Ukuvumelanisa (Alignment)
Alignment
Inqubo yokuqinisekisa ukuthi izinhloso, imiphumela, kanye nokuziphatha kohlelo lwe-AI kuvumelana nezinjongo nezindinganiso zabantu. Lokhu kubaluleke kakhulu ezinhlelweni ezithuthukile ezingase zithuthukise ukuziphatha okungahlosiwe ngokuqondile.
Isibonelo: Ukuqinisekisa ukuthi i-chatbot yezempilo yengqondo ayinikezi izenzo eziyingozi noma ngabe yimiphi imiyalezo.
Isikhombimsebenzisi sohlelo lokusebenza (API) (Application Programming Interface (API))
Application Programming Interface (API)
Iqoqo lemithetho nezinqubo ezichaziwe ezivumela izinhlelo ezihlukene zesoftware ukuthi zixhumane futhi zishintshane ngedatha.
Isibonelo: Ukusebenzisa i-API ye-OpenAI ukuthumela umyalezo nokuthola impendulo eyenziwe imodeli yolimi kuhlelo lwakho lokusebenza lwe-web.
Ubuhlakani Bokwenziwa Obujwayelekile (AGI) (Artificial General Intelligence (AGI))
Artificial General Intelligence (AGI)
Ifomu elithiyori le-AI elingakwazi ukwenza noma yimuphi umsebenzi wokuhlakanipha umuntu angawenza. Liyakwazi ukufunda kuzo zonke izindawo.
Isibonelo: Uhlelo lwe-AGI lungafunda ukwakhiwa komculo, lwenze ukuhlinzwa, futhi ludlule esivivinyweni sefilosofi ngaphandle kokuhlelwa okukhethekile.
Ubuhlakani Bokwenziwa (AI) (Artificial Intelligence (AI))
Artificial Intelligence (AI)
Ukulinganisa ubuhlakani bomuntu kumishini ehlelwe ukucabanga, ukucabanga, nokuziphatha ngokuzimela.
Isibonelo: I-AI isebenza abasizi bomuntu njengoSiri nezinhlelo zokushayela ezizimele njenge-Tesla Autopilot.
Izimiso Zokuziphatha ze-AI (AI Ethics)
AI Ethics
Isiyalo esikhathazekile ngemiphumela yokuziphatha yokuthuthukiswa nokusetshenziswa kwe-AI, okuhlanganisa ukulunga, ubumfihlo, ukuziphendula, nokungakhethi.
Isibonelo: Ukudala imihlahlandlela yokuvimbela izinhlelo zokuqasha ukukhetha ngokobulili noma ngokobuhlanga.
Ubuhlakani Obandisiwe (Augmented Intelligence)
Augmented Intelligence
Imodeli yokubambisana lapho i-AI ihlanganisa futhi yandisa ubuhlakani bomuntu kunokuba iyibuyisele esikhundleni sayo.
Isibonelo: Izinsiza ze-AI ezisekelwe ku-AI ezikhombisa izinto ezingajwayelekile kodokotela, abenza isinqumo sokugcina.
U-Agent Ozimele (Autonomous Agent)
Autonomous Agent
Uhlelo lwe-AI olukwazi ukwenza izinqumo zalo futhi luthathe izinyathelo ukuze lufeze izinjongo zalo ngaphandle kokuphazamiseka komuntu.
Isibonelo: Iloli lokulethwa elizimele elizulazula emigwaqweni yedolobha futhi ligweme izithiyo ngokuzimela.
Ukubuyela emuva (Backpropagation)
Backpropagation
Inqubo yokuqeqesha amanethiwekhi e-neural ngokuvuselela izisindo ngokubuyela emuva kusuka kumalungu okuphuma kuya kumalungu okufaka, kunciphisa amaphutha okubikezela.
Isibonelo: Kusetshenziswa ekuqeqesheni izihlukanisi zezithombe ukunciphisa izinga lamaphutha ekuboniseni izinombolo ezibhalwe ngesandla.
Ukuchema (Ukuchema Okubalwayo) (Bias (Algorithmic Bias))
Bias (Algorithmic Bias)
Ukuncishiswa okungahlosiwe nokuhlelekile kwemiphumela ye-AI ngenxa yedatha yokuqeqesha engalingani noma engamele.
Isibonelo: Uhlelo lokubona ubuso oluhluleka ukubona abantu abanemibala kaningi ngenxa yokungameleli kwedatha yokuqeqesha.
Idatha Enkulu (Big Data)
Big Data
Amaqoqo amakhulu kakhulu wedatha adinga amathuluzi akhethekile ukugcina, ukuhlaziya, nokukhipha inani, avame ukusetshenziselwa ukuqeqesha amamodeli e-AI.
Isibonelo: Ukusebenzisa izigidi zezinto ezisebenzisanayo ukuqeqesha izinjini zokuncoma ezindaweni zokuthenga nge-inthanethi.
Imodyeli yeBhokisi Elimnyama (Black Box Model)
Black Box Model
Uhlobo lwe-AI noma imodeli yokufunda ngomshini okungelula ukuyiqonda indlela yangaphakathi yayo kubantu, okwenza kube nzima ukuqonda ukuthi izinqumo zenziwa kanjani.
Isibonelo: Inethiwekhi ye-neural ejulile esetshenziselwa ukuvumela izikweletu kodwa ayinikezi incazelo ecacile yokuthi kungani umfakisicelo owodwa wamukelwa kanti omunye wanqatshwa.
Ukubala Okuhlakaniphile (Cognitive Computing)
Cognitive Computing
Izinhlelo ze-AI eziklanyelwe ukulinganisa izinqubo zokucabanga zabantu, njengokucabanga nokufunda, kusetshenziswa izindlela ezifana ne-NLP nokubona amaphethini.
Isibonelo: Uhlelo lokubala oluhlakaniphile olusiza ochwepheshe bezomthetho ukuhlaziya umthetho wecala futhi lubikezela imiphumela.
Umbono Wekhompyutha (Computer Vision)
Computer Vision
Isizinda sobuhlakani bokwenziwa esivumela amakhompyutha ukuthi achaze futhi acubungule idatha ebonakalayo njengezithombe nevidiyo.
Isibonelo: Izinhlelo zokubona ubuso ezibona abantu kumavidiyo okuqapha kusetshenziswa umbono wekhompyutha.
I-Corpus (Corpus)
Corpus
Iqoqo elikhulu lemibhalo ebhaliwe noma ekhulunyiwe esetshenziselwa ukuqeqesha izilimi zemodeli.
Isibonelo: I-Common Crawl dataset iyisikhombisi swebhu somphakathi esisetshenziselwa ukuqeqesha izilimi zemodeli ezinkulu ezifana ne-GPT.
Ukudonswa Kwedatha (Data Drift)
Data Drift
Isenzakalo lapho idatha yokufaka ishintsha ngokuhamba kwesikhathi, okubangela ukwehla kokusebenza kwemodeli.
Isibonelo: Imodeli yokugcinwa okubikezelayo yemishini yezimboni iba nokunemba okuncane njengoba ubuchwepheshe obusha bezinzwa bu introduced.
Ukulebula Idatha (Data Labelling)
Data Labelling
Inqubo yokufaka amalebula kudatha ngamalebula noma amagama-ukhiye ukuze ilungele ukufunda okubhekiselwe kuyo.
Isibonelo: Ukulebula izithombe zezimila eziyinkulungwane njengezingenawo umdlavuza noma ezinomdlavuza ukuqeqesha imodeli yokuthola umdlavuza.
Ukumba Idatha (Data Mining)
Data Mining
Inqubo yokuthola amaphethini anencazelo, ukuhlobana, nokungajwayelekile kumadatha amakhulu.
Isibonelo: Abathengisi basebenzisa ukumba idatha ukuthola ukuthi abantu abathenga izimpahla zezingane bavame ukuthenga nebhiya.
Ukufunda Okujulile (Deep Learning)
Deep Learning
Isizinda sokufunda ngomshini esisebenzisa amanethiwekhi e-neural anezigaba eziningi ukumodela amaphethini ayinkimbinkimbi kudatha.
Isibonelo: Ukufunda okujulile kusetshenziswa ezilimini zemodeli ezifana ne-GPT-4 namamodeli okukhiqiza izithombe ezifana ne-Stable Diffusion.
Amamodeli Okudiliza (Diffusion Models)
Diffusion Models
Isigaba samamodeli okukhiqiza afunda ukukhiqiza idatha ngokuguqula kancane umsindo ongahleliwe ube yimiphumela ehlelekile.
Isibonelo: I-Stable Diffusion idala izithombe ezingokoqobo kusuka kumiyalezo yombhalo kusetshenziswa izindlela zokudiliza.
Ukubeka (Embedding)
Embedding
Ukumelelwa kwedatha kwenombolo, kuvame ukusetshenziselwa ukubamba incazelo yezincazelo zamagama, izithombe, noma imisho.
Isibonelo: Ku-NLP, igama elithi 'ibhange' lingase libe nokubeka okufanayo negama elithi 'imali' kodwa lihlukile kwelithi 'umngcele womfula' kuye ngomongo.
I-Epoch (Epoch)
Epoch
Ukuphindwa okugcwele phezu kwedatha yokuqeqesha yonke phakathi nenqubo yokuqeqesha imodeli yokufunda ngomshini.
Isibonelo: Uma idatha inezibonelo eziyi-1,000 futhi imodeli ibona zonke izikhathi ezilodwa phakathi nokuqeqeshwa, lokho kungaba yi-epoch eyodwa.
AI Ezinzile (Ethical AI)
Ethical AI
Ifilosofi yokuklama nokusetshenziswa eqinisekisa ukuthi ubuchwepheshe be-AI busebenza ngokucacile, ngokulinganayo, futhi ngokuhambisana nezindinganiso zomphakathi.
Isibonelo: Ithuluzi lokuqasha i-AI elihlanganisa izinsiza zokukhetha ukuze kuvinjwe ukucwaswa kwabazobhapathizwa abavela ezimbalwa.
Uhlelo Lwezingcweti (Expert System)
Expert System
Uhlelo lwe-AI olulinganisa amakhono okwenza izinqumo engcweti yomuntu endaweni ethile kusetshenziswa imithetho nokucabanga.
Isibonelo: Uhlelo lwezingcweti olusetshenziswa kwezolimo ukuncoma ukwelashwa kwezitshalo ngokusekelwe kudatha yenhlabathi nomlando wezinambuzane.
AI Echazwayo (XAI) (Explainable AI (XAI))
Explainable AI (XAI)
Izinhlelo ze-AI eziklanyelwe ukwenza izinqubo zazo zangaphakathi kanye nezinqumo ziqondakale kubantu, zikhuphula ukwethenjwa nokuziphendula.
Isibonelo: I-AI yokuxilonga yezokwelapha enikeza isiphakamiso kuphela kodwa futhi ichaza ukuthi yiziphi izimpawu eziholele kuleso siphetho.
Ukufunda Okunemibono Embalwa (Few-shot Learning)
Few-shot Learning
Indlela yokufunda ngomshini lapho imodeli iqeqeshwa noma ilungiswa kusetshenziswa izibonelo ezimbalwa ezilebuliwe.
Isibonelo: Ukwenza ngezifiso i-LLM ukubhala ama-imeyili asemthethweni ngemuva kokuyibonisa izibonelo eziyi-10 kuphela.
Ukulungisa (Fine-tuning)
Fine-tuning
Inqubo yokuthatha imodeli eqeqeshwe ngaphambili bese uyiyoqeqesha kwi-dataset entsha, encane ukuze uyenze ibe nekhono kumsebenzi othile.
Isibonelo: Ukulungisa i-LLM ejwayelekile njenge-GPT kumadokhumenti angaphakathi asemthethweni ukuze kwakhiwe umsizi wokudweba wezomthetho.
Imodyeli Yesisekelo (Foundation Model)
Foundation Model
Imodeli enkulu eqeqeshwe kudatha ehlukahlukene futhi ebanzi engalungiswa kumisebenzi eminingi ephansi.
Isibonelo: I-GPT-4 ne-PaLM 2 yizilimi zemodeli eziyisisekelo ezikwazi ukufingqa, ukubuzwa nokuphendulwa, ukuhumusha, nokunye.
I-Fuzzy Logic (Fuzzy Logic)
Fuzzy Logic
Uhlobo lokucabanga olubhekana namazinga acishe abe yiqiniso kunokucabanga okuyiqiniso/okungelona iqiniso (binary), okusiza ekucabangeni ngaphansi kokungaqiniseki.
Isibonelo: Kusetshenziswa ezinhlelweni zokulawula isimo sezulu ukulungisa izinga lokushisa ngokusekelwe kumazinga afuzzy afana 'nokushisa kancane' noma 'okubandayo kakhulu'.
Inethiwekhi Yokukhiqiza Eyinqaba (GAN) (Generative Adversarial Network (GAN))
Generative Adversarial Network (GAN)
Isakhiwo semodeli yokukhiqiza lapho izinhlelo ezimbili - umkhiqizi kanye nomhlukanisi - ziqhudelana ukuthuthukisa ikhwalithi yomphumela.
Isibonelo: Ama-GAN asetshenziselwa ukudala amavidiyo we-deepfake noma ukukhiqiza izithombe zomkhiqizo ezingokoqobo kusuka emidwebeni.
I-Generative AI (Generative AI)
Generative AI
Isigaba sobuhlakani bokwenziwa esingakha okuqukethwe okusha - njengombhalo, izithombe, umculo, noma ividiyo - kusuka kudatha yokuqeqesha.
Isibonelo: I-ChatGPT ikhiqiza izindatshana zebhulogi noma i-Midjourney idala ubuciko bedijithali kusuka kumiyalezo yombhalo.
Isihumushi Esikhiqizayo Esihleliwe (GPT) (Generative Pre-trained Transformer (GPT))
Generative Pre-trained Transformer (GPT)
Isigaba sezilimi zemodeli ezinkulu ezakhiwe yi-OpenAI ezisebenzisa isakhiwo sesihumushi futhi iqeqeshwe ngaphambili kudatha yombhalo omkhulu ukuze yenze imisebenzi yolimi eyahlukahlukene.
Isibonelo: I-GPT-4 iyakwazi ukubhala izindatshana, ukuhumusha izilimi, nokufingqa amadokhumenti ngemiyalezo encane.
Umshini Wofuzo (Genetic Algorithm)
Genetic Algorithm
Indlela yokuthuthukisa ephefumulelwe ukukhethwa kwemvelo lapho izixazululo zithuthuka ngokuhamba kwesikhathi ngokusebenzisa ukuguqulwa, ukuhlanganiswa, nokukhethwa.
Isibonelo: Kusetshenziselwa ukuklama izakhiwo zenethiwekhi ezingcono ngokulinganisa ukusinda kwabanamandla kakhulu.
Ukuhalalisela (Hallucination)
Hallucination
Ukuphuma kokuqukethwe okuzwakala kahle kodwa okungalungile ngokweqiniso noma okungenangqondo yi-AI model.
Isibonelo: Imodyeli yolimi yakha isicatshulwa esingekho noma inikeza amaqiniso omlando angamanga.
I-Heuristic (Heuristic)
Heuristic
Indlela esebenzayo yokuxazulula izinkinga engaqinisekisi isixazululo esiphelele kodwa esanele izinjongo ezisheshayo.
Isibonelo: Ukusebenzisa umthetho wesithupha ukulinganisa isikhathi sokulethwa kuhlelo lokubikezela lwezinto zokuhambisa.
I-Hyperparameter (Hyperparameter)
Hyperparameter
Inani lokucushwa elibekwe ngaphambi kokuqeqeshwa kwemodeli yokufunda ngomshini, njengesilinganiso sokufunda noma inombolo yezigaba.
Isibonelo: Ukulungisa usayizi webheshi kusuka ku-32 kuya ku-128 ukuthuthukisa isivinini sokuqeqeshwa nokusebenza kwemodeli.
Ukuboniswa (Inference)
Inference
Inqubo yokusebenzisa imodeli yokufunda ngomshini eqeqeshwe ukwenza izibikezeli noma ukukhiqiza imiphumela kusuka kudatha yokufaka entsha.
Isibonelo: Ukusebenzisa imodeli ye-GPT elungisiwe ukudweba ama-imeyili eqembu losizo lwamakhasimende.
Ukutholwa Kwesifiso (Intent Detection)
Intent Detection
Umsebenzi ekuqondeni ulimi olwemvelo lapho uhlelo luthola injongo noma inhloso yomsebenzisi emyalezweni.
Isibonelo: Ku-chatbot, ukuqonda 'Ngifuna ukubhuka indiza' njengesifiso sokubhuka uhambo.
Inthanethi Yezinto (IoT) (Internet of Things (IoT))
Internet of Things (IoT)
Inethiwekhi yezinto ezibonakalayo ezixhunyiwe ezihlanganiswe izinzwa, isoftware, nezinye ubuchwepheshe ukuqoqa nokushintshana ngedatha.
Isibonelo: Ama-thermostat ahlakaniphile namafriji abika idatha yokusetshenziswa futhi alungise izilungiselelo kusetshenziswa izibalo ze-AI.
Ukuhunyushwa (Interpretability)
Interpretability
Izinga lapho umuntu angaqonda khona izindlela zangaphakathi zemodeli yokufunda ngomshini kanye nenqubo yayo yokwenza izinqumo.
Isibonelo: Isihlahla sokwenza izinqumo siqondakala kakhulu kunenethiwekhi ye-neural ejulile ngoba izinqumo zayo ziyalandeleka.
I-Jupyter Notebook (Jupyter Notebook)
Jupyter Notebook
Indawo yokubala yokusebenzisana evulekile evumela abasebenzisi ukuthi babhale ikhodi, babonise imiphumela, futhi babhale ukuhlaziywa kusikhombimsebenzisi esisodwa.
Isibonelo: Ososayensi bedatha basebenzisa ama-Jupyter Notebooks ukwenza amamodeli okufunda ngomshini nokwabelana ngemiphumela.
I-K-Nearest Neighbours (KNN) (K-Nearest Neighbours (KNN))
K-Nearest Neighbours (KNN)
I-algorithm elula, engaphathelene nokufunda ngomshini esetshenziselwa ukuhlukaniswa nokubuyela emuva. Yenza izinqumo ngokusekelwe ezibonelweni zokuqeqesha eziseduze esikhundleni sezici.
Isibonelo: Ukuhlukanisa isithelo esisha njenge-apula noma ipere, i-KNN ihlola ukuthi yiziphi izithelo ezilebuliwe eziseduze ngesimo nombala.
Imephu Yokuqonda (Knowledge Graph)
Knowledge Graph
Isakhiwo sedatha esisebenzisa izindawo namalungu ukumelela nokugcina izincazelo ezixhunyiwe zezinto kanye nobudlelwano bazo.
Isibonelo: Iphaneli yolwazi ye-Google inikwa amandla yimephu yolwazi exhuma izinto ezifana nabantu, izindawo, nezehlakalo.
Ukuthuthukiswa Kwemodeli Yokufunda Ulimi (LLMO) (Language Learning Model Optimisation (LLMO))
Language Learning Model Optimisation (LLMO)
Izindlela ezisetshenziselwa ukuthuthukisa ukusebenza, ukusebenza kahle, noma ukulungiswa kwezilimi zemodeli ezinkulu kumisebenzi noma izindawo ezithile.
Isibonelo: Ukusebenzisa ukubalwa nokulungiswa komyalezo ukuthuthukisa i-LLM ukusetshenziswa kwezingqungquthela.
Imodyeli Yolimi Olukhulu (LLM) (Large Language Model (LLM))
Large Language Model (LLM)
Uhlobo lwemodeli yokufunda ejulile eqeqeshwe ngamanani amakhulu edatha yombhalo ekwazi ukukhiqiza, ukuqonda, nokucabanga ngolimi lwabantu.
Isibonelo: I-ChatGPT ne-Claude yizilimi zemodeli ezinkulu eziqeqeshwe ukusiza ekubhaleni, ukuhlela, nokuphendula imibuzo.
Isikhala Esifihliwe (Latent Space)
Latent Space
Ukumelelwa okuyinkimbinkimbi okuyinkimbinkimbi lapho okufakiwe okufanayo kuqoqwa eduze, kusetshenziselwa amamodeli okukhiqiza kanye nokubeka.
Isibonelo: Ekukhiqizeni izithombe, ukulawula isikhala esifihliwe kungashintsha izici ezifana nokukhanya noma umuzwa.
Isilinganiso Sokufunda (Learning Rate)
Learning Rate
I-hyperparameter eyinhloko ekuqeqeshweni elawula ukuthi izisindo zemodeli zilungiswa kangakanani maqondana ne-gradient yelahleko.
Isibonelo: Isilinganiso sokufunda esiphakeme singase siholele ekudluleleni kwama-minima, kanti isilinganiso esiphansi kakhulu siyalibazisa inqubekela phambili yokuqeqeshwa.
Ukufunda Ngomshini (ML) (Machine Learning (ML))
Machine Learning (ML)
Isikhundla se-AI esivumela izinhlelo ukuthi zifunde kusuka kudatha futhi zithuthukise ukusebenza ngaphandle kokuhlelwa ngokuqondile.
Isibonelo: Izihlungi ze-Spam zisebenzisa ukufunda ngomshini ukuhlukanisa ama-imeyili njenge-spam noma cha ngokusekelwe ezibonelweni ezedlule.
Ukudonswa Kwemodeli (Model Drift)
Model Drift
Isenzakalo lapho ukunemba kwemodeli kwehla ngokuhamba kwesikhathi ngenxa yezinguquko kudatha noma imvelo.
Isibonelo: Imodeli yokuthola inkohlakalo iba nokunemba okuncane njengoba izindlela zenkohlakalo ziguquka.
Ukuqeqeshwa Kwemodeli (Model Training)
Model Training
Inqubo yokudlulisa idatha kumodeli yokufunda ngomshini nokulungisa iziphumo zayo ukunciphisa iphutha.
Isibonelo: Ukuqeqesha injini yokuncoma kumlando wokuthenga amakhasimende ukuze kunconywe imikhiqizo emisha.
I-Multimodal AI (Multimodal AI)
Multimodal AI
Izinhlelo ze-AI ezikwazi ukucubungula nokuhlanganisa izinhlobo eziningi zedatha ezifana nombhalo, izithombe, umsindo, nevidiyo.
Isibonelo: Imodeli efana ne-GPT-4 Vision ekwazi ukufunda umbhalo futhi iqonde izithombe ngesikhathi esifanayo.
Ukucubungula Ulimi Olwemvelo (NLP) (Natural Language Processing (NLP))
Natural Language Processing (NLP)
Isizinda se-AI esigxile ekusebenzisaneni phakathi kwamakhompyutha nezilimi zabantu (zemvelo). Ivumela imishini ukuthi ifunde, iqonde, futhi iphendule ngolimi lwabantu.
Isibonelo: I-NLP isetshenziswa kubasizi bezwi, izinhlelo zokusebenza zokuhumusha ulimi, kanye nama-chatbot.
Inethiwekhi Ye-Neural (Neural Network)
Neural Network
Imodeli yokufunda ngomshini ephefumulelwe isakhiwo sobuchopho bomuntu, equkethe izigaba zamalungu axhunyiwe (ama-neuron).
Isibonelo: Amanethiwekhi e-Neural angemuva kwamamodeli okufunda ajulile asetshenziswa ekubonisweni kwezithombe nasekuqashelweni kwezwi.
Umsindo (Noise)
Noise
Ulwazi olungahleliwe noma olungabalulekile kudatha olungafihla amaphethini anencazelo futhi luthinte kabi ukusebenza kwemodeli.
Isibonelo: Amaphutha ezinzwa noma okufakiwe kwedatha okuneziphutha kungabhekwa njengomsindo.
I-Ontology (Ontology)
Ontology
Isakhiwo esihlelekile esihlukanisa futhi sichaze ubudlelwano phakathi kwemibono endaweni ethile, evame ukusetshenziswa ezinhlelweni ze-AI ezihlakaniphile.
Isibonelo: I-ontology kwezokwelapha ingachaza ukuthi izimpawu zihlobana kanjani nezifo nokwelashwa.
Ukweqisa (Overfitting)
Overfitting
Iphutha lokumodela lapho imodeli yokufunda ngomshini ibamba umsindo wedatha yokuqeqesha futhi yenza kabi kudatha entsha.
Isibonelo: Imodeli egcina izimpendulo zokuqeqeshwa kodwa ayikwazi ukubhekana nedatha engabonakali ihlolwa ngokweqisa.
Ukubikezela Okuhlakaniphile (Predictive Analytics)
Predictive Analytics
Ukusetshenziswa kwedatha, izibalo, kanye ne-AI ukuthola ukufaneleka kwemiphumela yesikhathi esizayo ngokusekelwe kudatha yomlando.
Isibonelo: Abathengisi basebenzisa izibalo zokubikezela ukubikezela isidingo semikhiqizo ethile.
Ukuqeqeshwa Kwangaphambili (Pre-training)
Pre-training
Inqubo yokuqeqesha imodeli ekuqaleni kudatha enkulu, ejwayelekile ngaphambi kokuyilungisa kumisebenzi ethile.
Isibonelo: Amamodeli e-GPT aqeqeshwa ngaphambili kumakhorasi amakhulu ngaphambi kokwenziwa ngokwezifiso kwama-chatbot osizo lwamakhasimende.
Ubunjiniyela be-Prompt (Prompt Engineering)
Prompt Engineering
Ubuciko nesayensi yokwakha imiyalezo esebenzayo ukuze kuqondiswe imiphumela yezilimi zemodeli ezinkulu.
Isibonelo: Ukwengeza imiyalelo yohlelo efana 'Niphendule njengomfundisi ohloniphayo' kuyisibonelo sobunjiniyela be-prompt.
Ukubalwa (Quantisation)
Quantisation
Inqubo yokunciphisa imodeli eyenza izisindo nama-activations asebenzise izinombolo ezincane, kuthuthukisa ukusebenza kahle.
Isibonelo: Ukubala imodeli kusuka ku-32-bit kuya ku-8-bit kuthuthukisa ukusebenza kumadivayisi eselula.
Ukubala Kwe-Quantum (Quantum Computing)
Quantum Computing
Iphadigimu entsha yokubala esekelwe ku-quantum mechanics, enikeza amandla okucubungula okuphindaphindayo.
Isibonelo: Ukubala kwe-Quantum kungase kusheshe ukuqeqeshwa kwe-AI ngaphezu kwemikhawulo yakudala.
Injini Yokucabanga (Reasoning Engine)
Reasoning Engine
Uhlelo ku-AI oluthola iziphetho ezinengqondo kusuka kuqoqo leziqinisekiso noma idatha kusetshenziswa imithetho noma izibalo zokuboniswa.
Isibonelo: Ithuluzi lokuxilonga i-AI lisebenzisa injini yokucabanga ukuthola izimo zezokwelapha ezingenzeka ngokusekelwe ezimpawini.
Ukufunda Ngokuqinisa (RL) (Reinforcement Learning (RL))
Reinforcement Learning (RL)
Isizinda sokufunda ngomshini lapho ama-agent afunda ngokusebenzisana nemvelo yawo ukuze andise imiklomelo eqongelelekile.
Isibonelo: Inqola efunda ukuhamba ngokuzama nokuhluleka kusetshenziswa izindlela ze-RL.
Ukufunda Ngokuqinisa Ngempendulo Yomuntu (RLHF) (Reinforcement Learning with Human Feedback (RLHF))
Reinforcement Learning with Human Feedback (RLHF)
Indlela yokufunda lapho izintandokazi zabantu ziqondisa isignali yomklomelo we-AI, evame ukusetshenziswa ekulungiseni izilimi zemodeli.
Isibonelo: I-ChatGPT yaqeqeshwa nge-RLHF ukuze ikhiqize izimpendulo ezisizayo neziphephile.
Ukukhiqiza Okuhlanganiswe Nokubuyisa (RAG) (Retrieval-Augmented Generation (RAG))
Retrieval-Augmented Generation (RAG)
Indlela ehlanganisa ukubuyisa ulwazi nokukhiqiza, lapho i-LLM ithola amadokhumenti afanele ukuthuthukisa impendulo yayo.
Isibonelo: Umsizi we-AI ubuyisa futhi ucacisa izincazelo zomkhiqizo ngenkathi ekhiqiza impendulo kumbuzo wezobuchwepheshe.
Ukufunda Okuzisizayo (Self-Supervised Learning)
Self-Supervised Learning
Indlela yokuqeqesha lapho imodeli ifunda amaphethini ngokukhiqiza amalebula ayo kusuka kudatha eluhlaza, kunciphisa ukuncika kwidatha ehlaziywe ngabantu.
Isibonelo: I-BERT iqeqeshwe ngokufunda okuzisizayo ngokubikezela amagama alahlekile embhalweni.
Ukuboniswa Okuhlakaniphile (Semantic Search)
Semantic Search
Indlela yokusesha eqonda isifiso somsebenzisi kanye nencazelo yomongo, hhayi nje ukufanisa amagama-ukhiye.
Isibonelo: Ukusesha 'indlela yokulungisa ithephu eyophayo' kubuyisa imihlahlandlela noma ngabe igama elithi 'ithephu eyophayo' alikho kudokhumenti.
Ukuhlaziywa Kwemizwa (Sentiment Analysis)
Sentiment Analysis
Inqubo yokuthola imizwa, imibono, noma izimo zengqondo embhalweni, ngokuvamile ukuhlukanisa njengokuhle, okungalungile, noma okungathathi hlangothi.
Isibonelo: Ukuhlaziya ama-tweet ukuze kutholwe ukusabela komphakathi kumkhiqizo omusha.
I-Stochastic (Stochastic)
Stochastic
Okubandakanya ukungahleliwe noma ukuziphatha okungokomlando, kuvame ukusetshenziswa ku-generative AI nasezibalo zokuthuthukisa.
Isibonelo: Umphumela we-GPT-4 uyahlukahluka ngokuya ngokufaka okufanayo ngenxa yenqubo yayo yokudweba ye-stochastic.
I-Strong AI (Strong AI)
Strong AI
Kuye futhi kwaziwe njenge-Artificial General Intelligence (AGI), kubhekiselwa kumishini enamakhono okubala afana nawomuntu kuzo zonke izindawo.
Isibonelo: I-AI yesikhathi esizayo ezobhala ngokuzimela amanoveli, ihlele amadolobha, futhi ixazulule izinkinga zokuziphatha ngokufanayo.
Ubuhlakani Bokwenziwa Obuphezulu (SAI) (Super Artificial Intelligence (SAI))
Super Artificial Intelligence (SAI)
I-AI ethayori eyidlula kakhulu ubuhlakani bomuntu kuzo zonke izici - ukucabanga, ubuciko, ubuhlakani bempilo, njll.
Isibonelo: I-SAI ingathuthukisa ngokomthetho izifundo ezintsha kanye nefilosofi ngokuzimela.
Ukufunda Okubhekiselwe (Supervised Learning)
Supervised Learning
Indlela yokufunda ngomshini lapho amamodeli aqeqeshwa ngedatha enezimpawu ukuze afunde ukufaniswa kokufaka-okuphumayo.
Isibonelo: Ukufundisa imodeli ukuhlukanisa ama-imeyili njenge-spam noma cha kusetshenziswa izibonelo zomlando.
Idatha Eyenziwe (Synthetic Data)
Synthetic Data
Idatha eyenziwe ngokwenziwa efana nedatha yomhlaba wangempela, evame ukusetshenziselwa ukuqeqeshwa lapho idatha yangempela ingavamile noma ibucayi.
Isibonelo: Ukudala izithombe zezokwelapha ezenziwe ukuqeqesha amamodeli okuxilonga ngaphandle kokwephula ubumfihlo besiguli.
I-Token (Token)
Token
Iyunithi yombhalo ecubungulwa yi-LLMs - ngokuvamile igama noma ucezu lwegama.
Isibonelo: Umusho othi 'Sawubona lizwe!' uhlukaniswe izimpawu ezi-3: 'Sawubona', 'lizwe', kanye '!'.
I-Tokenisation (Tokenisation)
Tokenisation
Inqubo yokuhlukanisa umbhalo ube izimpawu ukuze kucutshungulwe imodeli.
Isibonelo: Ku-NLP, 'I-ChatGPT inhle' iba yi-['Chat', 'G', 'PT', 'iyinhle'].
Ukufunda Okudlulisiwe (Transfer Learning)
Transfer Learning
Ukusebenzisa ulwazi oluvela komunye umsebenzi ukuthuthukisa ukufunda komunye umsebenzi ohlobene, kunciphisa isikhathi sokuqeqeshwa nezidingo zedatha.
Isibonelo: Ukulungisa imodeli eqeqeshwe embhalweni wesiNgisi ukwenza ukuhlaziywa kwemizwa kolunye ulimi.
Isihumushi (Transformer)
Transformer
Isakhiwo senethiwekhi ye-neural esisebenzisa izindlela zokunakwa ukumodela idatha elandelanayo, esetshenziswa kabanzi kuma-LLM.
Isibonelo: I-BERT, i-GPT, ne-T5 zonke ziyizilimi zemodeli ezisekelwe kusihumushi.
Ukwehluleka (Underfitting)
Underfitting
Lapho imodeli ilula kakhulu ukubamba amaphethini wedatha yokuqeqeshwa, okuholela ekusebenzeni okungalungile.
Isibonelo: Imodeli elinganisa izimpendulo zokuqeqeshwa kodwa ayikwazi ukubhekana nedatha engabonakali ingase yehluleke.
Ukufunda Okungabhekiselwe (Unsupervised Learning)
Unsupervised Learning
Indlela yokufunda lapho amamodeli athola amaphethini noma amaqoqo kudatha engabhalwanga.
Isibonelo: Ukuhlukanisa amakhasimende ngokusekelwe ekuziphatheni kokuthenga ngaphandle kwamalebula achaziwe.
Isifiso Somsebenzisi (User Intent)
User Intent
Inhloso noma injongo ngemuva kokufuna noma ukusebenzisana komsebenzisi.
Isibonelo: Umsebenzisi othayipha 'indlela yokubhaka ikhekhe' cishe uhlose ukuthola iresiphi.
Isethi Yokuvidiyo (Validation Set)
Validation Set
Ingxenye yedatha esetshenziselwa ukuhlola ukusebenza kwemodeli phakathi nokuqeqeshwa nokulungisa ama-hyperparameters.
Isibonelo: Kusetshenziselwa ukuthola ukweqisa ngaphambi kokuhlolwa kokugcina.
Isizindalwazi Se-Vector (Vector Database)
Vector Database
Isizindalwazi esiklanyelwe ukugcina nokusesha ukubeka kwe-vector okusetshenziswa emisebenzini ye-AI njengokufuna okufanayo kanye ne-RAG.
Isibonelo: I-Pinecone ne-Weaviate yizizindalwazi ze-vector zokugcina ukubeka umbhalo noma izithombe.
Ukubeka kwe-Vector (Vector Embedding)
Vector Embedding
Ukumelelwa kwenombolo kwedatha okugcina incazelo kanye nobudlelwano esikhundleni se-vector.
Isibonelo: Amagama athi 'inkosi' kanye 'nendlovukazi' anokubeka okufanayo okunokuhlukahluka kobulili okuncane.
Umsizi We-Virtual (Virtual Assistant)
Virtual Assistant
I-agent yesoftware enikwe amandla yi-AI esiza abasebenzisi baqedele imisebenzi ngokusebenzisa ingxoxo noma imiyalo yezwi.
Isibonelo: I-Siri, i-Alexa, kanye ne-Google Assistant yizisizi ezidumile ze-virtual.
Ukuqashelwa Kwezwi (Voice Recognition)
Voice Recognition
Ubuchwepheshe obuchaza futhi buguqula ulimi olukhulunywayo lube yombhalo noma isenzo.
Isibonelo: Ukuthayipha ngezwi kanye nemiyalo yezwi kuncike ezinhlelweni zokuqashelwa kwezwi.
I-Weak AI (Weak AI)
Weak AI
Izinhlelo ze-AI eziklanyelwe ukwenza umsebenzi owodwa, owahlukaniswe ngokuqondile ngaphandle kobuhlakani obujwayelekile.
Isibonelo: I-AI yokudlala ibhodi le-chess engakwazi ukuqonda ulimi noma ukushayela imoto iyisibonelo se-AI eyi-weak.
Ukuklwebha Iwebhu (Web Scraping)
Web Scraping
Ukuqoqwa okuzenzakalelayo kolwazi kusuka kumawebhusayithi, kuvame ukusetshenziselwa ukuqoqa idatha yokuqeqeshwa noma ukuqapha okuqukethwe.
Isibonelo: Ukuklwebha izinhlu zezindlu ukuze kuqeqeshwe imodeli yokulinganisa impahla.
Isisindo (Weight)
Weight
Ipharamitha kunethiwekhi ye-neural enquma amandla omthelela wesinye isikhungo kwenye.
Isibonelo: Izisindo zilungiswa phakathi nokuqeqeshwa ukunciphisa iphutha lemodyeli.
Ukukhuluma (Whisper)
Whisper
Imodeli yokukhuluma ukuya kumbhalo eyakhiwe yi-OpenAI ekwazi ukubhala umsindo ngezilimi eziningi.
Isibonelo: Ukukhuluma kungabhala izinkulumo namaphodikasti ngokunemba okuphezulu.
YAML (YAML)
YAML
Ifomethi efundeka ngabantu yokuhlukaniswa kwedatha, evame ukusetshenziselwa amafayela okucushwa emisebenzini yokufunda ngomshini.
Isibonelo: Ukuchaza amapharamitha emodeli kufayela le-YAML ukuze kuqeqeshwe ku-PyTorch.
Ukufunda Okunolwazi Zero (Zero-shot Learning)
Zero-shot Learning
Ikhono lemodyeli lokwenza imisebenzi engakaze iqeqeshwe ngayo ngokusebenzisa ulwazi olujwayelekile.
Isibonelo: Imodeli ephendula imibuzo yezomthetho noma ngabe ayizange iqeqeshwe ngokukhethekile ngedatha yezomthetho.
I-Zettabyte (Zettabyte)
Zettabyte
Iyunithi yedatha yedijithali elingana nama-sextillion (10^21) ama-byte, evame ukusetshenziselwa ukuchaza isilinganiso sedatha ye-inthanethi.
Isibonelo: Umkhuhlane womhlaba wonke we-inthanethi udlule i-zettabyte eyodwa ngonyaka ngo-2016.
