Ukunqongophala (Alignment)
Alignment
Inkqubo yokuqinisekisa ukuba iinjongo, iziphumo, kunye neziphatha-kahle zenkqubo ye-AI ziyahambelana neenjongo kunye nemilinganiselo yabantu. Oku kubaluleke kakhulu kwiinkqubo eziphambili ezinokuthi ziphuhlise iziphatha-kahle ezingacwangciswanga ngokucacileyo.
Umzekelo: Ukuqinisekisa ukuba i-chatbot yenkxaso yempilo yengqondo ayisoze icebise ngezenzo eziyingozi nokuba zeziphi na izicelo.
Isijongano sohlelo lokusebenza (API) (Application Programming Interface (API))
Application Programming Interface (API)
Iseti yemigaqo kunye neenqobo ezichaziweyo ezivumela iinkqubo ezahlukeneyo zesoftware ukuba zinxibelelane kwaye zithathe idatha.
Umzekelo: Ukusebenzisa i-API ye-OpenAI ukuthumela isicelo kwaye ufumane impendulo eyenziwe ngumzekelo wolwimi kwi-app yakho yewebhu.
Ubukrelekrele obubanzi bokwenziwa (AGI) (Artificial General Intelligence (AGI))
Artificial General Intelligence (AGI)
Uhlobo oluthathwa njengobukrelekrele bokwenziwa olunako ukwenza nawuphi na umsebenzi wokucinga umntu angawenza. Iphumeza ukufunda kuzo zonke iindawo.
Umzekelo: Inkqubo ye-AGI ingafunda ukubhalwa komculo, yenze utyando, kwaye iphumelele uvavanyo lwefilosofi ngaphandle kokucwangciswa komsebenzi othile.
Ubukrelekrele bokwenziwa (AI) (Artificial Intelligence (AI))
Artificial Intelligence (AI)
Ukulinganisa ubukrelekrele bomntu koomatshini abacwangciswe ukuba bacinge, bacalule, kwaye benze ngokuzimeleyo.
Umzekelo: I-AI inika amandla abancedisi babucala njengoSiri kunye neenkqubo zokuhamba ngokuzimeleyo njengeTesla Autopilot.
I-AI Ethics (AI Ethics)
AI Ethics
Isiyobisi esijongene neziphumo zokuziphatha zokuphuhliswa kunye nokusetyenziswa kwe-AI, kubandakanya ubulungisa, imfihlo, uxanduva, kunye nokungabandlululi.
Umzekelo: Ukudala izikhokelo zokuthintela ii-algorithms zokuqeshwa ukuba zingabandlululi ngokwesini okanye ubuhlanga.
Ubukrelekrele obongeziweyo (Augmented Intelligence)
Augmented Intelligence
Imodeli yokusebenzisana apho i-AI ihambelana kwaye yongeza ubukrelekrele bomntu endaweni yokuyibuyisela.
Umzekelo: Izixhobo zokufota nge-AI eziphawula izinto ezingaqhelekanga kubagqirha, abenza ugqibo lokugqibela.
I-Autonomous Agent (Autonomous Agent)
Autonomous Agent
Inkqubo ye-AI enako ukwenza izigqibo zayo kwaye ithathe amanyathelo ukufezekisa iinjongo zayo ngaphandle kokungenelela komntu.
Umzekelo: Iloli yokuhambisa ngokuzimeleyo ehamba ezitratweni zedolophu kwaye iyayiphepha imiqobo ngokuzimeleyo.
Ukubuyela emva (Backpropagation)
Backpropagation
Indlela yokuqeqesha iinethiwekhi ze-neural ngokuhlaziya ubunzima ngokubuyela umva ukusuka kwisiphumo ukuya kwizigaba zokufaka, ukunciphisa iimpazamo zokubala.
Umzekelo: Isetyenziswa kuqeqesho lwezibonisi zemifanekiso ukunciphisa izinga lempazamo ekuboneni amanani abhalwe ngesandla.
Ukubandlulula (Ukubandlulula kwe-Algorithmic) (Bias (Algorithmic Bias))
Bias (Algorithmic Bias)
Ukuthanda okungacwangciswanga kunye nokucwangcisiweyo kwiziphumo ze-AI ngenxa yedatha yokuqeqesha engalinganiyo okanye engabonakaliyo.
Umzekelo: Inkqubo yokubonisa ubuso ephosakele abantu abanesikhumba esimnyama rhoqo ngenxa yokungabikho kwabo kwidatha yokuqeqesha.
Idatha enkulu (Big Data)
Big Data
Iidatha ezinkulu kakhulu ezifuna izixhobo ezikhethekileyo ukugcina, ukuhlalutya, kunye nokukhupha ixabiso, ezisetyenziswa rhoqo ukuqeqesha iimodeli ze-AI.
Umzekelo: Ukusebenzisa izigidi zokusebenzisana kwabasebenzisi ukuqeqesha ii-algorithms zokucebisa kwiindawo zokuthenga.
Imizobo yeBhokisi Elimnyama (Black Box Model)
Black Box Model
Uhlobo lwe-AI okanye imodeli yokufunda ngomatshini apho ingqiqo yangaphakathi ayichazeki lula ngabantu, kwenza kube nzima ukuqonda ukuba izigqibo zenziwa njani.
Umzekelo: Inethiwekhi ye-neural enzulu esetyenziselwa ukuvuma izibonelelo kodwa ayinikezi nkcazo ecacileyo yokuba kutheni omnye umfaki wamkelwe kwaye omnye wenqatshiwe.
Ukubala ngokuqonda (Cognitive Computing)
Cognitive Computing
Iinkqubo ze-AI ezicwangciswe ukulinganisa iinkqubo zokucinga zomntu, njengokubaleka nokufunda, kusetyenziswa iindlela ezifana ne-NLP kunye nokubonisa imizekelo.
Umzekelo: Inkqubo yokubala ngokuqonda enceda abasebenzi bezomthetho bahlalutye umthetho wamacala kwaye bafikelele kwiziphumo.
Umbono weKhompyutha (Computer Vision)
Computer Vision
Inkalo yobukrelekrele bokwenziwa evumela oomatshini ukuba bachaze kwaye bacubungule idatha ebonakalayo efana nemifanekiso kunye nevidiyo.
Umzekelo: Iinkqubo zokubonisa ubuso ezibonisa abantu kwifotografi yokhuseleko kusetyenziswa umbono wekhompyutha.
I-Corpus (Corpus)
Corpus
Ukuqokelelwa okukhulu kwezibhalo ezibhaliweyo okanye ezithethwayo ezisetyenziselwa ukuqeqesha iimodeli zolwimi.
Umzekelo: Idatha yeCommon Crawl yikhorpus yewebhu kawonononke esetyenziselwa ukuqeqesha iimodeli ezinkulu zolwimi ezifana neGPT.
Ukudrift kwedatha (Data Drift)
Data Drift
Isiganeko apho idatha yokufaka itshintsha ngokuhamba kwexesha, ibangela ukuba ukusebenza komodeli kwehla.
Umzekelo: Imfono yokuqikelela yokugcinwa kwezixhobo zemizi-mveliso iba ngaphantsi ngokuchanekileyo njengoko kuthe ngeniswa ubuchwepheshe obutsha bezinzwa.
Ukulebula idatha (Data Labelling)
Data Labelling
Inkqubo yokubeka iilebula kwidatha ukuze ilungele ukufunda okubekwe esweni.
Umzekelo: Ukulebula amawaka emifanekiso ye-tumour njenge-benign okanye i-malignant ukuqeqesha imodeli yokufumanisa umhlaza.
Imigodi yedatha (Data Mining)
Data Mining
Inkqubo yokufumana imizekelo ethembisayo, ubudlelwane, kunye nezinto ezingaqhelekanga kwidatha enkulu.
Umzekelo: Abathengisi basebenzisa imigodi yedatha ukufumanisa ukuba abantu abathenga iipampers badla nokuthenga nebhiya.
Ukufunda okunzulu (Deep Learning)
Deep Learning
Inxalenye yokufunda ngomatshini esebenzisa iinethiwekhi ze-neural ezinezigaba ezininzi ukumodela imizekelo eyinkimbinkimbi kwidatha.
Umzekelo: Ukufunda okunzulu kusetyenziswa kwiimodeli zolwimi ezifana neGPT-4 kunye neenjini zokudala imifanekiso ezifana neStable Diffusion.
Iimodeli zokuqhawuka (Diffusion Models)
Diffusion Models
Udidi lweemodeli zokudala ezifunda ukudala idatha ngokutshintsha ngokuthe ngcembe ingxolo engacwangciswanga kwiziphumo ezicwangcisiweyo.
Umzekelo: I-Stable Diffusion idala imifanekiso efana neyokwenyani ukusuka kwizicelo zombhalo kusetyenziswa iindlela zokuqhawuka.
Ukubekwa (Embedding)
Embedding
Impendulo yenombolo yedatha, esetyenziswa rhoqo ukubamba intsingiselo echazayo yamagama, imifanekiso, okanye izivakalisi.
Umzekelo: Kwi-NLP, igama elithi 'ibhanki' lingaba nokubekwa okufanayo negama elithi 'imali' kodwa kwahlukile kwelithi 'unxweme lomlambo' kuxhomekeke kwisimo.
I-Epoch (Epoch)
Epoch
Ukuhamba okugcwele phezu kwedatha yokuqeqesha yonke ngexesha lenkqubo yokuqeqesha imodeli yokufunda ngomatshini.
Umzekelo: Ukuba idatha inemizekelo eyi-1,000 kwaye imodeli iyibona yonke kanye ngexesha lokuqeqesha, loo nto yiyeyona epoch.
I-Ethical AI (Ethical AI)
Ethical AI
Ifilosofi yoyilo kunye nokusetyenziswa eqinisekisa ukuba ubuchwepheshe be-AI busebenza ngokucacileyo, ngokulinganayo, kwaye ngokuhambelana nemilinganiselo yoluntu.
Umzekelo: Isixhobo sokuqeshwa kwe-AI esibandakanya ukujongwa kokubandlulula ukuthintela ukubandlulula abaviwa abambalwa.
Inkqubo yengcali (Expert System)
Expert System
Inkqubo ye-AI elinganisa izakhono zokugqiba kwesigqibo sengcali yomntu kwindawo ethile kusetyenziswa imigaqo kunye nengqiqo.
Umzekelo: Inkqubo yengcali esetyenziswa kwezolimo ukucebisa unyango lwezityalo ngokusekelwe kwidatha yomhlaba kunye nembali yezinambuzane.
Ubukrelekrele obuchazekayo (XAI) (Explainable AI (XAI))
Explainable AI (XAI)
Iinkqubo ze-AI ezicwangciswe ukwenza iinkqubo zazo zangaphakathi kunye nezigqibo ziqondwe ngabantu, ukwandisa ukuthembana kunye noxanduva.
Umzekelo: I-AI yokufumanisa iimfihlelo enikezela ingcebiso kuphela kodwa ikwachaza ukuba zeziphi izimpawu ezikhokelele kweso sigqibo.
Ukufunda okune-shot (Few-shot Learning)
Few-shot Learning
Indlela yokufunda ngomatshini apho imodeli iqeqeshwa okanye ilungiswe kusetyenziswa imizekelo embalwa nje enelebula.
Umzekelo: Ukwenza ngokwezifiso i-LLM ukuba ibhale ii-imeyile zomthetho emva kokuyibonisa imizekelo eyi-10 nje.
Ukulungisa (Fine-tuning)
Fine-tuning
Inkqubo yokuthatha imodeli eqeqeshwe ngaphambili kwaye uyibuyisele kwidatha entsha, encinci ukuyenza ibe yeyomsebenzi othile.
Umzekelo: Ukulungisa i-LLM jikelele efana neGPT kwidatha yomthetho yangaphakathi ukudala uncedisi lokubhala umthetho.
Imizobo yesiseko (Foundation Model)
Foundation Model
Imodeli enkulu eqeqeshwe kwidatha eyahlukeneyo kunye nebanzi enokuthi ilungiswe kwimisebenzi emininzi.
Umzekelo: I-GPT-4 kunye ne-PaLM 2 yimizobo yesiseko enako ukufingqa, ukubuzwa nokuphendula, ukuguqulela, nokunye.
Ilizwi elingacacanga (Fuzzy Logic)
Fuzzy Logic
Uhlobo lwengqiqo olujongana neenqobo ezilinganiselweyo endaweni yengqiqo eyinyani/engubuxoki (binary), eluncedo ekubalekeni phantsi kobungaqinisekanga.
Umzekelo: Isetyenziswa kwiinkqubo zokulawula imozulu ukulungisa ubushushu ngokusekelwe kwizinto ezingacacanga ezifana 'nokufudumala kancinci' okanye 'ukubanda kakhulu'.
I-Generative Adversarial Network (GAN) (Generative Adversarial Network (GAN))
Generative Adversarial Network (GAN)
Uzakhiwo lwemodeli yokudala apho iinethiwekhi ezimbini—umvelisi kunye nomahluli—zikhathalelana ukuze ziphucule umgangatho wesiphumo.
Umzekelo: Ii-GANs zisetyenziselwa ukudala iividiyo ezingezizo zodumo okanye ukudala imifanekiso yemveliso efana neyokwenyani ukusuka kwimidwebo.
I-Generative AI (Generative AI)
Generative AI
Udidi lobukrelekrele bokwenziwa olunako ukudala umxholo omtsha—njengombhalo, imifanekiso, umculo, okanye ividiyo—ukusuka kwidatha yokuqeqesha.
Umzekelo: I-ChatGPT idala izithuba zebhlog okanye i-Midjourney idala ubugcisa bedijithali ukusuka kwizicelo zombhalo.
I-Generative Pre-trained Transformer (GPT) (Generative Pre-trained Transformer (GPT))
Generative Pre-trained Transformer (GPT)
Udidi lweemodeli ezinkulu zolwimi eziphuhliswe yi-OpenAI ezisebenzisa ubume be-transformer kwaye iqeqeshwe kwidatha eninzi yombhalo ukwenza imisebenzi eyahlukeneyo yolwimi.
Umzekelo: I-GPT-4 iyakwazi ukubhala izincoko, ukuguqulela iilwimi, kunye nokufingqa amaxwebhu kunye nezicelo ezincinci.
Umzobo weGenetic (Genetic Algorithm)
Genetic Algorithm
Indlela yokulungisa ephefumlelweyo yindalo apho izisombululo ziphuhlisa ngokuhamba kwexesha ngokusebenzisa ukuguquka, ukudibanisa, kunye nokukhetha.
Umzekelo: Isetyenziselwa ukuyila ii-algorithms ze-neural network ezisebenzayo ngokulinganisa ukusinda kweyona ilungileyo.
Ukubhanquka (Hallucination)
Hallucination
Ukuveliswa komxholo othembisayo kodwa ongachanekanga ngokwezibhalo okanye ongacacanga yimodeli ye-AI.
Umzekelo: Imfono yolwimi yenza isicatshulwa esingekho okanye ibonelele ngefakto zembali engachanekanga.
I-Heuristic (Heuristic)
Heuristic
Indlela esebenzayo yokusombulula ingxaki engaqinisekisi isisombululo esigqibeleleyo kodwa esanele kwiinjongo ezikhoyo ngoku.
Umzekelo: Ukusebenzisa umgaqo wokubala ixesha lokuhambisa kwinkqubo yokuhambisa i-logistics.
I-Hyperparameter (Hyperparameter)
Hyperparameter
Inqobo yokucwangciswa ethunyelwa ngaphambi kokuqeqesha imodeli yokufunda ngomatshini, njengomlinganiselo wokufunda okanye inani lezigaba.
Umzekelo: Ukulungisa ubungakanani bebhetshi ukusuka kwi-32 ukuya kwi-128 ukuphucula isantya sokuqeqesha kunye nokusebenza komodeli.
Ukubaleka (Inference)
Inference
Inkqubo yokusebenzisa imodeli yokufunda ngomatshini eqeqeshwe ukwenza izibikezelo okanye ukudala iziphumo kwidatha yokufaka entsha.
Umzekelo: Ukusebenzisa imodeli yeGPT elungisiweyo ukubhala ii-imeyile kwiqela lenkxaso yabathengi.
Ukubona injongo (Intent Detection)
Intent Detection
Umsebenzi kukuqonda ulwimi apho inkqubo ibona injongo okanye injongo yomsebenzisi kwisigijimi.
Umzekelo: Kwi-chatbot, ukuqaphela 'Ndifuna ukubhukisha indiza' njengolwazi lokubhukisha uhambo.
I-Internet of Things (IoT) (Internet of Things (IoT))
Internet of Things (IoT)
Inethiwekhi yezixhobo zomzimba ezidibene ezineenzinzwa, isoftware, kunye nobuchwepheshe obunye ukuqokelela nokutshintshiselana ngedatha.
Umzekelo: Ii-thermostat ezihlakaniphile kunye neefriji ezibika idatha yokusetyenziswa kwaye zilungelelanise useto kusetyenziswa uhlalutyo lwe-AI.
Ukuchazwa (Interpretability)
Interpretability
Ubungakanani apho umntu angayiqonda imicimbi yangaphakathi yemodeli yokufunda ngomatshini kunye nenkqubo yayo yokugqiba kwesigqibo.
Umzekelo: Umthi wesigqibo uchaneka ngakumbi kune-neural network enzulu kuba izigqibo zawo ziyalandeleka.
I-Jupyter Notebook (Jupyter Notebook)
Jupyter Notebook
Indawo yokubala evulekileyo, enokusebenzisana apho abasebenzisi bangabhala khona ikhowudi, babonise iziphumo, kwaye babhale uhlalutyo kwisijongano esinye.
Umzekelo: Abaphandi bedatha basebenzisa ii-Jupyter Notebooks ukwenza iimodeli zokufunda ngomatshini kunye nokwabelana ngezona ziphumo.
I-K-Nearest Neighbours (KNN) (K-Nearest Neighbours (KNN))
K-Nearest Neighbours (KNN)
Inkqubo elula, engaphandle komzobo yokufunda ngomatshini esetyenziselwa ukuhlela kunye nokubala. Yenza izigqibo ngokusekelwe kwimizekelo yokuqeqesha esondeleyo kwindawo yezinto.
Umzekelo: Ukuhlela isiqhamo esitsha njengeapile okanye ipere, i-KNN ijonga ukuba zeziphi iziqhamo ezinelebula ezikufutshane ngobume nangombala.
I-Knowledge Graph (Knowledge Graph)
Knowledge Graph
Uzakhiwo lwedatha olusebenzisa ii-nodes kunye nemiqolo ukumodela kunye nokugcina iinkcazo ezidibene zezinto kunye nobudlelwane bazo.
Umzekelo: Ipani yolwazi yeGoogle inikwa amandla yikhowudhi yobudlelwane eyidibanisa izinto ezifana nabantu, iindawo, kunye neziganeko.
Ukulungiswa koMzekelo wokufunda ulwimi (LLMO) (Language Learning Model Optimisation (LLMO))
Language Learning Model Optimisation (LLMO)
Iindlela ezisetyenziselwa ukuphucula ukusebenza, ukusebenza kakuhle, okanye ukulungiswa kweemodeli ezinkulu zolwimi kwimisebenzi ethile okanye iindawo.
Umzekelo: Ukusebenzisa ukubala kunye noqeqesho lwemiyalelo ukulungisa i-LLM kusetyenziso lweshishini.
Umzekelo omkhulu wolwimi (LLM) (Large Language Model (LLM))
Large Language Model (LLM)
Uhlobo lwemodeli yokufunda enzulu eqeqeshwe kwidatha eninzi yombhalo enako ukudala, ukuqonda, kunye nokubaleka ngolwimi lwabantu.
Umzekelo: I-ChatGPT kunye ne-Claude zii-LLM eziqeqeshwe ukuncedisa ekubhaleni, ekubhaleni ikhowudi, nasekubuzeni imibuzo.
Indawo efihliweyo (Latent Space)
Latent Space
Impendulo ephezulu-yobukhulu eyenziweyo apho izinto zokufaka ezifanayo ziqokelelwe kufuphi, zisetyenziswa kwiimodeli zokudala kunye nokubekwa.
Umzekelo: Kwi-generation yemifanekiso, ukulawula indawo efihliweyo kungatshintsha izinto ezifana nokukhanya okanye imvakalelo.
Isantya sokufunda (Learning Rate)
Learning Rate
I-hyperparameter ephambili kuqeqesho elawula ukuba ubunzima bemodeli buhlaziywa kangakanani ngokubhekiselele kwigradiyenti yelahleko.
Umzekelo: Isantya esiphezulu sokufunda singakhokelela ekugqithiseni iindawo ezisezantsi, ngelixa isantya esiphantsi kakhulu sicothisa inkqubela phambili yokuqeqesha.
Ukufunda ngomatshini (ML) (Machine Learning (ML))
Machine Learning (ML)
Icandelo le-AI elivumela iinkqubo ukuba zifunde kwidatha kwaye ziphucule ukusebenza ngaphandle kokucwangciswa ngokucacileyo.
Umzekelo: Ii-spam filters zisebenzisa ukufunda ngomatshini ukuhlela ii-imeyile njenge-spam okanye hayi ngokusekelwe kwimizekelo yangaphambili.
Ukudrift komzobo (Model Drift)
Model Drift
Isiganeko apho ukuchaneka komodeli kwehla ngokuhamba kwexesha ngenxa kotshintsho kwidatha okanye kwindawo.
Umzekelo: Imfono yokufumanisa inkohlakalo iba ngaphantsi ngokuchanekileyo njengoko iindlela zenkohlakalo ziphuhlisa.
Uqeqesho lomzobo (Model Training)
Model Training
Inkqubo yokondla idatha kwimodeli yokufunda ngomatshini kunye nokulungisa iiparameter zayo ukunciphisa impazamo.
Umzekelo: Ukuqeqesha injini yokucebisa kwimbali yokuthenga yomthengi ukucela iimveliso ezintsha.
I-Multimodal AI (Multimodal AI)
Multimodal AI
Iinkqubo ze-AI ezinako ukucubungula kunye nokudibanisa iintlobo ezininzi zedatha ezifana nombhalo, imifanekiso, ilizwi, kunye nevidiyo.
Umzekelo: Imfono efana neGPT-4 Vision enako ukufunda umbhalo kunye nokuchaza imifanekiso ngaxeshanye.
Ukucubungula ulwimi lwendalo (NLP) (Natural Language Processing (NLP))
Natural Language Processing (NLP)
Icandelo le-AI eligxile kudibano phakathi koomatshini kunye neelwimi zabantu (zendalo). Ivumela oomatshini ukuba bafunde, baqonde, kwaye baphendule ngolwimi lwabantu.
Umzekelo: I-NLP isetyenziswa kwiingcebisi zelizwi, ii-apps zokuguqulela ulwimi, kunye nee-chatbots.
Inethiwekhi ye-Neural (Neural Network)
Neural Network
Imodeli yokufunda ngomatshini ephefumlelweyo yisakhiwo sobuchopho bomntu, equka izigaba zeenodi ezidibene (ii-neurons).
Umzekelo: Iinethiwekhi ze-neural zisemva kweemodeli zokufunda ezinzulu ezisetyenziswa kumfanekiso kunye nokuqaphela ilizwi.
Ingxolo (Noise)
Noise
Ulwazi olungacwangciswanga okanye olungafunekiyo kwidatha olunako ukufihla imizekelo ethembisayo kwaye luthintele ukusebenza komodeli.
Umzekelo: Iimpazamo zezinzwa okanye idatha eneenqaku ezingalunganga ingathathwa njengengxolo.
I-Ontology (Ontology)
Ontology
Isakhiwo esicwangcisiweyo esihlela kwaye sichaze ubudlelwane phakathi kweengcamango kwindawo ethile, esetyenziswa rhoqo kwiinkqubo ze-AI ezichazayo.
Umzekelo: I-ontology kwimpilo ingachaza indlela izimpawu ezidibene ngayo nezifo kunye nonyango.
Ukugqithisa (Overfitting)
Overfitting
Impazamo yokumodela apho imodeli yokufunda ngomatshini ibamba ingxolo kwidatha yokuqeqesha kwaye yenza kakubi kwidatha entsha.
Umzekelo: Imfono egcina iimpendulo zokuqeqesha kodwa ayinakukwazi ukujongana nedatha engabonwanga yimodelayo igqithisile.
Ukuqikelela izinto (Predictive Analytics)
Predictive Analytics
Ukusetyenziswa kwedatha, ii-algorithms, kunye ne-AI ukufumanisa ukufana kweziphumo zexesha elizayo ngokusekelwe kwidatha yembali.
Umzekelo: Abathengisi basebenzisa ukuqikelela izinto ukubala imfuno yeemveliso ezithile.
Uqeqesho lwangaphambili (Pre-training)
Pre-training
Inkqubo yokuqeqesha imodeli ekuqaleni kwidatha enkulu, jikelele ngaphambi kokuyilungisa kwimisebenzi ethile.
Umzekelo: Iimodeli zeGPT ziqeqeshwa ngaphambili kwii-corpora ezinkulu ngaphambi kokuba zenzelwe ii-chatbots zenkxaso yabathengi.
Ubunjineli be-Prompt (Prompt Engineering)
Prompt Engineering
Ubugcisa kunye nesayensi yokwenza izicelo ezisebenzayo ukulawula iziphumo zeemodeli ezinkulu zolwimi.
Umzekelo: Ukongeza imiyalelo yenkqubo efana 'Ndiphendule njengomfundisi ohloniphekileyo' ngumzekelo wobunjineli be-prompt.
Ukubala (Quantisation)
Quantisation
Indlela yokunciphisa imodeli ethatha inxaxheba kwiibhitshi ezininzi ezisetyenziselwa ukumela ubunzima kunye nokwenza izinto, ukuphucula ukusebenza kakuhle.
Umzekelo: Ukubala imodeli ukusuka kwi-32-bit ukuya kwi-8-bit kuphucula ukusebenza kwizixhobo eziphathwayo.
Ukubala nge-Quantum (Quantum Computing)
Quantum Computing
Iparadigm entsha yokubala esekelwe kwi-quantum mechanics, eneziphumo zokubala ezikhawulezayo.
Umzekelo: Ukubala nge-quantum kungakhawulezisa uqeqesho lwe-AI ngaphaya kwemida yakudala.
Injini yokubaleka (Reasoning Engine)
Reasoning Engine
Inkqubo kwi-AI eyenza izigqibo ezichanekileyo ukusuka kwiseti yefakto okanye idatha kusetyenziswa imigaqo okanye ii-algorithms zokubaleka.
Umzekelo: Isixhobo sokufumanisa i-AI sisebenzisa injini yokubaleka ukuchaza izifo ezinokwenzeka ngokusekelwe kwizimpawu.
Ukufunda ngokuphindaphinda (RL) (Reinforcement Learning (RL))
Reinforcement Learning (RL)
Inkalo yokufunda ngomatshini apho ii-arhente zifunda ngokusebenzisana nemvelo yazo ukuze zonyuse umvuzo oqokegeleyo.
Umzekelo: Irhorho elifunda ukuhamba ngokuvavanya kunye nokubona kusetyenziswa iindlela ze-RL.
Ukufunda ngokuphindaphinda kunye neNkxaso yoMntu (RLHF) (Reinforcement Learning with Human Feedback (RLHF))
Reinforcement Learning with Human Feedback (RLHF)
Indlela yokufunda apho ukuthanda komntu kulawula isignali yomvuzo ye-AI, esetyenziswa rhoqo kulungiso lweemodeli zolwimi.
Umzekelo: I-ChatGPT yaqeqeshwa nge-RLHF ukuvelisa iziphumo eziluncedo nezikhuselekileyo.
Ukubuyisela-Augmented Generation (RAG) (Retrieval-Augmented Generation (RAG))
Retrieval-Augmented Generation (RAG)
Indlela edibanisa ukubuyisela ulwazi kunye nokudala, apho i-LLM ifumana amaxwebhu afanelekileyo ukuphucula impendulo yayo.
Umzekelo: Uncedisi we-AI ufumana kwaye ubhekiselele kwizinto zemveliso ngelixa udala impendulo kumbuzo wobuchwepheshe.
Ukufunda okuzenzekelayo (Self-Supervised Learning)
Self-Supervised Learning
Indlela yokuqeqesha apho imodeli ifunda imizekelo ngokudala iilebula zayo ezivela kwidatha ekrwada, inciphisa ukuxhomekeka kwidatha enelebula yomntu.
Umzekelo: I-BERT iqeqeshwe ngokufunda okuzenzekelayo ngokubikezela amagama alahlekileyo kumbhalo.
Ukukhangela okuchazayo (Semantic Search)
Semantic Search
Indlela yokukhangela yokuqonda injongo yomsebenzisi kunye nentsingiselo yesimo, hayi nje ukufanisa amagama angundoqo.
Umzekelo: Ukukhangela 'indlela yokulungisa iteyiphu evuzayo' kubonisa izikhokelo nokuba igama elithi 'itheyiphu evuzayo' alikho kuxwebhu.
Ukuhlalutya imvakalelo (Sentiment Analysis)
Sentiment Analysis
Inkqubo yokufumana iimvakalelo, izimvo, okanye izimo zengqondo kumbhalo, rhoqo ukuhlela njenge pozitiv, negative, okanye neutral.
Umzekelo: Ukuhlalutya ii-tweets ukuqonda ukusabela koluntu kwimveliso entsha.
I-Stochastic (Stochastic)
Stochastic
Ukubandakanya ukungahambi kakuhle okanye ukuziphatha okubalulekileyo, kusetyenziswa rhoqo kwi-generative AI kunye nee-algorithms zokulungisa.
Umzekelo: Isiphumo seGPT-4 siyahluka kwisicelo esifanayo ngenxa yenkqubo yayo yokuchaza i-stochastic.
I-Strong AI (Strong AI)
Strong AI
Kwakhona ibizwa ngokuba yi-Artificial General Intelligence (AGI), ibhekisa koomatshini abanezakhono zokucinga ezilingana nezinga lomntu kuzo zonke iindawo.
Umzekelo: I-AI yexesha elizayo enokuthi ngokuzimeleyo ibhale amanoveli, icwangcise izixeko, kwaye isombulule iingxaki zokuziphatha ngokufanayo.
Ubukrelekrele obungaphezulu bokwenziwa (SAI) (Super Artificial Intelligence (SAI))
Super Artificial Intelligence (SAI)
I-AI ethathwa njengengaphezulu kakhulu kubukrelekrele bomntu kuzo zonke iinkalo—ukubaleka, ubuchule, ubukrelekrele beemvakalelo, njl.njl.
Umzekelo: I-SAI ingakwazi ukudala iinzululwazi ezintsha kunye neefilosofiya ngokuzimeleyo.
Ukufunda okubekwe esweni (Supervised Learning)
Supervised Learning
Indlela yokufunda ngomatshini apho iimodeli ziqeqeshwe kwidatha enelebula ukuze zifunde ukudibanisa okufakwayo-okuphumayo.
Umzekelo: Ukufundisa imodeli ukuhlela ii-imeyile njenge-spam okanye hayi kusetyenziswa imizekelo yembali.
Idatha eyenziweyo (Synthetic Data)
Synthetic Data
Idatha eyenziweyo ngokwenziwa eyalinganisa idatha yehlabathi lokwenyani, esetyenziswa rhoqo ukuqeqesha xa idatha yokwenyani ingabikho okanye ibuthathaka.
Umzekelo: Ukudala imifanekiso yezonyango eyenziweyo ukuqeqesha iimodeli zokufumanisa ngaphandle kokwaphula imfihlo yomguli.
I-Token (Token)
Token
Iyunithi yombhalo ecubungulwe yi-LLM—ngokwesiqhelo ilizwi okanye isiqwenga selizwi.
Umzekelo: Isivakalisi esithi 'Hello world!' sihlulwe saba zii-tokens ezi-3: 'Hello', 'world', kunye '!'.
Ukubeka i-token (Tokenisation)
Tokenisation
Inkqubo yokwahlula umbhalo kwiitokeni ukuze zicubungulwe yimodeli.
Umzekelo: Kwi-NLP, 'I-ChatGPT inkulu' iba ngama-['Chat', 'G', 'PT', 'is', 'great'].
Ukufunda ngokuDluliselwa (Transfer Learning)
Transfer Learning
Ukusebenzisa ulwazi ukusuka komnye umsebenzi ukuze kuphuculwe ukufunda komnye umsebenzi ohambelanayo, kunciphisa ixesha lokuqeqesha kunye neemfuno zedatha.
Umzekelo: Ukulungisa imodeli eqeqeshwe kumbhalo wesiNgesi ukwenza uhlalutyo lwemvakalelo kolunye ulwimi.
I-Transformer (Transformer)
Transformer
Uzakhiwo lwe-neural network olusebenzisa izixhobo zokuqwalasela ukumodela idatha elandelelanayo, esetyenziswa kakhulu kwi-LLM.
Umzekelo: I-BERT, i-GPT, kunye ne-T5 zonke ziimodeli ezisekelwe kwi-transformer.
Ukungagqithisi (Underfitting)
Underfitting
Xa imodeli ilula kakhulu ukubamba imizekelo kwidatha yokuqeqesha, ibangela ukusebenza okungalunganga.
Umzekelo: Imfono eline-linear ezama ukubala ukubala kwemifanekiso eyinkimbinkimbi ingase ingagqithisi.
Ukufunda okungabhekiswanga (Unsupervised Learning)
Unsupervised Learning
Indlela yokufunda apho iimodeli zifumana imizekelo okanye ii-clusters kwidatha engenelebula.
Umzekelo: Ukudibanisa abathengi ngokusekelwe kwindlela yokuthenga ngaphandle kweelebula ezichaziweyo.
Injongo yomsebenzisi (User Intent)
User Intent
Injongo okanye injongo ngasemva kwesicelo okanye unxibelelwano lomsebenzisi.
Umzekelo: Umsebenzisi obhala 'indlela yokubhaka ikhekhe' mhlawumbi ujonge ukufumana iresiphi.
Isampuli yokuqinisekisa (Validation Set)
Validation Set
Isiqwenga sedatha esisetyenziselwa ukuvavanya ukusebenza komodeli ngexesha lokuqeqesha kunye nokulungisa ii-hyperparameters.
Umzekelo: Isetyenziselwa ukufumanisa ukugqithisa ngaphambi kovavanyo lokugqibela.
Idathabhesi yeVector (Vector Database)
Vector Database
Idathabhesi eyenzelwe ukugcina kunye nokukhangela ii-vector embeddings ezisetyenziswa kwimisebenzi ye-AI efana nokukhangela okufanayo kunye ne-RAG.
Umzekelo: I-Pinecone kunye ne-Weaviate ziziidathabhesi ze-vector zokugcina i-text okanye i-embedding zemifanekiso.
Ukubekwa kweVector (Vector Embedding)
Vector Embedding
Impendulo yenombolo yedatha egcina intsingiselo kunye nobudlelwane kwindawo ye-vector.
Umzekelo: Amagama athi 'ukumkani' kunye 'ne-queen' anokubekwa okufanayo kunye neenguqulelo zobulili ezicokisekileyo.
Uncedisi oluBonakalayo (Virtual Assistant)
Virtual Assistant
Ubuncedisi be-AI obunika amandla isoftware enceda abasebenzisi ukugqibezela imisebenzi ngokusebenzisa unxibelelwano okanye imiyalelo yelizwi.
Umzekelo: USiri, uAlexa, kunye noGoogle Assistant ngabancedisi ababonakalayo abadumileyo.
Ukuqaphela ilizwi (Voice Recognition)
Voice Recognition
Ubuchwepheshe obuchaza kwaye buguqula ulwimi oluthethwayo lube ngumbhalo okanye isenzo.
Umzekelo: Ukuchwetheza ngelizwi kunye nemiyalelo yelizwi ixhomekeke kwiinkqubo zokuqaphela ilizwi.
I-Weak AI (Weak AI)
Weak AI
Iinkqubo ze-AI ezicwangciswe ukwenza umsebenzi othile, ongabonakaliyo ngaphandle kobukrelekrele obubanzi.
Umzekelo: I-AI yokudlala itshosi engekho nako ukuqonda ulwimi okanye ukuhambisa imoto yimzekelo ye-AI ebuthathaka.
Ukukhangela iWebhu (Web Scraping)
Web Scraping
Ukukhutshwa okuzenzekelayo kolwazi kwiindawo zokubeka, kusetyenziswa rhoqo ukuqokelela idatha yokuqeqesha okanye ukujonga umxholo.
Umzekelo: Ukukhangela uluhlu lweendawo zokuhlala ukuqeqesha imodeli yokuxabisa impahla.
Ubunzima (Weight)
Weight
Ipharameter kwiinethiwekhi ze-neural ezichaza amandla othintelo olunye i-node olunalo kwenye.
Umzekelo: Ubunzima bulungiswa ngexesha lokuqeqesha ukunciphisa impazamo yomzekelo.
I-Whisper (Whisper)
Whisper
Imodeli yokuthetha-kwisicatshana eyenziwe yi-OpenAI enako ukubhala izandi ngeelwimi ezininzi.
Umzekelo: I-Whisper ingabhala iintetho kunye nee-podcasts ngokuchaneka okuphezulu.
I-YAML (YAML)
YAML
Ifomathi efundeka ngabantu yokubeka idatha, esetyenziswa rhoqo kwii-files zokucwangciswa kwiinkqubo zokufunda ngomatshini.
Umzekelo: Ukuchaza ii-parameters zomzekelo kwifayile ye-YAML yokuqeqesha kwi-PyTorch.
Ukufunda okune-zero-shot (Zero-shot Learning)
Zero-shot Learning
Ubume bemodeli bokwenza imisebenzi engazange iqeqeshwe ngayo ngokucacileyo ngokusebenzisa ulwazi olubanzi.
Umzekelo: Imfono iphendula imibuzo yomthetho nangona ingaqeqeshwanga ngokukhethekileyo kwidatha yomthetho.
I-Zettabyte (Zettabyte)
Zettabyte
Iyunithi yedatha yedijithali elingana ne-sextillion (10^21) bytes, esetyenziswa rhoqo ukuchaza ubungakanani bedatha ye-intanethi.
Umzekelo: Umsebenzi we-intanethi yehlabathi udlule i-zettabyte enye ngonyaka ngo-2016.
