Mën-mën yu feeñ ci misaali làkk yu mag
Mën-mën yuy feeñ mooy mën-mën yiy feeñ ci saasi ci misaali làkk yu mag yi ginaaw bi ñu weesee benn eskaal, doonte misaal yu ndaw yi wone wuñu leen benn màndarga.
Résumé
They matter because they make capabilities hard to predict from small-scale experiments.
Plongeur bu xóot
Ñu siiwal ko ci benn téere bu Wei ak ay naataangoom bind ci 2022, génne gi dafay wax ci liggéey yu performance yi di des ci wetu saas ngir model yu gëna ndaw ba noppi di tëb bu baax su model bi jéggee benn dayo ci parametre yi, done yi, wala calcul yi. Ay misaal yuñ joxe bokkuna ci xayma yu bari, yenn tegtali xalaat, ak topp tegtal yu bees. Li ci gënoon fës mooy ñàkka wéy: mën mën bi nekkul woon luy yokk ndànk, dafa melni lu amul ba noppi nekk. Benn toppu bu 2023 bu Schaeffer ak ay naataangoom wax ni yenn feeñte yi bokk nañu ci ay mbiri natt, ndax ay metrik yu dëgër yu mel ni exact-match dañuy juum ci ay tëb yu gaaw yuy nuru lu nooy ci suufu poñ yu gëna woyof. Werante bi soppali na anam wi gëstukat yi di joxee rapoor ci resultaa yi ak tànneefi evaluation yi.
Gis-gis xarala
Ndax génn gi 'real' lay faral di aju ci metric bi. Liggéey buñu joxe poñ ci match exact du jox benn credit ba jéego bu nekk jub, kon benefiis bu dëgër ci njubte ci token bu nekk mën na feeñ ni tëb bu gaaw. Weccoo ci benn metric bu wéy lu melni niveau token wala crédit partiel, te courbe bi dafay faral di nuru lu nooy. Kon génn gi dafay wane jaxasoo bi am ci digganté màggug kàttan dëgg ak ñàkka wéy giñ tabax ci sàrtu poñ biñ tànn.
njeextalu pexe
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.
Ëlëgu Mën-mën yiy Feeñ ci Royuwaayi Lakk yu Yaatu
Leegi gëstukat yi dañuy boole njàngum scaling ak metrics yu bari ngir tàqale coppite fase yu dëggu yi ak artifact yi, ba noppi ñuy saytu ban kàttan mooy yegsi ci scale bi. Li gëna am solo mooy gëna mëna xam luy am ci kaaraange, ndax mën nañu am luñu xamul luy am risk. Xaarandil liggéey bu gëna bari ci yooni scaling yiy wax luy waaja am ci kanam, boole ci jëmmal benchmark bu jaar yoon suko defee ñu wax ni 'feññ' dafay wane doxalinu model bi te baña nekk benn quirk ci natt gi.
Doxal ci àdduna dëgg
Modèle yu mag yiy saafara jafe-jafe baat yu bari te version yu ndaw yi tontu ci niveau chance.
Benn model buy jekki saa rek topp ay tegtal yu jafee xam, yu kenn musul gis ginaaw bimu jàllee benn ëttu balans.
Caden-de-pensée biy ñaax nit ñi ñu xalaat ludul su model yi yeggee ci dayo bu doy.
Gëstukat yi dañu defaraat ap tëb bu 'ci saasi' ak poñ yu xaaj ci kredi ba noppi gis courbe bu neex.
Risk yi ak balustrade yi
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.
Roadmap ngir samp gi
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
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 Emergent Abilities of Large Language Models 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
Gis bi ci topp
Sycophancy ci misaali làkk
Laaj yi ñuy faral di laaj
What is Emergent Abilities of Large Language Models?
Mën-mën yuy feeñ mooy mën-mën yiy feeñ ci saasi ci misaali làkk yu mag yi ginaaw bi ñu weesee benn eskaal, doonte misaal yu ndaw yi wone wuñu leen benn màndarga. Dañu am solo ndax dañu tax ñu baña mëna wax luy am ci jàngat yu ndaw.
Lan mooy màndargaal mën mën buy feeñ ci misaali làkk yu yaatu?
Mën-mën yiy feeñ daanaka duñu am ci model yu gëna ndaw yi, ba noppi ñuy feeñ bu baax su eskaal bi jàll benn buntu.
Lan la toppu 2023 bi Schaeffer ak ay naataango wax?
Wane nañu ni metrics all-or-nothing mën na tax benefiis yu baax yi nuru ay tëb yu gaaw.
Lan moo waral poñ yi méngoo mënna juum ci génn gi?
Match exact du jox benn credit ngir yokkute bu ndaw, kon yokkute bu dëgër ci token bu nekk dafay nuru tëb bu gaaw.
Ban facteur scaling ñoo méngoo ak génnug àdduna?
Dañuy xamle tëb yu am solo yi model yi di màgg ci parametre yi, ci done yiñ tàggat ak ci xayma yi.
Lan moo waral feeñte am solo ci kaaraange IA?
Sudee kàttan yi mën nañu feeñ ci lu gaaw ci eskaal bi, yenn yu ñu xamul mën nañu nekk lu jafe, di waral ñu gëna mëna wax luy am ci kanam.