๐Ÿ“ข ๊ฒ€์ƒ‰ ๊ธฐ๋Šฅ ์ถ”๊ฐ€ ์˜ˆ์ •

์ฃผ๋ฏผ๊ฑด

๊ธฐ๋ก = ์„ฑ์žฅ

zoomg

ํŠธ๋žœ๋“œ๋ฅผ ์ฝ๋Š” ๋ฐฉ๋ฒ•

์‹œ๋ชฌ์Šค ๋ถ€์‚ฌ์žฅ ๊น€์„ฑ์ค€

Impression (๋ฉ”์„ธ์ง€ ์ „๋‹ฌ) โ†’ Perception (๋Š๋‚Œ ์ „๋‹ฌ)

06:56; ๋ชจ๋‘๊ฐ€ ๊ธฐ์–ตํ•˜๋Š” ๊ทธ ๊ด‘๊ณ  ๋น„ํ•˜์ธ๋“œ ์Šคํ† ๋ฆฌ

์–ด๋–ค ์ปจํ…์ธ ๋ฅผ ๋งŒ๋“ค์ง€ ์ƒ๊ฐํ•˜๊ธฐ๋ณด๋‹จ ํ”Œ๋žซํผ ํŠน์„ฑ์„ ์šฐ์„ ์ ์œผ๋กœ ์ดํ•ด

  • ๊ทธ ๋‹น์‹œ TV ๊ด‘๊ณ ๋Š” ๋ฉ”์„ธ์ง€๋ฅผ ์ผ๋ฐฉ์ ์œผ๋กœ ์ „๋‹ฌํ•˜๊ณ ์ž ํ•จ
  • ๋„ˆ๋ฌด๋‚˜ ๋งŽ์€ ์ •๋ณด๊ฐ€ ๋“ค์–ด์˜ค๋ฉด ์ฑ„๋„์„ ๋Œ๋ฆฌ๊ฒŒ ๋จ
  • ์›ํ•˜๋Š” ๋Š๋‚Œ๋งŒ ์ „๋‹ฌ ํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•˜์ž

์‹œ๋ฅ˜(ํŠธ๋ Œ๋“œ?)๋ฅผ ์ฝ๋Š” ํž˜

16:07; ๋””์ž์ธ ํŠธ๋ Œ๋“œ ๋Œ€์‹  ์‹œ๋ฅ˜ ํŒŒ์•…ํ•˜๊ธฐ

์‚ฌํšŒ ํ˜„์ƒ์— ๋Œ€ํ•ด ์™œ๋ฅผ ๋Š์ž„ ์—†์ด ํƒ๊ตฌ

๊ธฐ์‚ฌ๋ฅผ ๋งŽ์ด ์ฝ์Œ

  • ์™œ ์ด๋Ÿฐ ํ‚ค์›Œ๋“œ๊ฐ€ ๋งŽ์ด ๋ณด์ผ๊นŒ
  • ์™œ ์ด๋Ÿฐ ๊ธฐ์‚ฌ๊ฐ€ ๋งŽ์ด ๋ณด์ผ๊นŒ

ํ•ซํ•œ ๋™๋„ค์— ๊ฐ€์„œ ์‚ฌ๋žŒ๋“ค์„ ๊ด€์ฐฐํ•จ

  • ์ค„ ์„œ์žˆ๋Š” ์‚ฌ๋žŒ๋“ค์˜ ํ‘œ์ •๊ณผ ์˜ท ์ž…๊ณ  ์žˆ๋Š” ๊ฒƒ
  • ์–ด๋–ค ์‚ฌ๋žŒ๋“ค์ด ์ €๊ธฐ ์ง„์ž…ํ•˜๋Š”์ง€
  • ๋ญ˜ ๋“ค๊ณ  ๋‚˜์˜ค๋Š”์ง€
  • ์—์–ดํŒŸ์„ ๊ผฝ๋Š”์ง€
  • ์™œ ์„ ์ด ์žˆ๋Š” ์ด์–ดํฐ์ด ๋‹ค์‹œ ๋“ฑ์žฅํ•œ๊ฑด์ง€

30% 70% ๋ธŒ๋žœ๋”ฉ ์ „๋žต

18:37; ์š”์ฆ˜์— ๋ธŒ๋žœ๋”ฉ์€ ์ •์˜๋‹นํ•  ์ˆ˜ ์žˆ์–ด์•ผ ํ•œ๋‹ค
  • ๋ธŒ๋žœ๋”ฉ์€ ์ปค๋ฎค๋‹ˆ์ผ€์ด์…˜ ์ „๋žต์ด๊ธฐ ๋•Œ๋ฌธ์— ์šฐ๊ธฐ๋ฉด ์•ˆ๋œ๋‹ค
  • ๋ฐฉํ–ฅ์„ฑ์„ ์œ„ํ•ด ํฐ ๊ทธ๋ฆผ ์ •๋„(30%)๋งŒ ๊ทธ๋ ค์ฃผ๊ณ 
  • ๋‚˜๋จธ์ง€ 70%๋Š” ์†Œ๋น„์ž๋“ค ํ˜น์€ ๋ถˆํŠน์ • ๋‹ค์ˆ˜์™€ ํ”ผ๋“œ๋ฐฑ์„ ์ฃผ๊ณ  ๋ฐ›์œผ๋ฉด์„œ ์ „๋žต ์ˆ˜์ •

ํŒ€ ์šด์˜์—๋„ ์‚ฌ์šฉํ•˜๊ณ  ๊ณ„์‹  ์ „๋žต

  • ๋ฐฉํ–ฅ์„ฑ์„ ์œ„ํ•ด ํฐ ๊ทธ๋ฆผ ์ •๋„(30%)์ธ ํ•˜์ง€ ๋ง์•„์•ผ ๋  ๊ฒƒ, ๋ฌด์กฐ๊ฑด ํ•ด์•ผ ๋  ๊ฒƒ์„ ์•Œ๋ ค์ฃผ๊ณ 
  • ๋‚˜๋จธ์ง€ 70%๋Š” ๊ตฌ์„ฑ์›์ด ํ•ด์˜ค๋Š”๋Œ€๋กœ ํ”ผ๋“œ๋ฐฑ ์ฃผ๊ณ  ๋ฐ›์œผ๋ฉด์„œ ์ „๋žต ์ˆ˜์ •

๊ตฌ์„ฑ์›์˜ ์ฐฝ์˜์„ฑ์„ ๋†’ํž ์ˆ˜ ์žˆ๋Š” ์ „๋žต

  • Must TODO list (or Not TODO list): ์ง„์งœ (์•ˆ)ํ•ด์•ผ๋˜๋Š” ๋ถ€๋ถ„, ๋‚˜๋จธ์ง€๋Š” ์ž์œ ๋กญ๊ฒŒ ํ•ด๋„ ๋จ
  • ๊ผญ ๋ˆ ๋ฐ›๊ณ  ํŒ”์•„์•ผ๋ผ = ๊ณต์งœ๋กœ ๋‚˜๋ˆ ์ฃผ๋ฉด ๋‚ด๊ฐ€ ์„ ํƒํ•œ ๊ด€๊ณ„๊ฐ€ ์•„๋‹ˆ๊ธฐ์— ๊ธฐ์–ต ๋ชปํ•จ
  • ๊ผญ ์‹ผ ๊ฐ’์— ํŒ”์•„์•ผ๋ผ = ๊ฑฐ์˜ ๊ณต์งœ ๋Š๋‚Œ์œผ๋กœ, ํ•˜์ง€๋งŒ ๊ตฌ๋งค ์„ ํƒ์€ ์†Œ๋น„์ž๊ฐ€
  • ๊ผญ ์‹œ๋ชฌ์Šค ์˜›๋‚  ๋กœ๊ณ  ๋˜๋Š” 150์ด๋ผ๋Š” ๊ธ€์ž๋Š” ๋„ฃ์–ด์•ผ๋ผ = 150 ์ฃผ๋…„๋„ ์•Œ๋ฆฌ๊ณ , ์‹œ๋ชฌ์Šค๋„ ์•Œ๋ ค์•ผ๋˜๋‹ˆ๊นŒ
  • ๊ผญ ์™ธ๊ด€์— ๋†๊ตฌ๋Œ€ ํ•˜๋‚˜๋งŒ ๋‹ฌ์ž = ์‚ฌ์ง„ ์ฐํ˜€์„œ ๋Œ์•„๋‹ค๋‹ํ…๋ฐ ๋ˆ„๊ฐ€ ๋ด๋„ ํ•˜๋“œ์›จ์–ด ์Šคํ† ์–ด์ธ์ง€ ์•Œ์ •๋„๋กœ ์‹œ๊ทธ๋‹ˆ์ฒ˜ํ•œ ๋””์ž์ธ ์š”์†Œ๋ฅผ ๋„ฃ์ž + ๋†๊ตฌ๋Œ€=ํž™ํ•ฉ

ํ˜์‹ ์˜ ์ •์˜์™€ ๊ธฐํš์˜ ์ค‘์š”์„ฑ

์ŠคํŽ˜์ด์Šคํ”Œ๋ž˜๋‹ ๋Œ€ํ‘œ ์ •์šฐ์„

19:12; ์ €๋„ ์ด์ œ ์ฒ˜์Œ์— ์‚ฌ์—… ์‹œ์ž‘ํ•  ๋•Œ '๋‚˜๋Š” ๋ชจํ…”์„ ํ˜์‹ ํ•ด์•ผ ๋œ๋‹ค'๋ผ๋Š” ์ƒ๊ฐ์ด ํ•ญ์ƒ ์žˆ์—ˆ๊ฑฐ๋“ ์š”? 'ํ˜์‹ ์ด ๊ทธ๋Ÿผ ๋ญ˜๊นŒ?' ์ด๋Ÿฐ ๊ฑธ ์ƒ๊ฐ์„ ํ•ด๋ณธ ์ ์ด ์žˆ์–ด์š”. ๊ทธ ํ˜์‹ ์— ๋Œ€ํ•œ ์ •์˜๋ฅผ ์ €๋Š” 'ํ”ผํ„ฐ ๋“œ๋Ÿฌ์ปค'ํ•œํ…Œ์„œ ์ฐพ์•˜์–ด์š”
ํ˜์‹ ์„ ์ชผ๊ฐœ๋ฉด ๋ญ๊ฐ€ ๋˜๋ƒ๋ฉด ํ๊ธฐํ•˜๋Š” ๊ฒƒ๊ณผ ๊ทธ ํ๊ธฐํ•œ ๊ฒƒ์œผ๋กœ ๋‚จ์€ ์œ ํšจ ๋ฆฌ์†Œ์Šค๋ฅผ ๋‚ด๊ฐ€ ์‹œ์žฅ์˜ ํŠธ๋ Œ๋“œ๋ฅผ ๋ฐ˜์˜ํ•ด์„œ ๋‚ด ๋น„์ง€๋‹ˆ์Šค์— ์œ ๋ฆฌํ•˜๊ฒŒ ์ ์šฉํ•  ์ˆ˜ ์žˆ๋Š” ๊ฒƒ์— ํˆฌ์žํ•˜๋Š” ๊ฒƒ. ๊ทธ ๋‘ ๊ฐœ๋กœ ๋‚˜๋‰˜์–ด์š”. ์ด์ œ๋Š” ๋” ์ด์ƒ ํ•˜์ง€ ์•Š์•„๋„ ๋  ๊ฒƒ๋“ค.
๊ทผ๋ฐ ๋ชจํ…”์€ ๊ทธ๊ฒŒ ๋„ˆ๋ฌด ๋งŽ์•„์š”. ์•ˆ์ข‹์€ ์„œ๋น„์Šค, ๋ถˆํ•„์š”ํ•œ ์ผํšŒ์šฉํ’ˆ, ํ™”๋ คํ–ˆ๋˜ ์น˜์žฅ์œผ๋กœ ๋˜์–ด์žˆ๋Š” ๋ชฉ๊ณต์‚ฌ ์ด๋Ÿฐ ๊ฒƒ๋“ค์ด ๋‹ค ํ๊ธฐํ•  ๊ฑฐ์˜ˆ์š”. ๊ทธ๋ž˜์„œ ๋ชจ๋“  ์นดํ…Œ๊ณ ๋ฆฌ๋ณ„๋กœ ํ๊ธฐํ•  ๊ฒƒ๋“ค์„ ์ •๋ฆฌํ•˜๋‹ˆ๊นŒ ์ด๊ฒŒ ํ•œ ๊ฐ€๋“์ด์—์š”. ๊ทผ๋ฐ ์–˜๋„ค๋งŒ ํ๊ธฐํ•ด๋„ ์˜ˆ์‚ฐ์ด ์—„์ฒญ ์•„๊ปด์งˆ ์ˆ˜๊ฐ€ ์žˆ๋Š”๊ฑฐ์˜ˆ์š”.
์ œ๊ฐ€ ์‚ฌ์—…์„ ์‹œ์ž‘ํ•  ๋•Œ๋Š” ๊ทธ๋•Œ ๋‹น์‹œ ์ฝ”๋กœ๋‚˜๊ฐ€ ๋ฐœ๋ฐœํ–ˆ๋˜ ์‹œ๊ธฐ์˜ˆ์š”. ๊ทธ๋Ÿฌ๋‹ค ๋ณด๋‹ˆ๊นŒ ๋ฌด์Šจ ํŠน์ง•์ด ์žˆ์—ˆ๋ƒ๋ฉด ์ด ์‚ฐ์—…์—๋Š”, ์ „์ฒด ์ด์ œ ๊ณต๊ฐ„ ์‚ฐ์—…์—๋Š” PC๋ฐฉ, ๋…ธ๋ž˜๋ฐฉ, ๊ฒŒ์ž„๋ฐฉ ์ด๋Ÿฐ ๊ฒƒ๋“ค์ด ์ „๋ถ€ ๋‹ค ์ง‘ํ•ฉ๊ธˆ์ง€ ์ œํ•œ์ด ๊ฑธ๋ ค์š”. ๊ทธ๋ž˜์„œ ์ปจํ…์ธ ๋Š” ๋„ˆ๋ฌด ์ข‹์€๋ฐ ๊ทธ๊ฑธ ๋Œ€๊ด€์„ ๋ชปํ•ด์ฃผ๋Š” ์ƒํ™ฉ์ด ๋œ๊ฑฐ์˜ˆ์š”. ๊ทผ๋ฐ PC๋ฐฉ ํ”„๋žœ์ฐจ์ด์ฆˆ ์—„์ฒญ ๋งŽ์ž–์•„์š”. ๋…ธ๋ž˜๋ฐฉ ํ”„๋žœ์ฐจ์ด์ฆˆ ์—„์ฒญ ๋งŽ์ž–์•„์š”. ๊ทธ ๋งŒํผ ๊ทธ ์ปจํ…์ธ ๋Š” ๊ฒ€์ฆ์ด ๋œ ๊ฑฐ๊ฑฐ๋“ ์š”. ย ๊ทธ๋Ÿผ ๊ทธ ์ปจํ…์ธ ๋ฅผ ์ง€๊ธˆ ์ง‘ํ•ฉ๊ธˆ์ง€ ์ œํ•œ์— ์ œํ•œ์ด ์—†๋Š” ๋ชจํ…”์—๋‹ค๊ฐ€ ๊ฒฐํ•ฉ์‹œํ‚ค๋ฉด ๋˜๊ฒ ๋‹ค. ๊ทธ๊ฒŒ ์ €๋Š” ํ˜์‹ ์ด๋ผ๊ณ  ๋ดค์–ด์š”. ๊ทธ๋ ‡๊ฒŒ ์ด์ œ ์ปจํ…์ธ  ํ˜ธํ…”์ด ๋‚˜์˜ค๊ฒŒ ๋œ ๊ฑฐ์˜ˆ์š”.
๊ทธ ํ•„์š” ์—†๋Š”๊ฑฐ๋ฅผ ์˜›๋‚ ์—๋Š” ์™œ ํ–ˆ์„๊นŒ์š”?
๊ทธ๊ฒŒ ์ด์ œ ๋ถˆ์•ˆํ•จ์—์„œ ์˜ค๋Š”๊ฑฐ์˜ˆ์š”. ์–ด๋””๋‹ค๊ฐ€ ์šฐ์„ ์ˆœ์œ„๋ฅผ ํ•ด์•ผ ํ• ์ง€ ๋ชจ๋ฅด๊ฒ ๊ณ , ๋‚จ๋“ค๋ณด๋‹ค๋Š” ๋‹๋ณด์—ฌ์•ผ ๋˜๊ณ , ๊ทธ๋ฆฌ๊ณ  ๊ฐ€์žฅ ํฐ ์ด์œ  ์ค‘์— ํ•˜๋‚˜๊ฐ€ ์˜ˆ์ „์—๋Š” ์ด๋Ÿฐ ์˜จ๋ผ์ธ์œผ๋กœ ์ˆ™์†Œ๋ฅผ ์˜ˆ์•ฝํ•˜๋Š” ๊ฑด ์•„๋‹ˆ์—ˆ์ž–์•„์š”. ์™ธ์žฅ์„ ๋ณด๊ณ  ์ด ํ˜ธํ…”์˜ ์‹œ์„ค์„ ๊ฐ์•ˆํ•˜๋Š” ์‹œ์ ˆ์ด ์žˆ๊ธด ํ–ˆ์ฃ . ๊ทธ ๋•Œ๋Š” ์™ธ์žฅ์ด ํ™”๋ คํ•˜๋ฉด ์ €๊ธด ๋‚ด๋ถ€๋„ ์ข‹๊ฒ ์ง€. ๊ทผ๋ฐ ์ง€๊ธˆ์€ ์˜์‚ฌ๊ฒฐ์ • ๊ตฌ์กฐ๊ฐ€ ์™„์ „ํžˆ ๋ฐ”๋€Œ์—ˆ์–ด์š”.

36:25; ๊ทธ๋ž˜์„œ ๊ธฐํš์ด ์ค‘์š”ํ•˜๋‹ค๊ณ  ์ƒ๊ฐํ•ด์š” ์ €๋Š”. '๊ณต๊ฐ„์„ ์–ด๋–ป๊ฒŒ ์ธํ…Œ๋ฆฌ์–ดํ•œ๋‹ค'๊ฐ€ ์ค‘์š”ํ•œ ๊ฒŒ ์•„๋‹ˆ๊ณ , ์šฐ๋ฆฌ๊ฐ€ ์™œ ์ด ๊ณต๊ฐ„์„ ๋งŒ๋“ค๊ณ  ์–ด๋–ป๊ฒŒ ๋งŒ๋“ค ๊ฒƒ์ธ๊ฐ€ ๊ทธ๊ฑฐ๋ฅผ ๋จผ์ € ์ €ํฌ๊ฐ€ ์ •ํ•˜๊ณ  ๋ชจ๋“  ๊ทธ๋ฆผ์„ ๊ทธ๋ฆฌ๋Š” ๊ฒƒ์„ ์‹œ์ž‘ํ•ด์•ผ ๋ผ์„œ ๋‹จ์ˆœํžˆ ๋ฉ‹์ง„ ๊ฑธ ๋งŒ๋“ ๋‹ค๋ผ๊ธฐ๋ณด๋‹ค ์ด๊ฒƒ๋“ค์— ๋Œ€ํ•œ ๊ธฐ๋Šฅ์„ ์ข€ ์ •์˜ํ•ด ๋ณด๊ณ  ์–ด๋–ป๊ฒŒ ํ•  ๊ฑด์ง€ ๊ทธ๊ฑธ ๋จผ์ € ์ •ํ•˜๋Š” ๊ฒŒ ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค.

ํ˜์‹  = ๋ฆฌ์†Œ์Šค ํšจ์œจํ™” + ํŠธ๋žœ๋“œ ๋ฐ˜์˜

ํŠธ๋žœ๋“œ ๋ฐ˜์˜ = ํŠธ๋žœ๋“œ ๋ถ„์„ + ๊ธฐํš

๊ธฐํš = ์™œ + ์–ด๋–ป๊ฒŒ

AI ๋ชจ๋ธ ๋ถ•๊ดด ํ˜„์ƒ

ํ˜„์ƒ๊ณผ ๋ฌธ์ œ

  • ๋ชจ๋ธ์—์„œ ์ƒ์„ฑํ•œ ๋ฐ์ดํ„ฐ๋กœ ํ•™์Šต๋œ ๋ชจ๋ธ์—์„œ ์ƒ์„ฑํ•œ ๋ฐ์ดํ„ฐ๋กœ ๋‹ค์‹œ ํ•™์Šตํ•œ ๋ชจ๋ธ์„ ๋งŒ๋“ค ๊ฒฝ์šฐ
  • ์ถœํ˜„ ํ™•๋ฅ ์ด ๋‚ฎ์€ ๋ฐ์ดํ„ฐ๋Š” ํ•™์Šต์„ ๊ณ„์† ํ• ์ˆ˜๋ก ์ ์  ์‚ฌ๋ผ์ง€๊ฒŒ ๋œ๋‹ค
  • ์–ธ์–ด ๋ชจ๋ธ์— ์œ„ ๊ณผ์ •์„ ์ ์šฉํ•˜๋ฉด ์˜ฌ๋ฐ”๋ฅธ ์ˆœ์„œ์˜ ๋ฌธ์žฅ์„ ์ƒ์„ฑํ•  ํ™•๋ฅ ์ด ๋‚ฎ์•„์ง„๋‹ค

ํ•ด๊ฒฐ ๋ฐฉ์•ˆ

  • ํ•™์Šต ๋ฐ์ดํ„ฐ์˜ ๋ถ„ํฌ๋ฅผ ์‚ฌ์ „์— ํ™•์ธํ•ด์„œ ๊ด€๋ฆฌ ํ•  ์ˆ˜ ์žˆ์œผ๋ฉด ์ข‹๊ฒ ๋‹ค
  • ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ๋ฅผ ์ œ์ž‘ํ•  ๋•Œ ํ•˜๋‚˜์˜ LLM์„ ํ†ตํ•ด ๋งŒ๋“ค๊ธฐ ๋ณด๋‹ค๋Š” ์—ฌ๋Ÿฌ LLM์—์„œ ๋‚˜์˜จ ๊ฒฐ๊ณผ๋“ค์„ ์ข…ํ•ฉํ•˜๋Š” ํ˜•ํƒœ๊ฐ€ ์œ„ ํ˜„์ƒ์„ ๋ฐฉ์ง€ํ•  ์ˆ˜ ์žˆ๊ฒ ๋‹ค

๊ธฐํƒ€ ์ƒ๊ฐ

  • ์‚ฌ๋žŒ์€ ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ์—์„œ ์–ป์–ด์ง„ ๋ฐ์ดํ„ฐ๊ฐ€ ๋‹ค์–‘ํ•œ ํ˜•ํƒœ๋กœ ๋ณ€ํ™˜๋˜์–ด ์ €์žฅ๋œ๋‹ค
  • ์ด๋ฏธ์ง€๋ฅผ ๋ณด๊ณ  ํ…์ŠคํŠธ๋ฅผ ๊ธฐ์–ต์— ์ €์žฅํ•˜๊ฑฐ๋‚˜, ํ…์ŠคํŠธ๋ฅผ ๋ณด๊ณ  ์ด๋ฏธ์ง€๋ฅผ ๊ทธ๋ ค์„œ ์ €์žฅํ•œ๋‹ค
  • ์ƒํ™œ ์†์—์„œ ๊ฐ๊ฐ ๊ธฐ๊ด€์„ ํ†ตํ•ด ๋Š์ž„์—†์ด ๋“ค์–ด์˜ค๋Š” ์ •๋ณด๋Š” ํ•™์Šต ๋ฐ์ดํ„ฐ๋กœ ์น˜ํ™˜ํ•  ๊ฒฝ์šฐ ๊ทธ ์šฉ๋Ÿ‰์ด ์–ด๋งˆ์–ด๋งˆํ•˜๋‹ค
  • ๊ทธ๋ž˜์„œ ๋‡Œ๋Š” ๋ชจ๋‘ ์ €์žฅํ•˜๊ธฐ ๋ณด๋‹ค ์„ ๋ณ„ํ•ด์„œ ์ €์žฅํ•˜๋Š” ๊ธฐ๋Šฅ์ด ํŠนํ™”๋˜์–ด ์žˆ๋Š” ๋“ฏํ•˜๋‹ค

๋ฐ•์ฐฌ์„ฑ๋‹˜์˜ WWDC 2024 ์ƒ๊ฐ

  • ์• ํ”Œ์€ Adapter๋ฅผ ์–ด๋–ป๊ฒŒ ๊ตฌ์„ฑํ•˜๊ณ  ๋ฐฐ์น˜ํ•œ๊ฑธ๊นŒ?
  • iOS ๊ฐœ๋ฐœ์ด ํ•˜๊ณ  ์‹ถ์–ด์ง„๋‹ค (์• ํ”Œ ์ƒํƒœ๊ณ„์— ์ƒ์‚ฐ์ž๋กœ ๋“ค์–ด๊ฐ€๊ณ  ์‹ถ์–ด์ง„๋‹ค)

LLM ๊ฐœ๋ฐœ ๊ด€๋ จ ์ž๋ฃŒ ์•„์นด์ด๋น™

๊ฒฝํ—˜ ๊ณต์œ 

  • 2024.06.08; maven ํŒŒ์ธํŠœ๋‹ ๊ฐ•์—ฐ์ž๊ฐ€ ์ž‘์„ฑ์ž๋กœ ์žˆ๋Š” โ€œ1๋…„ ๋™์•ˆ LLM ๊ตฌ์ถ•ํ•˜๋ฉฐ ๋ฐฐ์šด ๋‚ด์šฉโ€ ๊ฒฝํ—˜ ๊ณต์œ  ๊ธ€
Applied LLMs - What Weโ€™ve Learned From A Year of Building with LLMs
A practical guide to building successful LLM products, covering the tactical, operational, and strategic.

(๋ฒˆ์—ญ)

1๋…„ ๋™์•ˆ LLM๊ณผ ํ•จ๊ป˜ ๊ตฌ์ถ•ํ•˜๋ฉฐ ๋ฐฐ์šด ์  | GeekNews
๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ(LLM)์„ ์‚ฌ์šฉํ•œ ๊ฐœ๋ฐœ์ด ํฅ๋ฏธ๋กœ์šด ์‹œ๊ธฐ์ž„์ง€๋‚œ 1๋…„ ๋™์•ˆ LLM์ด ์‹ค์ œ ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์— โ€์ถฉ๋ถ„ํžˆ ์ข‹์€โ€ ์ˆ˜์ค€์ด ๋˜์—ˆ์œผ๋ฉฐ, ๋งค๋…„ ๋” ์ข‹์•„์ง€๊ณ  ์ €๋ ดํ•ด์ง€๊ณ  ์žˆ์Œ์†Œ์…œ ๋ฏธ๋””์–ด์˜ ๋ฐ๋ชจ์™€ ํ•จ๊ป˜, 2025๋…„๊นŒ์ง€ AI์— ์•ฝ 2000์–ต ๋‹ฌ๋Ÿฌ๊ฐ€ ํˆฌ์ž๋  ๊ฒƒ์œผ๋กœ ์ถ”์ •๋จ์—…์ฒด๋“ค์˜ API๋กœ ์ธํ•ด LLM์ด ๋” ์ ‘๊ทผํ•˜๊ธฐ ์‰ฌ์›Œ์ ธ, ML ์—”์ง€๋‹ˆ์–ด์™€ ๊ณผํ•™์ž๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ๋ชจ๋‘๊ฐ€ ์ œํ’ˆ
  • 2024.06.04; ๋‹น๊ทผ์˜ LLM ๊ธฐ๋Šฅ ๊ฐœ๋ฐœ ๊ฒฝํ—˜ ๊ณต์œ  ์˜์ƒ
  • 2024.05.28; ์—˜๋ฐ•์Šค์˜ LLM ์„œ๋น„์Šค ๊ฐœ๋ฐœ๊ธฐ
LLM ๊ธฐ๋ฐ˜ Application, LBox AI ๊ฐœ๋ฐœ๊ธฐ
๋น ๋ฅด๊ฒŒ ๋ฐœ์ „ํ•˜๊ณ  ์žˆ๋Š” LLM ๊ธฐ๋ฐ˜์œผ๋กœ Application ์„ ๋งŒ๋“ค์—ˆ๋˜ ๊ฒฝํ—˜์„ ๊ณต์œ ํ•ด ๋ณด๊ณ ์ž ํ•ฉ๋‹ˆ๋‹ค. ์ƒˆ๋กœ์šด ๊ธฐ์ˆ ์ด ๋‚˜์˜ค๋ฉด ๋Š˜ ๊ทธ๋ ‡๋“ฏ, ๋‹ค์–‘ํ•œ ์‹œํ–‰์ฐฉ์˜ค๋ฅผ ๊ฑฐ์น˜๊ฒŒ ๋ฉ๋‹ˆ๋‹ค. ์ด ๊ธ€์—์„œ๋„ ์ •๋‹ต์ด ์•„๋‹Œ, ์—ฌ๋Ÿฌ ๊ฐ€์ง€ ๊ณ ๋ฏผ๋“ค์— ๋Œ€ํ•œ ์‹œํ–‰์ฐฉ์˜ค๋ฅผ ๊ณต์œ ํ•˜๊ณ ์žโ€ฆ
  • 2024.07.01; RAG์˜ ์„ฑ๋Šฅ๊ณผ ํšจ์œจ์„ฑ์„ ๋™์‹œ์— ์žก์„ ์ „๋žต
Searching for Best Practices in Retrieval-Augmented Generation
Retrieval-augmented generation (RAG) techniques have proven to be effective in integrating up-to-date information, mitigating hallucinations, and enhancing response quality, particularly in specialized domains. While many RAG approaches have been proposed to enhance large language models through queโ€ฆ
  • 2024.07.10; Notion AI, Evaluation of LLM Application
  • 24.07.24; Liner ์ง€๋‚œ 1๋…„ LLM ์–ดํ”Œ๋ฆฌ์ผ€์ด์…˜ ๊ฐœ๋ฐœ ํšŒ๊ณ 
  • 2024.08.18; Project Pluto, AI ์–ดํ”Œ๋ฆฌ์ผ€์ด์…˜์„ ๊ฐœ๋ฐœํ•˜๋Š” ์Šคํƒ€ํŠธ์—…์˜ ์ƒ๊ฐ (f. ์Šคํƒ€ํŠธ์—…์—๊ฒŒ Moat์ด๋ž€?)
  • 2024.08.25; Project Pluto, RAG์—์„œ R์˜ ์ •ํ™•์„ฑ์ด ๋งค์šฐ ์ค‘์š”, ๊ทธ ๋‹ค์Œ์€ ๊ฒ€์ƒ‰๋œ ์ •๋ณด๋ฅผ ์–ด๋–ค ์ˆœ์„œ๋กœ ์—ฎ์„๊ฑด์ง€๊ฐ€ ์ค‘์š”
Hyun Hong on LinkedIn: ๊ธฐ๊ด€์—๊ฒŒ ๋ฌป๊ณ , ๊ธฐ๊ด€์—๊ฒŒ ๋‹ตํ•˜๋‹ค --- It's not all about RAG
๐Ÿญ. ์จ์น˜์— ์žˆ์–ด์„œ ๐—ฅ๐—”๐—š๊ฐ€ ๋ฌด์Šจ ๐˜€๐—ถ๐—น๐˜ƒ๐—ฒ๐—ฟ ๐—ฏ๐˜‚๐—น๐—น๐—ฒ๐˜์ธ ๋งˆ๋ƒฅ ํ‘œํ˜„๋˜๋Š”๋ฐ ๊ทธ๋ ‡์ง€ ์•Š๋‹ค. '๊ทธ๋ƒฅ ๋ž˜๊ทธ๋กœ ๊ฐ–๋‹ค๋ถ™์—ฌ์„œ ํ•˜๋Š”๊ฑฐ ์ •๋„๋Š” ์šฐ๋ฆฌ๋„ ํ•  ์ˆ˜ ์žˆ์–ด์š”'๋ผ๊ณ  ๋งํ•˜๋Š” ๊ธฐ์—… ๋‚ด ํ…Œํฌ ๋‹ด๋‹น์ž๋“ค์—๊ฒŒ๋Š” ์ด๋ ‡๊ฒŒ ์„ค๋ช…ํ•˜๋Š” ํŽธ์ด๋‹ค. ๐—ฅ๐—”๐—š๋Š” ์ •๋ณด์˜โ€ฆ
  • 2024.08.08; ์žกํ”Œ๋ž˜๋‹›, LLM Agent ๊ตฌ์ถ•๊ธฐ
Tell-i : LLM Agent ์„œ๋น„์Šค ๊ตฌ์ถ•๊ธฐ (1) - ์žกํ”Œ๋ž˜๋‹› ํ…Œํฌ๋ธ”๋กœ๊ทธ
LLM AI Agent ๋„์ž…์— ํ•„์š”ํ•œ ๊ธฐ์ˆ ์„ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค.(MSA, RAG, Embedding) | Data&AI, Engineering
  • 2024.11.02; Naver, ๊ฒฝ๋Ÿ‰ํ™” ๋ ˆ์‹œํ”ผ
NAVER D2
  • 2024.11.04; Liner, AI ๊ฒ€์ƒ‰์—”์ง„ ๊ตฌ์ถ•๊ธฐ

ํ•ด์„

  • 2024.04.11; Transformer ํ•ด์„
๐Ÿง  ChatGPT์˜ ๋‹ต๋ณ€ ์กฐ์ข…์„ ์œ„ํ•œ Superposition Hypothesis
10์–ต๋ช…์˜ ์‚ฌ์šฉ์ž๋ฅผ ๊ฐ€์ง„ ChatGPT์˜ ๋‹ต๋ณ€์„ ์กฐ์ข…ํ•  ์ˆ˜ ์žˆ๋‹ค๋ฉด ์–ด๋–จ๊นŒ์š”? ๊ฐ€๋ น ๋Œ€ํ™”์— ์€๊ทผ์Šฌ์ฉ ๊ด‘๊ณ ๋ฅผ ๋ผ์›Œ ๋„ฃ๋Š”๋‹ค๊ฑฐ๋‚˜, ์„ ๊ฑฐ์— ์˜ํ–ฅ์„ ์ค„ ์ˆ˜๋„ ์žˆ๊ฒ ์ฃ . ์ด๋ ‡๊ฒŒ AI์— ๋Œ€ํ•œ ์ธ๊ฐ„์˜ ๊ฐœ์ž… ๋Šฅ๋ ฅ์ด ์ƒ๊ธด๋‹ค๋ฉด, ์ด๋Š” ๋ถ„๋ช… ์—„์ฒญ๋‚œ ๊ถŒ๋ ฅ์ž…๋‹ˆ๋‹ค.
  • 2024.06.30; In-context learning์ด ๋™์ž‘ํ•˜๋Š” ์ด์œ 
ํŠธ๋žœ์Šคํฌ๋จธ ๋ฆฌ๋ฒ„์Šค ์—”์ง€๋‹ˆ์–ด๋ง์œผ๋กœ In-context Learning ์ดํ•ดํ•˜๊ธฐ
์ง€๋‚œ 4์›” ๊ธ€์—์„œ ์˜ˆ์ƒํ•œ ๊ฒƒ์ฒ˜๋Ÿผ Anthropic๊ณผ OpenAI ๊ฐ๊ฐ ๋‘ ๋‹ฌ์ด ์ง€๋‚˜์ง€ ์•Š์•„ ๋Œ€ํ‘œ ํ”Œ๋ž˜๊ทธ์‹ญ ๋ชจ๋ธ์˜ ๋‰ด๋Ÿฐ ๋ถ„์„์„ ์ง„ํ–‰ํ–ˆ์Šต๋‹ˆ๋‹ค.
  • 2024.07.22; chain of thought๊ฐ€ ์ž˜ ๋™์ž‘ํ•˜๋Š” ์ด์œ  (Beyond Memorization: The Challenge of Random Memory Access in Language Models, ACL 2024)
  • 2024.08.07; LLM (ํ˜น์€ AI)์€ ๊ท€๋‚ฉ์  task์— ๋Œ€ํ•ด ํƒ์›”ํ•˜๋‹ค, ์—ฐ์—ญ์  task์— ๋Œ€ํ•ด์„œ๋Š” ๋„์›€์ด ํ•„์š”ํ•˜๋‹ค
Transformer Explainer: LLM Transformer Model Visually Explained
An interactive visualization tool showing you how transformer models work in large language models (LLM) like GPT.
  • 2024.09.02; LLM์€ ํ…์ŠคํŠธ ์••์ถ•, ๋ณต์› ์—ญํ• ์„ ํ•œ๋‹ค
  • 2024.08.07; LLM์€ ์ถ”๋ก  ๋Šฅ๋ ฅ์ด ์•„๋‹Œ ํŒจํ„ด ๋งค์นญ ๋Šฅ๋ ฅ์ด ์ข‹์•„์ง„ ๊ฒƒ, Apple
GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models
Recent advancements in Large Language Models (LLMs) have sparked interest in their formal reasoning capabilities, particularly in mathematics. The GSM8K benchmark is widely used to assess the mathematical reasoning of models on grade-school-level questions. While the performance of LLMs on GSM8K hasโ€ฆ
  • 2024.06.22; ํ”„๋กฌํ”„ํŒ… ์ตœ์ ํ™”์—๋Š” ์˜ˆ์ œ ์‚ฌ์šฉ์ด ์ตœ๊ณ , NeurIPS 2024, Google
Teach Better or Show Smarter? On Instructions and Exemplars in Automatic Prompt Optimization
Large language models have demonstrated remarkable capabilities, but their performance is heavily reliant on effective prompt engineering. Automatic prompt optimization (APO) methods are designed to automate this and can be broadly categorized into those targeting instructions (instruction optimizatโ€ฆ
์ผ๋ฐ˜ ์ฑ„ํŒ… ์ธํ„ฐํŽ˜์ด์Šค๋กœ LLM์„ ํ™œ์šฉํ•˜๋Š” ์‚ฌ๋žŒ ์ž…์žฅ์—์„œ ๋ณด์ž๋ฉด, ํ”„๋กฌํ”„ํŠธ ์ž์ฒด๋ฅผ ํŠœ๋‹ํ•˜๋ ค๋Š” ๋…ธ๋ ฅ์„ ๊ธฐ์šธ์ด๊ธฐ ์ „์—, ์ ์ ˆํ•œ ์˜ˆ์‹œ๋ฅผ ์ œ์‹œํ•˜๋Š” ๋…ธ๋ ฅ์„ ์šฐ์„ ์ ์œผ๋กœ ํ•˜๊ณ , ๊ทธ ๋‹ค์Œ์— ํ”„๋กฌํ”„ํŠธ๊นŒ์ง€ ๋ฐ”๊ฟ”๊ฐ€๋ฉด์„œ ์ตœ์ ํ™”ํ•˜๋Š” ๊ฒƒ์ด ๊ฐ€์žฅ ์ข‹์€ ๊ฒฐ๊ณผ๋ฅผ ๊ฐ€์ ธ์˜ฌ ์ˆ˜ ์žˆ๋‹ค๋Š” ์ด์•ผ๊ธฐ์ž…๋‹ˆ๋‹ค.

์„œ๋ฒ ์ด ๋…ผ๋ฌธ

  • 2024.06.17; ํ”„๋กฌํ”„ํŒ… ์„œ๋ฒ ์ด ๋…ผ๋ฌธ, 4796๊ฐœ์˜ ๋…ผ๋ฌธ ๋ชฉ๋ก ์ค‘ 1595๊ฐœ์˜ ๋…ผ๋ฌธ์œผ๋กœ ์ •๋ฆฌํ•œ ๋‚ด์šฉ
The Prompt Report: A Systematic Survey of Prompting Techniques
Generative Artificial Intelligence (GenAI) systems are being increasingly deployed across all parts of industry and research settings. Developers and end users interact with these systems through the use of prompting or prompt engineering. While prompting is a widespread and highly researched concepโ€ฆ
  • 2024.04.28; AI Assistants์˜ ์œค๋ฆฌ ๋ฐ ์‚ฌํšŒ์  ์œ„ํ—˜
The Ethics of Advanced AI Assistants
This paper focuses on the opportunities and the ethical and societal risks posed by advanced AI assistants. We define advanced AI assistants as artificial agents with natural language interfaces, whose function is to plan and execute sequences of actions on behalf of a user, across one or more domaiโ€ฆ

RAG

  • 2024.03.12; RAG, GraphRAG ์žฅ์ ๊ณผ ํ•œ๊ณ„ ๋ถ„์„ ์ •๋ฆฌ
From RAG to GraphRAG , What is the GraphRAG and why i use it? - graphwoody

๊น€์ธ์ค‘ ๊ต์ˆ˜๋‹˜์˜ ํ˜„ LLM ์ƒ๊ฐ

  • ํ˜„์žฌ LLM์˜ ๋ฐœ์ „ ์ˆ˜์ค€์„ ๊ฐ๊ด€์ ์ด๊ณ  ๋ƒ‰์ •ํ•˜๊ฒŒ ๋ถ„์„ํ•œ ๊ธ€
  • ํ•œ ๋ฐœ์ง ๋ฌผ๋Ÿฌ์„œ์„œ ๊ด€์ฐฐ์ž ์‹œ์ ์œผ๋กœ ์ƒํ™ฉ์„ ์ •๋ฆฌํ•œ ๊ธ€
  • ๋ณด์ด๋Š” ๊ฒƒ๋งŒ ๋ณด๊ณ , ๋ณด๊ณ ์‹ถ์€ ๊ฒƒ๋งŒ ๋ณด๊ฒŒ ๋˜๋Š” ์ƒํƒœ์— ๋Œ€ํ•œ ๊ฒฝ๊ฐ์‹ฌ์„ ์ฃผ๋Š” ๊ธ€

LangChain ์ดํ•ด ์ž๋ฃŒ

๊ต์œก๊ณผ ์ฝ”์นญ์— ๋Œ€ํ•œ ์ƒ๊ฐ

์ด์ „ ์ง์žฅ์ธ ์•Œ๊ณ ๋ฆฌ์ฆ˜๋žฉ์Šค์—์„œ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๋ฌธ์ œํ•ด๊ฒฐ๋Šฅ๋ ฅ ์ฝ”์น˜์— ์ง‘์ค‘ํ•  ๋•Œ ์ž‘์„ฑํ–ˆ๋˜ ์ž๋ฃŒ๋‹ค. ์ฝ”์น˜์™€ ์ˆ˜๊ฐ•์ƒ์„ ์œ„ํ•œ LMS๋Š” ์–ด๋–ป๊ฒŒ ์ œ์ž‘๋˜๋ฉด ์ข‹์„์ง€์— ๋Œ€ํ•œ ๊ณ ๋ฏผ์„ ์ •๋ฆฌํ–ˆ์—ˆ๋‹ค. ๊ต์œกํ•™์— ๋Œ€ํ•œ ๋ฐฐ๊ฒฝ์ด ์—†์–ด์„œ ์–ด๋–ป๊ฒŒ ๋ณด๋ฉด ๋‹น์—ฐํ•œ ๋ง์„ ์จ๋†“์€ ๊ฑธ ์ˆ˜๋„ ์žˆ์ง€๋งŒ, ๋จธ๋ฆฟ์†์—์„œ ์•”๋ฌต์ ์œผ๋กœ ๋™์ž‘ํ•˜๋Š” ์ ˆ์ฐจ๋ฅผ ๋ˆˆ์— ๋ณด์ด๊ฒŒ ๋งŒ๋“œ๋‹ˆ ๊ทธ๊ฐ„์˜ ์ฝ”์นญ ์Šคํ‚ฌ์ด ์ •๋ฆฝ๋˜๋Š” ๋Š๋‚Œ์ด๋ผ ์ข‹์•˜๋‹ค.

์ž๋ฃŒ๋ฅผ ์™„์„ฑํ•œ ๋‹น์‹œ์—๋Š” ์ฃผ๋ณ€์— ๊ฐ€๊นŒ์šด ์‚ฌ๋žŒ๋“ค์—๊ฒŒ๋งŒ ๊ณต์œ  ํ–ˆ์—ˆ๋‹ค. ๋”ฑํžˆ ๋„์›€ ๋œ๋‹ค๋Š” ๋ฐ˜์‘์ด ์ƒ๊ฐ๋งŒํผ ์˜ค์ง€ ์•Š์•„์„œ ์ปดํ“จํ„ฐ์—๋งŒ ์—ฌํƒœ ๋‚จ๊ฒจ๋’€์—ˆ๋‹ค. ์–ผ๋งˆ์ „ ๋ชจ๋‘์—ฐ SpiritusLAB์—์„œ ํ•ด๋‹น ์ž๋ฃŒ๋กœ ๋ฐœํ‘œํ•  ๊ธฐํšŒ๊ฐ€ ์žˆ์—ˆ๋‹ค. ๋žฉ์›๋ถ„๋“ค์ด ํฅ๋ฏธ๋กญ๊ฒŒ ๋“ค์–ด์ฃผ์…”์„œ ์ž๋ฃŒ๋ฅผ ๊ณต์œ ํ•˜๋ฉด ๋ˆ„๊ตฐ๊ฐ€์—๊ฒŒ๋Š” ๋„์›€์ด ๋ ์ˆ˜๋„ ์žˆ๊ฒ ๋‹ค๊ณ  ์ƒ๊ฐํ–ˆ๋‹ค.

์ž๋ฃŒ ๋‚ด์šฉ์€ ์ถœ์ฒ˜๊ฐ€ ๋ชจ๋‘ ๋‚ด ๊ฒฝํ—˜๊ณผ ๋จธ๋ฆฟ์†์— ์ผ์–ด๋‚œ ์ผ์ด๊ธฐ๋•Œ๋ฌธ์— ๊ฐ๊ด€์ , ๊ณผํ•™์  ๊ทผ๊ฑฐ์ž๋ฃŒ๊ฐ€ ์—†๋‹ค. ๋”ฐ๋ผ์„œ ๊ฒฝํ—˜๋‹ด ์ •๋„๋กœ๋งŒ ์ƒ๊ฐํ•˜๊ณ  ์ž๋ฃŒ๋ฅผ ์ฝ์–ด์ฃผ์…จ์œผ๋ฉด ํ•˜๋Š” ๋ฐ”๋žŒ์ด ์žˆ๋‹ค.

Great! Youโ€™ve successfully signed up.

Welcome back! You've successfully signed in.

You've successfully subscribed to zoomg.

Success! Check your email for magic link to sign-in.

Success! Your billing info has been updated.

Your billing was not updated.