Luqadda AI HAGAHA

Habaar celinta LLMs

Habaarka rogaal celiska ah waa hab guuldarro la yaab leh halkaas oo nooca luqadda barta 'A waa B' uusan si kalsooni leh uga jawaabi karin 'B waa A'. Waxay daaha ka qaadaysaa in LLM-yadu u kaydiyaan xaqiiqooyinka sidii ururro hal-jiho ah, oo aan ahayn aqoon summeedka.

Dhageyso hagahanAI-waa laga sheekeeyay · ~2 min

Boggaan2 daqiiqo akhri
  1. quusid qoto dheer
  2. Dhaqangelinta Adduunka-dhabta ah
  3. Mustaqbalka habaar celinta ee LLMs
  4. Ku dhaqangelinta
  5. Sii wad Sahaminta
  6. Su'aalaha soo noqnoqda

quusid qoto dheer

Waxaa lagu diiwaangeliyay warqad 2023 Berglund iyo asxaabtiisa, inkaarta rogaal celiska ah waxay muujineysaa in haddii model lagu tababaro 'Tom Cruise hooyadeed waa Mary Lee Pfeiffer,' waxay badanaa ku guuldareysataa marka la weydiiyo 'Waa ayo ina Mary Lee Pfeiffer?' inkastoo ay jawaabtu si macquul ah isku mid tahay. Saameyntu waxay ku sii socotaa cabbirrada moodeelka iyo xitaa ka dib marka si fiican loo habeeyo boqollaal xaqiiqooyin ah. Ma aha farqiga xusuusta: qaabku wuxuu arkay macluumaadka, laakiin kaliya hal nidaam. Sababtoo ah tababarku waxa uu wanaajiyaa saadaasha xigta ee habka saxda ah ee xogta, isku xirka tirakoobka ee A ilaa B si toos ah uma abuurayo isku xirka B ilaa A. Helitaanka fikrado caqabad ku ah oo kaliya ayaa soo saara dabacsanaan, fikrado aadanaha u eg oo ku saabsan xaqiiqda.

Dhaqangelinta Adduunka-dhabta ah

chatbot-ku wuxuu si sax ah u sheegaa qofka caanka ah waalidkiis laakiin wuu ku guul daraystay markii la waydiiyo inuu magacaabo ilmaha caanka ah ee waalidka.

Qaabka ayaa akhrinaya 'madaxweynihii sagaalaad waxa uu ahaa William Henry Harrison' weli wuu ku turunturooday 'madaxweyne nambarkee ahaa William Henry Harrison'.

Kaaliyaha codaynta ee bartay khariidaynta-shaqaynta-sharraxaadda kama soo celin karo magaca shaqada sharraxaadda oo keliya.

Nadaamka QA-da caafimaad ee lagu tabobaray 'Daawada X waxay dawaysaa Xaaladda Y' waxay ku guul daraysatay inay liis garayso Dawada X markii la waydiiyo waxa lagu daweeyo Xaaladda Y.

Mustaqbalka habaar celinta ee LLMs

Yaraynta daraasadda waxaa ka mid ah kordhinta xogta labada jiho (ku-darka weedho rogan), ujeedooyinka tababbarka ee saadaaliya calaamadaha labada dhinac, iyo hababka dib u soo celinta kuwaas oo u eegaya xaqiiqada si isku mid ah halkii ay ku tiirsanaan lahaayeen miisaanka la xafiday. Qaar ka mid ah dhismayaasha cusub iyo tijaabooyinka hore ee tababarka ayaa yareynaya farqiga. Filo in habaarku yaraado laakiin ma baabi'in doono, maadaama ay soo bandhigayso ismaandhaaf qotodheer oo u dhexeeya barashada calaamadda xigta iyo qaab-dhismeedka summeedka ee xidhiidhka adduunka dhabta ah.

Ku dhaqangelinta

Maxay muhiim u tahay

  • Xawaaraha iyo miisaanka. Socodka shaqada luqaddu si dhakhso leh ayay u socon kartaa iyada oo aan la hurayn joogteynta.
  • Helitaanka iyo gaarsiinta. Waxay balaadhisaa gelitaanka luqadaha iyo qaababka isgaarsiinta.
  • Go'aamo cad. Kooxuhu waxay waqti badan ku qaadan karaan xukunka halka otomaatiggu uu qabanayo ku celcelinta.

Iska ilaali

  • Xaqiiqooyinka dhalanteed waxay si deggan u geli karaan warbixinnada, taageerada socodka, ama natiijooyinka cilmi-baarista.
  • Dareenka degdega ahi wuxuu abuuri karaa natiijooyin aan iswaafaqayn codsiyada la midka ah.
  • Xogta qoraalka xasaasiga ah ayaa laga yaabaa in la kashifo haddii kontaroolada gelitaanka ay daciif yihiin.

Tallaabooyinka xiga

  1. Qeex qaabka wax soo saarka, codka, iyo heerarka tayada ka hor inta aan la baahin.
  2. Jawaabaha salka ku haya ilo lagu kalsoon yahay mar kasta oo saxnidu ay muhiim tahay.
  3. Hayso isbaarada dib u eegista bini aadamka ee wax soo saarka sare.
  4. Lasoco qaababka guuldarada oo dib u leyli dardargelinta ama socodka shaqada si joogto ah.

Sii wad Sahaminta

Warqad bilaash ah

Hel warbixinta maalinlaha ah ee AI

Saddex sheeko oo la xaqiijiyay oo AI ah maalin kasta oo shaqo ah, oo lagu qoray Ingiriis fudud. Bilaash had iyo jeer, aan xayaysiis lahayn.

Hal iimayl maalin kasta oo shaqo ah. Ka bax hal gujin. Marna ma iibinno ama ma wadaagno ciwaankaaga.

Is tijaabi

Take the Reversal Curse in LLMs quiz

Fikrad degdeg ah oo ku saabsan jawaab kasta, iyo shahaado la wadaajin karo oo leh aqoonsi la xaqiijin karo markaad ka gudubto koorso.

Bilow kedis

Taageer waxbarashada AI bilaashka ah. AI Understanding waa 501(c)(3) aan faa'iido doon ahayn — xayaysiis la'aan, albaab lacag ah ma jiro, had iyo jeer. Samee deeq

Su'aalaha soo noqnoqda

What is Reversal Curse in LLMs?

Habaarka rogaal celiska ah waa hab guuldarro la yaab leh halkaas oo nooca luqadda barta 'A waa B' uusan si kalsooni leh uga jawaabi karin 'B waa A'. Waxay daaha ka qaadaysaa in LLM-yadu u kaydiyaan xaqiiqooyinka sidii ururro hal-jiho ah, oo aan ahayn aqoon summeedka.

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.