01
I-AI Foundations
Qonda ukuthi iyini i-AI, ukuthi amasistimu afunda kanjani, lapho ehluleka khona, kanye nendlela yokwahlulela izimangalo ngaphandle kokukhohlisa.
Umtapo wezincwadi we-AI wamahhala
Ubuchwepheshe abaqondisiMahhala phakade.
177 imihlahlandlela yesiNgisi esicacile, izindlela zokufunda ezihlelekile, nomtapo wolwazi ovulekile — wakhiwe inhlangano engenzi nzuzo ye-501(c)(3) ezimele ukuze noma ubani aqonde i-AI yesimanje.
177Iziqondiso zamahhala
1Amathrekhi esihloko
~2 minNgomhlahlandlela
~6hIsikhathi sokufunda
Qala lapha
Ezinhlanu izifundo ezisekelwe emiphumeleni
Isifundo ngasinye sihlanganisa imiphumela esobala, amakhono afakwe kumephu, imisebenzi yokuzilolonga, kanye netshe eliyinhloko elisetshenzisiwe.
02
Umsebenzisi we-AI onesibophezelo
Sebenzisa i-AI ngendlela ekhiqizayo kuyilapho uvikela ubumfihlo, ubheka imiphumela, futhi ulondoloza ukuziphendulela komuntu.
03
I-AI emsebenzini
Linganisa izimo zokusetshenziswa kwendawo yokusebenza, sebenzisa abashayeli bezindiza abaphephile, kala inani, futhi uxhumane nezinguquko ngokuzibophezela.
04
Inqubomgomo ye-AI Nomphakathi
Hlaziya izinhlelo ze-AI ngokusebenzisa amalungelo, ukulingana, ukubusa, ukuphepha, kanye nemiphumela yenzuzo yomphakathi.
05
Ukwakha nge-AI Systems
Qonda amamodeli olimi, ukubuyisa, abasebenzeli, ukuhlola, izindleko, nokuvikela ukusetshenziswa ngokuklama kwesistimu okusebenzayo.
Amathrekhi esihloko
Phequlula ngethrekhi
Gxumela endaweni oyikhathalelayo. Wonke amathrekhi anemihlahlandlela eminingi yesiNgisi esilula.
OkuyisisekeloImiqondo eyinhloko: ukuthi i-AI isebenza kanjani ezingeni eliphezulu.84 abaqondisi Ulimi lwe-AIAma-LLM, i-NLP, i-ChatGPT, ukwaziswa, nokuqonda umbhalo.155 abaqondisi I-AI ebonakalayoUkwakhiwa kwesithombe, umbono wekhompyutha, ividiyo, kanye nomklamo.133 abaqondisi Umsindo we-AIIzwi, umculo, i-podcasting, kanye ne-AI yokuqala yomsindo.117 abaqondisi IzinkampaniAmaphrofayili we-OpenAI, Anthropic, Google, Meta, nokunye.121 abaqondisi Izinhlelo zokusebenzaIvela kanjani i-AI ekubhaleni amakhodi, usesho, ama-ejenti, nempilo yansuku zonke.111 abaqondisi IzimboniI-AI kwezokunakekelwa kwempilo, ezezimali, imfundo, amandla, nokuningi.91 abaqondisi UbuchwephesheAma-chips, ukulungisa kahle, i-RAG, ama-vector, nengqalasizinda ye-AI.177 abaqondisi UmphakathiUkuchema, ubumfihlo, i-copyright, ukuphepha, nekusasa le-AI.30 abaqondisi
Umtapo wolwazi ogcwele
Zonke iziqondiso
177 kwe 1019 imihlahlandlela ebonisiwe. Hlunga ngethrekhi noma sesha ngenhla.
Ubuchwepheshe
Ukusakaza-bukhoma okuqagelayo kanye nokubikezela kwamathokheni amaningi
Ukusakaza-bukhoma okuqagelayo nokubikezela kwamathokheni amaningi kusheshisa ukwakhiwa kwemodeli yolimi ngokuqagela amathokheni ambalwa esikhathi esizayo kanye nokuwaqinisekisa ngasikhathi sinye...
UbuchwephesheI-BERTScore kanye nokuhlola kweSemantic
I-BERTScore ikala ukuthi umbhalo okhiqizwe kahle ufanelana kanjani nesithenjwa ngokuqhathanisa incazelo, hhayi amagama aqondile.
UbuchwephesheUkunakwa Komugqa kanye Nezinhlamvu Zomdlali
Ukunakwa komugqa kungena esikhundleni sokunaka kwe-quadratic softmax kuma-Transformers ngeqhinga lezibalo elikala ngokulandelana ngobude bokulandelana.
UbuchwephesheIsandiso sobude be-YaRN kanye Nokuqukethwe
I-YaRN (Nokho enye isandiso se-RoPE) iyindlela ephumelelayo yokwelula iwindi lomongo elisebenzisekayo lemodeli lize lidlule lapho liqeqeshwe khona.
UbuchwephesheUkuhunyushwa Kwesikhundla Komongo Omude
I-Positional Interpolation (PI) iyindlela elula, enethonya enweba iwindi lomongo we-Transformer ngokucindezela izinkomba zendawo entsha kububanzi...
UbuchwephesheIngxube Yochwepheshe be-LoRA
Ingxube Yezazi Ze-LoRA (i-MoLE) ihlanganisa ama-adaptha amaningi amancane, aqeqeshwe ngokushibhile nerutha efundiwe ukuze imodeli eyodwa yesisekelo ikwazi ukusebenza ngokukhethekile kuyo yonke indawo…
UbuchwephesheUkuhlela Okuqagelwayo Kwamamodeli Ekhodi
Ukuhlela okuqagelayo kwenza ukuhlela ikhodi ye-AI kuzwakale ngokushesha ngokubikezela ukuthi iningi lefayela lizohlala lingashintshiwe futhi liqinisekisa izingxenye ezincane kuphela ezihlukile.
UbuchwephesheI-Activation Steering kanye Nobunjiniyela Bokumela
Isiqondiso sokwenza kusebenze sigudluza impatho yemodeli ngokwengeza ngokuqondile noma ngokukhipha ama-vectors ngaphakathi kokwenza kusebenze okufihliwe ngesikhathi sokusebenza, akukho ukuziqeqesha kabusha okudingekayo.
UbuchwephesheSparse Autoencoder for Ukutolika
Ama-Sparse autoencoder (ama-SAE) iyithuluzi elihlukanisa ukusebenza kwangaphakathi okuphithene kwenethiwekhi ye-neural ibe yisethi enkulu yokuhlanza...
UbuchwephesheI-Superposition kanye ne-Polysemanticity ku-AI Ukutolika
Funda i-superposition kanye ne-polysemanticity ekuchazekeni kwe-AI: kungani amamodeli epakisha izici eziqondisweni ezabiwe, kungani ama-neurons eba i-polysemantic…
UbuchwephesheI-Logit Lens kanye ne-Tuned Lens
Ilensi yelogithi namalensi ashuniwe kungamasu okutolika abheka ungqimba lwezimo ezifihliwe ze-transformer ngongqimba ukuze abone ukuthi imodeli 'icabangani'...
UbuchwephesheI-DeepSpeed kanye ne-Megatron Training Stacks
I-DeepSpeed (Microsoft) ne-Megatron-LM (NVIDIA) izitaki zesofthiwe ezenza amamodeli okuqeqesha anezigidigidi zamapharamitha ezinkulungwaneni zama-GPU...
Uqedile ukufunda? Ake usikhombise sibone.
Hlola okufundile ngombuzo wesihloko, bese uhlola izifundo zethu ezihlelekile noma usebenzela ukuthola isitifiketi. Wonke umhlahlandlela uhlala ukhululekile ukufunda.