Riba:
- Ikon yin ƙididdige hira, bincike da bayanan lura tare da hankali na wucin gadi da fitar da jigogi da alamu
- Ƙarfin tsara mutum, tafiya mai amfani da bayanan basira tare da goyan bayan basirar wucin gadi
- Kariya daga ƙirƙira da son zuciya ta hanyar haɗawa da tabbatar da bayanan da aka samar ta hanyar bayanan wucin gadi zuwa ga ɗanyen bayanai
Ana haifar da samfurori masu kyau daga ainihin buƙatar mai amfani, ba daga dandano mai zane ba. Hanyar fahimtar wannan buƙatar ita ce ta hanyar binciken mai amfani: tambayoyi, bincike, kallon filin, gwajin amfani. Amma mafi wahalar binciken ba shine tattara bayanan ba, amma yin ma'ana. Yana ɗaukar gogewa da lokaci don karanta dubun-dubatar bayanan tambayoyin da ɗaruruwan martanin bincike da fitar da tsari na gaske, buƙatu da fahimta (hankali; ma'anar "aha" a bayan bayanan). Wannan shine inda hankali na wucin gadi shine mataimaki mai ƙarfi mai ƙarfi: yana ƙididdige manyan gungu-gungu na rubutu da sauri, jigogi ƙungiyoyi, yana nuna alamu. Amma wannan iko ɗaya, idan ba a yi amfani da shi a hankali ba, yana haifar da zance mai ban tsoro, karkatacciyar girmamawa, da son zuciya. Wannan rukunin yana bayanin yadda ake amfani da AI azaman mai haɓakawa a cikin haɗin bincike da kuma dalilin da yasa yakamata ku haɗa kowane fahimta baya ga ɗanyen bayanai.
Menene haɗin bincike kuma a ina AI ya dace?
Bayanan binciken da aka gano ba su da kyau: wani mai amfani ya ce ya "neman kebul na caji kowane lokaci," wani ya ce "hasken dare yana da haske sosai," wani ya yi korafin cewa "ba zai iya samun maɓallan a cikin duhu ba." Haɗin kai shine aikin haɗa waɗannan maganganun guda ɗaya ("amfani da dare yana da matsala") da juya su zuwa fahimta ("masu amfani suna amfani da wannan samfur mafi kyau a cikin duhu, amma an tsara samfurin don amfanin rana"). A cikin hanyar gargajiya, masu zanen kaya suna yin haka tare da zane mai alaƙa (ruɗin jiki / dijital na irin wannan bayanin kula).
AI yana haɓaka matakai da yawa na wannan tsari:
- Coding: Sanya alamar jigo ga kowace sanarwa ("amfani da dare", "wahalar shigar"). AI pre-tags daruruwan jimloli a cikin mintuna.
- Jigo: Tattara tags zuwa manyan saiti. AI yana nuna alamu masu maimaitawa.
- Zane mai fa'ida: Ba da shawarar jimlolin "aha" daga jigogi. AI yana samar da zane-zane, kuna tace.
- Mutum da ƙayyadaddun tafiye-tafiye: Zayyana bayanan bayanan mai amfani na yau da kullun da matakan amfani.
Mahimman ƙa'ida: AI ta taƙaita da ƙungiyoyi, amma bayanan da ba su da tushe sun faɗi gaskiya. Kowane jigo da kowane zance ya kamata a haɗa su zuwa ainihin rubutun hirar kuma a tabbatar da su.
Haɗari biyu mafi girma: maganganun karya da son zuciya
Akwai haɗari guda biyu masu haɗari na AI a cikin haɗin bincike. Na farko an yi magana ne (hallucination): AI na iya haifar da ƙima mai amfani wanda ba ya cikin bayanan amma "sauti daidai." Wannan shine kuskure mafi haɗari a cikin bincike saboda kuna yin yanke shawara na ainihi bisa ga shaidar karya. Na biyu shi ne murdiya da son zuciya: AI na iya yin karin girman ra'ayi mafi rinjaye kuma ya shafe mahimman buƙatun ƴan tsiraru; ko stereotypes a cikin bayanan horo ("tsofaffin mutane suna tsoron fasaha") na iya shiga cikin bincikenku. Hanya daya tilo don kiyaye duka biyun: toshe kowane fitarwa zuwa cikin danyen bayanan kuma tambaya "a cikin wace zance, akan wane layi, wannan maganar ta faru?" shine yin tambaya.
Tsanaki: Kada a haɗa da kowane abin da aka ƙirƙiro na mai amfani da AI a cikin rahoto, mutum ko gabatarwa ba tare da ganowa da tabbatar da su baki ɗaya a cikin ɗan rubutun hira ba. Ƙimar apocryphal tana ba da bayanin ainihin yanke shawara.
Amintaccen kira yana gudana mataki-mataki
- Anonymize: Tsaftace bayanan sirri (suna, lamba, gano dalla-dalla) daga kwafin tambayoyin. Bayanan mai amfani bayanan sirri ne; Sirri yana da mahimmanci.
- Encode, amma nemi tushen: Tambayi AI don buƙatar ɗanyen magana da layukan layi/madogarar da ta dace da kowane tag.
- Kirga jigogi: "Masu amfani nawa ne suka ambaci wannan?" tambaya; Kada mutum ɗaya ya gabatar da ra'ayinsa a matsayin tsarin gaba ɗaya.
- Daure fahimtar baya: A ƙarƙashin kowace jimla mai fa'ida, sanya ainihin abin da aka faɗi a kai.
- Nemo misalai na gaba: Tambayi AI "akwai maganganun da suka saba wa wannan jigon, faɗi akasin haka?" tambaya; wannan yana rushe son zuciya da wuce gona da iri.
uku mini lokuta
Shari'a 1 - An kama abin ƙira. Ƙungiya ta taƙaita tattaunawa 12 tare da AI; AI ta haifar da zance mai ban sha'awa, "Daya daga cikin masu amfani ya ce, 'Zan sayi wannan samfurin a matsayin kyauta,'" kuma ƙungiyar ta so ta haɗa shi a cikin gabatarwa. Mai bincike yana duba danyen rubutun; Ba a samun irin wannan jumla a kowace zance. AI ta “yi” zance daga madaidaicin sautin gabaɗaya. An cire maganar. Darasi: Ana tabbatar da kowane zance a zahiri a cikin ɗanyen bayanai.
Hali na 2 - Goge buƙatun tsirarun. Ƙungiyar ƙira ta gudanar da nazarin kayan aikin dafa abinci tare da masu amfani da 20. Takaitaccen bayanin AI yana ba da haske game da taken "masu amfani sun damu da sauri." Amma lokacin da ƙungiyar ta koma ga ɗanyen bayanan, sun ga cewa masu amfani da 3 sun ambaci buƙatar amfani da hannu ɗaya (wanda ke da hannu mai nakasa, wanda ke ɗaukar jariri kullum); AI ta yi watsi da wannan a matsayin "yan tsiraru". Koyaya, wannan shine mahimmancin fahimtar isa ga wanda zai bambanta samfurin. Darasi: nauyin lamba ba koyaushe yana nufin mahimmanci ba; 'yan tsiraru na iya zama mafi mahimmancin fahimta.
Case 3 - Daidaitaccen amfani. Wani mai bincike ya ba da amsoshi 300 masu buɗe ido ga AI kuma ya ce "yi lakabin kowane amsa ga waɗannan jigogi, ƙidaya amsoshi nawa ne suka faɗi ga kowane jigo, kuma a faɗi 3 ainihin samfurin amsoshi a zahiri daga kowane jigo." AI yana haifar da jigogi 8, lambobi da ƙididdiga na gaske. Mai binciken yana duba samfuran ƙididdiga a cikin ɗanyen bayanan, duk gaskiya ne. Wannan ya sa tebur ya zama farkon binciken kusanci. Darasi: Tsarin gaggawar neman ƙididdigewa + ƙididdige ƙididdiga na zahiri yana haifar da abin dogaro kuma mai iya tabbatarwa.
Samfuran gaggawar da za a iya kwafi
SECURE COODING TEMPLATE "Tag waɗannan bayanan tambayoyin (ba a san su ba) cikin jigogi. Ga kowane bayani: (1) jigon da kuka sanya, (2) ainihin rubutun bayanin, (3) madogararsa [lambar hira/layi]. KADA ku yi duk wata sanarwa da ba ta cikin rubutun. Rubuta 'rashin tabbas' bayanin da ba ku da tabbas. "
THEME Counting TEMPLATE"Fitar da jigogi daga waɗannan bayanan da aka ƙulla. Ga kowane jigo: nawa masu amfani da yawa suka bayyana shi, misali 3 kalmomi na zahiri (tare da tushe).
MISALI / SON ZUCIYA BREAKING TEMPLATE "Ina so in gwada samfurin haske mai zuwa: '[bas'a]'. Yi lissafin maganganun da ke cikin bayanan da suka dace da wannan fahimta, yana nuna akasin haka, ko kuma a cikin 'yan tsiraru amma yana iya zama mahimmanci. Manufar ita ce a kama overgeneralization da son zuciya. Bayanai: [manna]."
HANKALI BACK-HADIN TEMPLATE "Rubuta wannan jimla mai fa'ida: [haske]. Nan da nan a ƙasa, sanya ainihin ƙididdiga (verbatim, tare da tushe) waɗanda ke KYAUTA wannan fahimtar. Idan babu isassun shedar kai tsaye, faɗi 'rashin wadatar shaida'. KADA KA KARA shaidar da aka yi.
Rauni mai ƙarfi / Ƙarfi mai ƙarfi
RAUNI GASKIYA: "Ka taƙaita waɗannan tattaunawa kuma ka ba ni haske."
KYAUTA MAI KARFI: "Bincika tambayoyin da ba a san su ba 12 da ke ƙasa. (1) Sanya bayanan cikin jigogi. (2) Ƙididdige yawan masu amfani da aka ambata kowane jigo. (3) Ba da 2 EXACT quotes ga kowane jigo (tare da lambar hira). (4) Tattara tsiraru amma masu yuwuwar buƙatu a ƙarƙashin wani keɓaɓɓen taken. (5) Alama ba a tantance ba tare da isassun bayanai ba. sama da duk wani bayanin da ba ya wanzu: [manna]."
Rawanin hanzari yana ba AI 'yanci don karkata da ƙirƙira; Yana sa a iya tabbatar da haɗakarwa tare da ƙidayar gaggawa mai ƙarfi, zance na zahiri, kariya ga tsiraru, da "hana ƙirƙira".
Kuskuren gama gari
- Yin amfani da ƙididdiga ta AI ba tare da duba shi a cikin ɗanyen bayanan ba. Maganar ƙira ita ce kuskuren bincike mafi haɗari.
- Tunanin cewa rinjayen lambobi shine kawai ma'aunin mahimmanci. Bukatar tsirarun galibi shine mafi mahimmancin fahimta.
- Ba da bayanan sirri ga AI ba tare da ɓoye suna ba. Bayanan mai amfani bayanan sirri ne; Asiri ya zama tilas.
- Ba neman misalan ƙiyayya ba. Haɗin da ba ya gwada akasin haka yana ƙarfafa son zuciya.
- Kuskure taƙaitaccen AI don "bincike." Maganar farko ita ce; Hankali yana dogara ne akan ɗanyen bayanai.
Tukwici: Ƙarƙashin kowane fahimta, tambayi "waɗanne zance na gaske 2-3 ke goyan bayan wannan?" Rubuta tambaya. Idan babu amsa, wannan fahimtar har yanzu hasashe ce, ba bincike ba.
A takaice
AI mai haɓakawa ne mai ƙarfi a cikin haɗin binciken mai amfani: yana ƙididdige ɗaruruwan jimloli, jigogi na rukuni, kuma yana samar da fahimta da mutane. Amma akwai manyan haɗari guda biyu: ƙirƙira ƙirƙira da murdiya / son zuciya. Hanya don kiyaye waɗannan ita ce haɗa kowane jigo da faɗar baya zuwa ga ɗanyen bayanai, ƙidaya mitar, adana ƴan tsiraru, da neman misalan misalan. Sanya bayanan mai amfani, tambayi AI don fa'ida da tushe na zahiri, sanya dokar "babu ƙirƙira", kuma kada ku ƙidaya kowane fahimta azaman bincike ba tare da goyan bayansu da shaida ba.
Aikin aikace-aikace
Samun gajerun shaidar mai amfani 8-10, na gaske ko na almara. Tun farko saka su duka. Yi jigogi masu alamar AI tare da samfurin "amintaccen coding"; sannan a nemi mitar da magana ta zahiri tare da samfurin “Theme count”. Bincika maganganun da AI ta bayar a cikin ɗanyen rubutu ɗaya bayan ɗaya kuma yi alama ko na gaske ne ko a'a. A ƙarshe, tare da samfurin “Counter-emple”, nemi maganganun da suka saba wa babban jigon ku kuma rubuta ko kun sami buƙatu marasa rinjaye.
jerin abubuwan dubawa
- [ ] Na ɓoye bayanan mai amfani kafin in ba AI.
- [ ] Na tambayi AI don maganganun magana a zahiri da nassoshi tushen kowane jigo.
- [ ] Na tabbatar da kowane zance ta hanyar nemo shi a cikin ɗanyen bayanan (tabbacin dacewa).
- [ ] Na ƙidaya yawan jigogi kuma na ajiye buƙatun ƴan tsiraru dabam.
- [ ] Na gwada son zuciya ta hanyar neman misalan misalan manyan jigogi na.
- [ ] Na sanya alamar da ba a dogara da shaida ba a matsayin "hasashen" kuma ban ƙidaya su azaman binciken ba.