Ubuyobozi

Guhindura no kwikosora

Reflexion nubuhanga aho umukozi wa AI agaragaza mukwandika kunanirwa kwayo kandi akagaburira ayo masomo mugerageza gukurikira.

2 min somaIbiherutse kuvugururwa

Incamake

It matters because it lets agents improve on a task without retraining the underlying model.

Kwibira cyane

Reflexion, yatangijwe mu mpapuro 2023 na Shinn na bagenzi be, iha umukozi umuzingo: igerageza umurimo, yakira ikimenyetso cyukuntu yakoze (ibisubizo byikizamini, ibihembo, cyangwa kunegura), hanyuma yandika imvugo ngufi 'imvugo' yerekana ururimi rusanzwe rusobanura ibitaragenze neza nibigomba gukurikiraho. Ibyo bitekerezo bibikwa murwibutso kandi byateganijwe kubutaha bwo kugerageza. Icy'ingenzi, uburemere bw'icyitegererezo ntibwigera buhinduka; kwiga bibaho rwose mumadirishya yimiterere nkinyandiko. Uku 'kwiga gushimangira amagambo' kureka abakozi bagasubiramo ibibazo bya code, kugendagenda kurubuga, hamwe nimirimo yo gutekereza. Ku gipimo ngenderwaho cya code ya HumanEval, uburyo bwo kwikosora bwifashishije uburyo bwo kwikosora bwasunikishije igipimo cyatsinzwe cyane kuruta kugerageza inshuro imwe, gusa kureka umukozi agakosora amakosa yacyo mugihe gito.

Ubushishozi

Reflexion itandukanya inshingano eshatu: Umukinnyi utanga ibikorwa, Isuzuma ryerekana amanota (ibizamini byibice, igenzura rihuye neza, cyangwa umucamanza wa LLM), hamwe na Moderi yo Kwiyerekana ihindura ayo manota isomo ryamasomo. Isomo ryaguye muri episodic yibuka buffer yongeye kugeragezwa ubutaha. Kuberako ibitekerezo ari ururimi aho kuba gradients, ntamahugurwa ya GPU arakenewe, ariko biterwa cyane nikimenyetso cyizewe cyo kwirinda kugirango ushimangire ibitekerezo byizewe ariko bitari byo.

Ingaruka z'Ingamba

Build choices

Igishushanyo-cy-urwego rugena niba AI itezimbere ibisubizo nyabyo.

Itsinda hamwe nakazi

Guhuza ibikorwa byiza bikora umusaruro wunguka abakoresha bashobora kwizera.

Risk and safety

Gukoresha neza ibibazo bigabanya umunaniro wimpinduka hamwe ningaruka zo gushyira mubikorwa.

Ejo hazaza h'ibitekerezo no kwikosora

Kwikosora bigenda bihinduka urwego rwibanze aho kuba amayeri yubushakashatsi. Witegereze kwishyira hamwe hamwe no kugenzura byikora, nka kode yumusenyi, agasanduku gasanzwe, hamwe no kugarura ibintu byemeza ukuri, bityo ibitekerezo bigashingira kubimenyetso bifatika aho kwerekana icyitegererezo ubwa kabiri. Inzitizi zifunguye ni ukwirinda aho umukozi 'akosora' umusaruro udashira, guhitamo igihe cyo guhagarika gusubiramo, no kubuza gutekereza gutembera mu buryo bwumvikana-butumvikana ariko butemewe.

Gushyira mu bikorwa Isi

Umukozi wa coding ukora ibizamini, asoma ibyananiranye, yandika inyandiko kuri bug, kandi ahindura code yayo mbere yo kongera gukora suite.

Umufasha wubushakashatsi ufata ibyemezo bya salusiyo mugihe igenzura ryananiwe, hanyuma rivugurura igisubizo cyo gukoresha amasoko yemewe gusa.

Urubuga-rwogukoresha (urugero, kurwego rwa AlfWorld cyangwa Urubuga rwa WebShop) rwandika 'Nakanze muyunguruzi rutari rwo' kandi rukirinda iryo kosa ryo gusubiramo.

Imibare-ikemura ibibazo igenzura igisubizo cyayo cyanyuma ikabuza imbogamizi, ikabona ikosa ryikimenyetso, kandi igakora intambwe ijyanye.

Ingaruka & Kurinda

Gutangiza inzira yamenetse birashobora kongera ibibazo bihari.

Amakipe arashobora gukora cyane kandi agakuraho ibitekerezo byabantu bikenewe.

Ubwiza burashobora gutemba niba ibisubizo bidahwema gusuzumwa.

Igishushanyo mbonera

1

Shushanya ibikorwa byubu hanyuma umenye intambwe-yo guterana hejuru.

2

Sobanura aho abantu bagenzura mbere yo kwikora byuzuye.

3

Hugura abakoresha kubisobanuro, inzira zo kuzamuka, hamwe nubuziranenge.

4

Kurikirana ibisubizo-urwego rwibisubizo kugirango wemeze agaciro karambye.

Komeza Ubushakashatsi

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.

Tangira ikibazo

Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation

Ubuyobozi bukurikira

Kwiyerekana wenyine mubakozi

Ibibazo bikunze kubazwa

What is Reflexion and Self-Correcting Agents?

Reflexion nubuhanga aho umukozi wa AI agaragaza mukwandika kunanirwa kwayo kandi akagaburira ayo masomo mugerageza gukurikira. Nibyingenzi kuko ireka abakozi bakitezimbere kumurimo batongeye kwitoza icyitegererezo.

Ni ubuhe busobanuro buranga uburyo umukozi wa Reflexion 'yiga'?

Reflexion rimwe na rimwe yitwa 'amagambo yo gushimangira amagambo' kubera ko amasomo abikwa nk'inyandiko-mvugo karemano yibuka, ntabwo yashyizwe mubipimo by'icyitegererezo.

Muburyo bwa Reflexion, Evaluator ikora iki?

Isuzuma (ikizamini cyibice, igenzura-rihuye neza, cyangwa umucamanza wa LLM) ritanga ikimenyetso cyuko intambwe yo Kwiyerekana ihinduka isomo ryanditse.

Ni ukubera iki ubwiza bwibimenyetso byo gusuzuma bifite akamaro kanini muri Reflexion?

Niba abasesengura batizewe, umukozi ashobora kwandika ibitekerezo byizewe ashingiye kubitekerezo bibi, kuyobora ibizagerwaho mugihe kibi.

Ni ibihe bipimo Reflexion-yuburyo bwo kwikosora yatezimbere cyane igipimo cya code?

HumanEval ni igipimo cyerekana ibisekuruza, kandi ureke umukozi agerageza kugerageza inshuro nyinshi yazamuye igipimo cyiza hejuru yimikorere imwe.

Niyihe ngaruka zifunguye zifatika zo kwikosora?

Hatabayeho itegeko ryiza ryo guhagarika, umukozi arashobora gukomeza gusubiramo ibisubizo nyabyo, guta compte kandi rimwe na rimwe bitesha agaciro umusaruro mwiza.