Ntugharị na ndị na-emezi onwe ha
Ntugharị bụ usoro ebe onye ọrụ AI na-atụgharị uche n'ide ihe na ọdịda nke ya ma na-azụghachi nkuzi ndị ahụ na mbọ ya ọzọ.
Nchịkọta
It matters because it lets agents improve on a task without retraining the underlying model.
Ime miri emi
Ntugharị, ewepụtara na akwụkwọ 2023 nke Shinn na ndị ọrụ ibe ya, na-enye onye nnọchi anya akaghị: ọ na-anwa ọrụ, na-anata akara gbasara otu o siri mee (nnwale ule, ụgwọ ọrụ, ma ọ bụ nkatọ), wee dee ‘ntụgharị uche’ dị nkenke nke asụsụ okike na-akọwa ihe na-ezighi ezi na ihe ị ga-anwale ọzọ. A na-echekwa echiche ahụ na ebe nchekwa ma kwadoo ngwa ngwa ọzọ. N'ụzọ dị mkpa, ihe atụ nke ihe nlereanya adịghị agbanwe agbanwe; mmụta na-eme kpam kpam na mpio okirikiri dị ka ederede. 'Mmụta nkwado okwu' a na-eme ka ndị ọrụ na-atụgharị uche na nsogbu nzuzo, igodo webụ, na ọrụ ịtụgharị uche. Na akara nrịbama nke HumanEval, mmezi onwe onye n'ụdị reflexion kpaliri ọnụego ngafe dị elu karịa mbọ otu agba, naanị site n'ikwe ka onye ọrụ mebie mmejọ nke ya n'ofe ole na ole.
Nghọta nka nka
Ntugharị na-ekewa ọrụ atọ: Onye na-eme ihe nkiri na-emepụta omume, Onye nyocha nke na-enyocha nsonaazụ ya (nnwale otu, nlele egwuregwu kpọmkwem, ma ọ bụ onye ọka ikpe LLM), na ihe nlere onwe onye na-atụgharị akara ahụ ka ọ bụrụ nkuzi ederede. Ihe mmụta a na-adaba na ebe nchekwa ebe nchekwa etinyegharịrị na nnwale ọzọ. N'ihi na nzaghachi bụ asụsụ kama gradients, ọ dịghị ọzụzụ GPU achọrọ, mana ọ dabere na mgbama nyocha a pụrụ ịdabere na ya iji zere ime ka ntụkwasị obi sie ike mana echiche na-ezighi ezi.
Mmetụta atụmatụ
Mee nhọrọ
Nhazi ọkwa-ngwa na-ekpebi ma AI ọ na-eme ka ezigbo nsonaazụ.
Team na usoro ọrụ
Ngwakọta arụmọrụ dị mma na-emepụta uru nrụpụta ọrụ ndị ọrụ nwere ike ịtụkwasị obi.
Ihe ize ndụ na nchekwa
Usoro eji eme ihe nke ọma na-ebelata ike ọgwụgwụ mgbanwe na ihe ize ndụ mmejuputa.
Ọdịnihu nke ntụgharị uche na ndị na-edozi onwe ha
Ndozi onwe onye na-aghọ oyi akwa ndabara na usoro ndị nnọchi anya karịa aghụghọ nyocha. Na-atụ anya mwekota siri ike na ihe nnwapụta akpaghị aka, dị ka igbe mkpuchi koodu, ndị na-enyocha akwụkwọ, na iweghachite nke na-akwado eziokwu, yabụ ntughari dabere na akara ebumnobi kama ịche ụdị n'onwe ya. Ihe ịma aka ndị mepere emepe na-ezere loops ebe onye nnọchi anya na-emezi nrụpụta ọrụ na-enweghị ngwụcha, na-ekpebi mgbe ị ga-akwụsị ikwugharị, yana igbochi ntụgharị uche ịbanye n'ime ụda na-ada ụda mana enweghị nkwenye.
Mmejuputa n'ezie n'ụwa
Onye na-ahụ maka koodu na-eme ule nkeji, na-agụ nkwupụta dara ada, na-ede ndetu na ahụhụ ahụ, ma dezie koodu ya tupu ịmalitegharịa ụlọ ọrụ ahụ.
Onye enyemaka nyocha nke na-enweta nkwubi okwu nhụsianya mgbe nlele iweghachite dara, wee degharịa azịza ya iji naanị isi mmalite enwetara.
Onye na-ahụ maka ihe nchọgharị weebụ (dịka, na AlfWorld ma ọ bụ WebShop benchmarks) nke na-edekọ 'm pịrị nzacha na-ezighi ezi' wee zere nzọụkwụ ahụ na-ezighi ezi na ịnwale.
Onye na-edozi nsogbu mgbakọ na mwepụ nke na-enyocha azịza ikpeazụ ya megide mmachi, na-ahụ mperi akara, wee rụgharịa nzọụkwụ dị mkpa.
Ihe ize ndụ & okporo ụzọ nche
Ime ka usoro gbajiri agbaji nwere ike ịbawanye nsogbu ndị dị adị.
Otu dị iche iche nwere ike megharịa ma wepụ ikpe mmadụ chọrọ.
Ogo nwere ike ịfegharị ma ọ bụrụ na enyochaghị nsonaazụ ya.
Map mmejuputa
Map usoro ọrụ dị ugbu a wee chọpụta usoro mgbagha kachasị elu.
Kọwaa ebe nlele mmadụ tupu akpaaka zuru oke.
Zụlite ndị ọrụ na mkpali, ụzọ mmụba, na ụkpụrụ ịdị mma.
Soro nsonaazụ ọkwa-ọrụ iji kwado uru na-adịgide adịgide.
Nọgide na-eme nchọpụta
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Reflexion and Self-Correcting Agents quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Ntuziaka na-esote
Ntụle onwe onye na Loops Agent
Ajụjụ a na-ajụkarị
What is Reflexion and Self-Correcting Agents?
Ntugharị bụ usoro ebe onye ọrụ AI na-atụgharị uche n'ide ihe na ọdịda nke ya ma na-azụghachi nkuzi ndị ahụ na mbọ ya ọzọ. Ọ dị mkpa n'ihi na ọ na-ahapụ ndị nnọchi anya imeziwanye otu ọrụ na-azụghachighị ihe nlereanya dị n'okpuru.
Gịnị bụ njirimara na-akọwapụta ka onye nnọchi anya Reflexion si 'amụta'?
A na-akpọ ntugharị mgbe ụfọdụ 'mmụta nkwado okwu' n'ihi na a na-echekwa nkuzi dị ka ederede asụsụ okike na ebe nchekwa, ọ bụghị itinye n'ime akara ngosi.
N'ime usoro ntụgharị uche, gịnị ka onye nleba anya na-eme?
Onye nnwale (nnwale otu, nlele egwuregwu kpọmkwem, ma ọ bụ onye ọka ikpe LLM) na-ewepụta akara na nzọụkwụ ntụgharị uche onwe na-atụgharị ghọọ nkuzi ederede.
Kedu ihe kpatara njirimara nlebanya ji dị mkpa na Ntugharị?
Ọ bụrụ na onye na-enyocha ya enweghị ntụkwasị obi, onye ọrụ ahụ nwere ike dee echiche obi ike dabere na nzaghachi ọjọọ, na-eduzi mbọ n'ọdịnihu n'ụzọ na-ezighi ezi.
Kedu ihe nrịbama ka mmezi onwe nke ụdị reflexion ji kwalite ọnụego ngafe koodu?
HumanEval bụ akara nrịbama-ọgbọ, yana ikwe ka onye nnọchi anya mebie n'ọtụtụ nnwale welitere ọnụego ngafe karịa arụmọrụ otu agba.
Kedu ihe bụ ezigbo ihe egwu mepere emepe na ndị ọrụ na-agbazi onwe ha?
Na-enweghị ezigbo usoro nkwụsị, onye nnọchi anya nwere ike ịga n'ihu na-emegharị azịza ziri ezi, na-efunahụ mgbakọ na mwepụ mgbe ụfọdụ na-eweda ezigbo mmepụta.