Ci bi mu melGis bi ci topp
Bisu dagg xam-xam ci LLMs
IA làkk
Làkk AI GUIDE
LLMs
Bañ bu ñu delloo ap anam bu yeeme la ndax xeetu lakk bu jang 'A moy B' munul tontu ci anam wu wóór 'B moy A.' Dafay wane ni LLM yi dañuy denc mbir yi ni kuréel yu am benn yoon, duñu ni xam-xam bu symétrique.
Déglul tegtal biiIA-nettali · ~2 min
Ci xët wii2 simili jàng
Plongeur bu xóot
Berglund ak ay naataangoom bind nañu ko ci benn këyitu atum 2023, mbañeel bi defay wone ni su benn mannekin taggat ci 'Yaayu Tom Cruise moy Mary Lee Pfeiffer' defay faral di lajj su ñu ko laaje 'Kan mooy doomu Mary Lee Pfeiffer?' doonte tontu li dafay nuru. Efe bi dafay wéy ci tolluwaayu model yi ba ci ginaaw ñu defaree téemeeri mbir yu mel noonu. Du benn bërëb ci fàttaliku: model bi gis na xibaar bi, waaye ci benn yoon kese. Ndax tàggat yaram dafay gëna baaxal xam-xam bi ci topp ci njubluwaay baat yi ci done yi, lëkkalekaay biy tàmbalee ci A ba ci B du sos ci saasi lëkkalekaay bu bawoo ci B ba ci A. Gis-gis bi dafa werante xalaat yi ne escalier kese mooy defar xalaat yu yomb, yu niro ak nit ci kaw dëgg.
Doxal ci àdduna dëgg
Chatbot dafay wax waajur bu siiw ci anam wu jaar yoon, waaye suñu ko laajee tuddee doomu waajur bu siiw bi, du wax.
Benn xeetu nit neena ' jiroom ñeenteelu njiitu réew ma moy William Henry Harrison' waaye nak mungi fakkastalu ci ' ban njiit moo nekkoon William Henry Harrison.'
Assistant codage bi jàng mapping fonction-à-description mënul am tur fonction ci description bi kese.
Benn sistem QA medikaal buñu tàggat ci 'Drog X dafay faj feebar Y' mënul lim Drog X suñu ko laajee lan mooy faj feebar Y.
Ëlëgu mbañeel ci LLMs
Mitigations yi ñuy jàngat bokkuna ci yokk done ci ñaari yoon (yokkum ay frase yuñ delloo), mébetu tàggat yiy wax luy waaja am ci ñaari yoon yi, ak sistem yuy seet mbir yi ci anam wu symétriquement moo gën ñuy wéeru ci diisaay yuñ xam. Yenn architecture yu bees ak jàngat yu ñuy njëkka def ngir wàññi bërëb bi. Xaarandil ni musiba bi dina wàññeeku waaye du ni mes, ndax dafay wane ñàkka méngoo bu xóot bi am ci digganté njàngum token bu ci topp ak structure symétrique ci diggante àdduna dëgg.
Teg ko ci jëf
Lu tax mu am solo
- Gaawaay ak yaatuwaay. Liggéeyukaay yi ci làkk yi mën nañu gëna gaaw te duñu yàq deggoo gi.
- Dugg ak yegg. Dafay yaatal jëfandikoo gi ci làkk yi ak ci anam yi ñuy jokkoo.
- dogal yu gëna leer. Ekip yi mën nañu gëna yàgg ci àtte ci jamono ji otomatisation di liggéey ci baamtu.
Moytul
- Lépp lu jaarul yoon mën na dugg ci rapoor yi, jàppale ci liggéey bi, wala ci njariñu gëstu bi.
- Sensibilite bu gaaw mën na jur njariñ yu wuute ci laajte yu noonu mel.
- Done yu am solo mën nañu feeñ sudee seytu jëfandikoo gi néew doole.
Jéego yi ci topp
- Mandargal formaa génne gi, melokaan bi, ak standard kalite yi laata ngay dugal ko.
- Tontu yu am solo ak balluwaay yu wóor saa yu dëggu bi di am solo.
- Fexeel am barabu xool nit ñi ngir am njariñ yu am solo.
- Toppal anami gacce yi ak di faral di tàggataat ay laaj wala def-liggéey.
Weyal di banneexu
Bulletin bu amul fayda
Jàngal xibaar AI bu bes
Ñatti jaar-jaari IA yuñ firndeel bis bu nekk ci ayu-bis bi, ñu bind leen ci Àngle bu leer. Gratis ba fàww, amul yëgle.
Benn email bis bu nekk ci ayu-bis bi. Benn klik nga dindi abonemaa bi. Dunu musa jaay wala séddoo sa adres.
Nattal sa bopp
Take the Reversal Curse in LLMs quiz
Feedback ci saasi ci bépp tontu, ak sertifikaa buñ mëna séddoo ak ID buñ mëna saytu soo jàllee ci njàng mi.
Jàppale njàngum IA te doo fay. AI Understanding mbootaay la buy def te Yàlla tax — amul yëgle, amul paywall, musul am. Defal ab sarax
Laaj yi ñuy faral di laaj
What is Reversal Curse in LLMs?
Bañ bu ñu delloo ap anam bu yeeme la ndax xeetu lakk bu jang 'A moy B' munul tontu ci anam wu wóór 'B moy A.' Dafay wane ni LLM yi dañuy denc mbir yi ni kuréel yu am benn yoon, duñu ni xam-xam bu symétrique.
A model learned “A is B” but fails to answer “B is A.” What is this called?
The paper names this directional failure the reversal curse.
What does the reversal curse show about directional associations?
The paper reports that forward and reverse statements can behave differently.
Is the reversal curse simply a failure because the model never saw the fact?
The research tests failure on a reverse query after forward exposure.
Why might next-token training contribute to the effect?
The paper discusses sequence direction as relevant to the failure.
What does failure on a reverse query imply about model knowledge?
The finding challenges assumptions about flexible use of stored facts.
Weyal di jàng
Gid yu jëm ci loolu
Tann nañu yeneen njiit ngir topic bii