi AI ƒe kpɔɖeŋua?

Kakaɖedzitɔe. Numekulawo ɖee fia be model extraction attacks siwo ƒe asi mede dɔlar 2,000 o le kɔmpiuta dzi, si wɔe be amesiame kloe ate ŋu akpɔ wo. Asitsaha sue siwo zãa kafukafumɔ̃ siwo wowɔ ɖe ɖoɖo nu, asiɖoɖoanyi ƒe mɔnuwo, alo ametafatafa didi ƒe mɔnuwo nyea taɖodzinu siwo hea amewo le esi tututu zi geɖe la, dedienɔnɔ si sɔ na dɔwɔƒewo mele wo si o ta. Mɔ̃ siwo ƒe asi bɔbɔ abe Mewayz ene, siwo dzea egɔme tso $19/mo dzi le app.mewayz.com, kpena ɖe ƒuƒoƒo suewo ŋu be woawɔ dɔwɔwɔ ƒe dedienɔnɔ sesẽ wu.

Nukae wòle be mawɔ ne mesusu be wogblẽ nye AI ƒe kpɔɖeŋua me?

Dze egɔme kple API ƒe mɔɖeɖe ƒe nuŋlɔɖiwo me dzodzro hena biabia ƒe agbɔsɔsɔme siwo mebɔ o alo ɖoɖowɔɖi ƒe nyawo tsɔtsɔ de eme ƒe ɖoɖo siwo do susu ɖa be wodze agbagba be woaɖee ɖa. Trɔ API safuiwo enumake eye nàwɔ seɖoƒe sesẽwo ɖe agbɔsɔsɔme ŋu. Kpɔe ɖa be kpɔɖeŋu ƒe emetsonuwo dze le hoʋlilawo ƒe adzɔnuwo me hã. Bu tsidede dzesi na etsɔme kpɔɖeŋu ƒe tɔtrɔwo ŋu be nàdi zazã si ŋu womeɖe mɔ ɖo o, eye nàbia kɔmpiuta dzi dedienɔnɔ ŋuti nunyala bibi aɖe be wòada alesi gbegbe sedzidadaa keke tae bliboe eye wòasesẽ wò ametakpɔnuwo.