Notice
Recent Posts
Recent Comments
ยซ   2024/12   ยป
์ผ ์›” ํ™” ์ˆ˜ ๋ชฉ ๊ธˆ ํ† 
1 2 3 4 5 6 7
8 9 10 11 12 13 14
15 16 17 18 19 20 21
22 23 24 25 26 27 28
29 30 31
Archives
Today
In Total
๊ด€๋ฆฌ ๋ฉ”๋‰ด

A Joyful AI Research Journey๐ŸŒณ๐Ÿ˜Š

[7] 241107 Generative AI [Goorm All-In-One Pass! AI Project Master - 4th Session, Day 7] ๋ณธ๋ฌธ

๐ŸŒณAI & Quantum Computing Bootcamp 2024โœจ/AI Lecture Revision

[7] 241107 Generative AI [Goorm All-In-One Pass! AI Project Master - 4th Session, Day 7]

yjyuwisely 2024. 11. 7. 15:48

241107 Thu 7th class

์˜ค๋Š˜ ๋ฐฐ์šด ๊ฒƒ ์ค‘ ๊ธฐ์–ตํ•  ๊ฒƒ์„ ์ •๋ฆฌํ–ˆ๋‹ค.


๋…ธ์ฝ”๋”ฉ์œผ๋กœ ๊ฐ€๋Š” ํšŒ์‚ฌ, ์‚ฌ๋žŒ๋“ค์ด ๋งŽ๋‹ค. 

๋งฅ๋ฝ์„ ๋ณด๊ณ  ์ „๋žต์ ์œผ๋กœ ํ•œ๋‹ค 
ํ˜„์žฌ ์œ„์น˜๋ฅผ ๋ช…ํ™•ํ•˜๊ฒŒ ํŒŒ์•…ํ•œ๋‹ค
๊ณต๋ถ€์™€ ์‹ค๋ ฅ์€ ์ฐจ์ด๊ฐ€ ์žˆ๋‹ค 
์›ํ•˜๋Š” ๊ณณ์ด ์•ˆ ๋˜๋ฉด ๋ฐ‘์œผ๋กœ ๊ฐ„๋‹ค 

์ผ ํ•˜๋ ค๊ณ  ํ•˜๋ฉด ๊ธฐ์ค€์„ ๋†’๊ฒŒ ์„ธ์šฐ์ง€X 

๋Œ€๊ธฐ์—…: ์ฝ”๋”ฉ ํ…Œ์ŠคํŠธ
์ค‘์†Œ๊ธฐ์—…: ์‹ค๋ ฅ ์Œ“๊ณ  -> ํฐ ๊ธฐ์—… 

๋งˆ์น˜๊ณ  3-5์‹œ๊ฐ„ ๊ณต๋ถ€, ๋…ผ๋ฌธ ๋ชจ์ž„ ๋“ฑ
1๋…„ ํ›„ ์ด์ง (๊ด€๋ จ ํšŒ์‚ฌ)

๋ถ„์•ผ ์ž˜ ์ •ํ•˜๊ธฐ 

๋ญ˜ ํ•˜๊ณ  ์‹ถ๋‹ค ๋ชฉํ‘œ ์ž˜ ์ •ํ•˜๊ธฐ

๊ทธ ๋ถ„์•ผ์˜ ๋…ผ๋ฌธ ์ฝ๋Š” ์ˆ˜์ค€์œผ๋กœ ๊ฐ„๋‹ค

๋ช‡ ๋…„ ๊ฑธ๋ฆฐ๋‹ค ๋ชจ๋ฅด๋Š” ๊ฒŒ ๋‹น์—ฐํ•œ ๊ฑฐ๋‹ค 

์‹ค๋ฌด์ž์˜ ์ˆ˜์ค€์œผ๋กœ ๊ณต๋ถ€ํ•œ๋‹ค ์‹ค๋ฌด์ž๋„ ๋‹ค ๋ชจ๋ฅธ๋‹ค ๊ณต๋ถ€ ์—„์ฒญํ•œ๋‹ค 

์ˆ˜์ค€์ด ์•„๋‹Œ ๋ฐ ๊ทธ๊ฑธ ๋ณด๊ณ  ์žˆ๋‹ค (๋…ผ๋ฌธ, ์Šคํƒ ํฌ๋“œ ๊ฐ•์˜, ๋‚ด ์ˆ˜์ค€์— ๋งก๋Š” ์ฑ…)

์„œ์ ์— ๊ฐ€์„œ ์ง์ ‘ ๊ฐ€์„œ ์ฝ์–ด๋ด์•ผ ํ•œ๋‹ค 

์ฑ…์€ ๋‘ ์„ธ๋ฒˆ ๋ด์•ผ ์˜๋ฏธ ์žˆ๋‹ค. 

๊ฐ์˜ค๊ฐ€ ์•ˆ ๋œ ์ฑ…์ด๋ฉด reference ์ฑ…์„ ๋ณธ๋‹ค ์ˆ˜์ค€์— ๋งž๋Š” ๊ฑธ๋กœ ๋ณธ๋‹ค 

์˜คํ”„๋ผ์ธ, ์‹ค์‹œ๊ฐ„ ์˜จ๋ผ์ธ ๊ฐ•์˜๊ฐ€ ๋‚ซ๋‹ค 

์ „๋ฌธ๊ฐ€๊ฐ€ ์•„๋‹ˆ์–ด๋„ ์ œํ’ˆ ๋งŒ๋“ ๋‹ค (ํ”„๋กฌํ”„ํŠธ ์—”์ง€๋‹ˆ์–ด๋ง) 

๊ฒฝ์Ÿ์ด ๋ถ™๊ธฐ ์‹œ์ž‘ํ•˜๋ฉด ๋กฑ๋Ÿฐ์ด ์–ด๋ ค์›Œ์ง„๋‹ค 

ํ”„๋กœ์ ํŠธ ์ž๊ธฐ ์ˆ˜์ค€X

์ด์Šˆํ™” ์‹œํ‚ด, ์ œ๋Œ€๋กœ ํ•˜๋ฉด ์ด์Šˆ๊ฐ€ ๋ฐœ์ƒํ•œ๋‹ค (์„ฑ๋Šฅ ๋“ฑ)

์ง€์‹์ด ๋Š˜์–ด๋‚˜์•ผ ํ•œ๋‹ค 

PPT ์ž๋ฃŒ ๋งŒ๋“ค๊ณ  ๋ฐœํ‘œ -> ๋™๋ฌด ๋จ 


์ด ๋…ผ๋ฌธ์— ๋Œ€ํ•ด ์•„์ฃผ ์‰ฝ๊ฒŒ ์„ค๋ช…์„ ํ•ด์ฃผ๋ ด. 


[ํ•™์Šต ๋ชฉํ‘œ]
์ƒ์„ฑํ˜• AI์˜ ์ด๋ฏธ์ง€ ์ƒ์„ฑ์— ๋Œ€ํ•ด ์ดํ•ดํ•ด ๋ด…๋‹ˆ๋‹ค.
LangChain์„ ํ™œ์šฉํ•˜์—ฌ LLM์„ ๊ตฌํ˜„ํ•ด ๋ด…๋‹ˆ๋‹ค.
๋จธ์‹ ๋Ÿฌ๋‹์˜ ๊ธฐ๋ณธ๊ณผ LLM์„ ์ดํ•ดํ•ด ๋ด…๋‹ˆ๋‹ค.


[ํ•™์Šต ๋‚ด์šฉ]
============================
(1) ์ง€๋‚œ ์‹œ๊ฐ„ ๋ฆฌ๋ทฐ(ํ…์ŠคํŠธ ๋ฐ์ดํ„ฐ ์ฒ˜๋ฆฌ, ChatGPT API)
(2) ์ด๋ฏธ์ง€ Prompt Engineering์— ๋Œ€ํ•ด ์•Œ์•„๋ด…๋‹ˆ๋‹ค.
(3) ์ƒ์„ฑํ˜• AI์— ๋Œ€ํ•ด ์•Œ์•„๋ด…๋‹ˆ๋‹ค.
(4) LangChain์„ ํ™œ์šฉํ•œ LLM์„ ๊ตฌํ˜„ํ•ด ๋ด…๋‹ˆ๋‹ค.
(5) ๋จธ์‹ ๋Ÿฌ๋‹ ๊ธฐ๋ณธ๊ณผ LLM์„ ์ดํ•ดํ•ด ๋ด…๋‹ˆ๋‹ค.
(6) ํŒŒ์ผ๋Ÿฟ ํ”„๋กœ์ ํŠธ ์ˆ˜ํ–‰ํ•˜๊ธฐ 
============================


https://www.adobe.com/products/firefly/features/text-to-image.html

 

AI Image Generator: Text to Image Online - Adobe Firefly

Easily create AI-generated images online for free with Adobe Firefly's new Image 3 model. Input simple text prompts and our AI image generator will do the rest.

www.adobe.com


ํ• ๋ฃจ์‹œ๋„ค์ด์…˜
- Prompt Engineering
- RAG


ํ”Œ๋ ˆ์ด ๊ทธ๋ผ์šด๋“œ ์‚ฌ์šฉํ•˜๊ธฐ
https://platform.openai.com/background

https://platform.openai.com/playground/chat?models=gpt-3.5-turbo


์ด๋ฏธ์ง€ ์ƒ์„ฑ AI ์ดํ•ด ๋ฐ ์‹ค์Šต(1), (2)
https://rowan-sail-868.notion.site/AI-fff7d480b59380dfafbcca54da08e616?pvs=74 

 

์ด๋ฏธ์ง€ ์ƒ์„ฑ AI ์ดํ•ด ๋ฐ ์‹ค์Šต | Notion

Notion ํŒ: ํŽ˜์ด์ง€๋ฅผ ์ƒ์„ฑํ•  ๋•Œ๋Š” ๋ช…ํ™•ํ•œ ์ œ๋ชฉ๊ณผ ๊ด€๋ จ๋œ ๋‚ด์šฉ์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค. ์ธ์ฆ๋œ ์ •๋ณด๋ฅผ ์‚ฌ์šฉํ•˜๊ณ , ํŽ˜์ด์ง€ ์ฃผ์ œ๋ฅผ ํ™•์‹คํžˆ ํ•˜๊ณ , ์ฃผ์š” ์ด์Šˆ์— ๋Œ€ํ•œ ์˜๊ฒฌ์„ ๊ณต์œ ํ•˜์„ธ์š”.

rowan-sail-868.notion.site

 

Prompt ์—”์ง€๋‹ˆ์–ด๋ง - ์ด๋ฏธ์ง€
 
https://rowan-sail-868.notion.site/Prompt-10e7d480b593804d9a49cdfbae95f762?pvs=74

 

Prompt ์—”์ง€๋‹ˆ์–ด๋ง - ์ด๋ฏธ์ง€ | Notion

01 ์ด๋ฏธ์ง€ ํ”„๋กฌํ”„ํŠธ ์—”์ง€๋‹ˆ์–ด๋ง ๊ฐœ์š”

rowan-sail-868.notion.site

 


Stable Diffusion Online(์Šคํ…Œ์ด๋ธ” ๋””ํ“จ์ „)

https://stablediffusionweb.com/


์ˆ˜์น˜ํ™”: ๋ฏธ๋“œ์ €๋‹ˆ ์„ฑ๋Šฅ์ด ์ข‹๋‹ค (์ˆซ์ž๋กœ ๊ฐ€์ค‘์น˜๋ฅผ ๋‘”๋‹ค) 

  • ์ˆซ์ž๋กœ ๊ฐ€์ค‘์น˜๋ฅผ ๋‘˜ ์ˆ˜ ์žˆ๋‹ค.
    • (Cute:1.5), (Puppy:0.7), (Sea:0.3)

๋‹ค๋ฅธ ์‚ฌ๋žŒ๋“ค์ด ์ƒ์„ฑํ•œ ํ”„๋กฌํ”„ํŠธ ํ™•์ธํ•ด ๋ณด๊ธฐ

CIVITAI : https://civitai.com/


2-2 ์ด๋ฏธ์ง€ ๋งŒ๋“ค์–ด๋ณด๊ธฐ

์‹ค์Šต 2-3 ๋‚˜๋งŒ์˜ ์ด๋ฏธ์ง€ ๋งŒ๋“ค๊ณ  ๊ณต์œ ํ•˜๊ธฐ


https://civitai.com/images/38668944

 

Image posted by kokumi

 

civitai.com

์Šคํ…Œ์ด๋ธ” ๋””ํ“จ์ „์— ๋„ฃ์–ด๋ณด๊ธฐ

์‚ฌ์šฉํ•œ ํ”„๋กฌํ”„ํŠธ

retro poster of a realistic cute cat in kimono kneeling,
looking at viewer,japanese text, advertise,"BUZZ ME PLEASE!" 
<lora:flux-lora-80sJPadvertise:1.0>

๊ธ์ • ํ”„๋กฌํ”„ํŠธ๋Š” ์›ํ•˜๋Š” ์ถœ๋ ฅ์˜ ํŠน์ง•์„ ์„ค๋ช…ํ•˜๋ฉฐ, ๋ถ€์ • ํ”„๋กฌํ”„ํŠธ๋Š” ์›ํ•˜๋Š” ์Šคํƒ€์ผ์ด๋‚˜ ๊ฐ์ฒด๋ฅผ ์ œ๊ฑฐํ•˜๋Š” ๋ฐฉ์‹์œผ๋กœ ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค. ๊ธ์ • ํ”„๋กฌํ”„ํŠธ๋ฅผ ํ†ตํ•ด ์›ํ•˜๋Š” ์ด๋ฏธ์ง€์˜ ๋””์ž์ธ ์š”์†Œ(์บ๋ฆญํ„ฐ, ์†Œํ’ˆ, ์ƒ‰์ƒ ๋“ฑ)๋ฅผ ๊ฐ•์กฐํ•˜๊ณ , ๋ถ€์ • ํ”„๋กฌํ”„ํŠธ๋ฅผ ํ†ตํ•ด ๋ถˆํ•„์š”ํ•œ ์š”์†Œ๋“ค์„ ๋ช…์‹œํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ฐฉ๋ฒ•์„ ํ†ตํ•ด ๋ณด๋‹ค ๋ช…ํ™•ํ•œ ๊ฒฐ๊ณผ๋ฅผ ์–ป์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.


The notation `<lora:flux-lora-80sJPadvertise:1.0>` is a prompt tag that is commonly used in text-to-image generative AI models, such as those in stable diffusion or other machine learning-based image generation models. Let's break it down:

1. **Lora**: Refers to the concept of "LoRA," which stands for "Low-Rank Adaptation." It is a specialized technique for training AI models that allows for easier and quicker training to specific styles or elements without needing the entire model to be retrained. In image generation, LoRA can be used to fine-tune a model to generate images in a certain style or theme.

2. **flux-lora-80sJPadvertise**: This is likely a specific name given to a LoRA model or a checkpoint. The name suggests a theme based on "80s Japanese advertising" (`80sJPadvertise`). It implies that this particular LoRA checkpoint is trained or fine-tuned to generate images in a style reminiscent of Japanese advertising from the 1980s, which often has a retro, vibrant, and unique aesthetic.

3. **1.0**: This number indicates the weight or influence that the specified LoRA should have in generating the image. A value of `1.0` usually means that the model will apply the LoRA exactly as intended, with full strength. A lower value like `0.5` would mean less influence, and a higher value would increase the influence.

So, in the context of your prompt:

- **"<lora:flux-lora-80sJPadvertise:1.0>"** is a directive that tells the model to use a pre-trained LoRA model named `flux-lora-80sJPadvertise` with a weight of `1.0`. This will ensure that the generated image has stylistic elements reminiscent of 1980s Japanese advertising.

In simpler terms, this part of the prompt tells the image generator to create the poster with a retro, 80s-inspired Japanese advertising vibe. The rest of the prompt ("retro poster of a realistic cute cat in kimono kneeling, looking at viewer, japanese text, advertise, 'BUZZ ME PLEASE!'") describes the content and scene, while the LoRA tag specifies the style in which the image should be created.


๊ฒฐ๊ณผ) Stable diffusion

GPT-4o 

Stable Diffusion Web์—์„œ ์Œ์„ฑ ํ”„๋กฌํ”„ํŠธ ์ถ”๊ฐ€

cartoonish, abstract, distorted, low quality, extra limbs

๊ฒฐ๊ณผ)

 



https://rowan-sail-868.notion.site/Prompt-10e7d480b593804d9a49cdfbae95f762?pvs=74

 

Prompt ์—”์ง€๋‹ˆ์–ด๋ง - ์ด๋ฏธ์ง€ | Notion

01 ์ด๋ฏธ์ง€ ํ”„๋กฌํ”„ํŠธ ์—”์ง€๋‹ˆ์–ด๋ง ๊ฐœ์š”

rowan-sail-868.notion.site

๊ฐ€์ค‘์น˜๋ฅผ 5๋ฐฐ๋ฅผ ๋ถ€์—ฌํ•˜๊ฒ ๋‹ค. 2๋ถ€ํ„ฐ 10๊นŒ์ง€ ์ž…๋ ฅ์ด ๊ฐ€๋Šฅ. (์ผ๋ฐ˜์ ์œผ๋กœ 0~10๊นŒ์ง€)

ํŠน์ • ํ‚ค์›Œ๋“œ์˜ ๊ฐ’์ด ๋†’์„ ์ˆ˜๋ก ๋‹ค๋ฅธ ํ‚ค์›Œ๋“œ๋Š” ์ด๋ฏธ์ง€์— ์ ์šฉ์ด ๋˜์ง€ ์•Š์„ ๊ฐ€๋Šฅ์„ฑ์ด ์žˆ๋‹ค.

Elegant cliff-side villa overlooking pristine coastline, 
in the style of Vincent van Gogh, rainbow color::5 --style raw

๊ฐ€์žฅ ์ค‘์š”ํ•˜๋‹ค๊ณ  ์ƒ๊ฐํ•˜๋Š” ํ”„๋กฌํ”„ํŠธ ๋‚ด์šฉ์„ ๊ฐ€์žฅ ์•ž์— ๋‘”๋‹ค.


01 ๋‚ด๊ฐ€ ๋งŒ๋“ค๊ณ ํ•˜๋Š” ์ด๋ฏธ์ง€

๊ณ ์–‘์ด (๊ณ ๋“ฑ์–ด ํƒœ๋น„, ๋Ÿฌ์‹œ์•ˆ ๋ธ”๋ฃจ) ๋‘ ๋งˆ๋ฆฌ๊ฐ€ ํ•ด๋ณ€์˜ ํ•ด๋จน์— ์žˆ์Œ 

02 Prompt

a brown tabby cat with a white fur face, 
emerald eyes and russian blue cat with emeral eyes laying down in the beige hammock 
between a palm tree in front to beautiful beach, look in the distance

์Œ์„ฑ ํ”„๋กฌํ”„ํŠธ 

cartoonish, abstract, distorted, low quality, extra limbs


03 ๋‚ด๊ฐ€ ๋งŒ๋“  ๊ฒฐ๊ณผ
(์–ด๋””๋ฅผ ์ผ๋‹ค.)

Stable diffusion web


https://www.instagram.com/ambre_ai/?hl=ko

https://www.instagram.com/shael.ai/

https://www.instagram.com/bonnycarrera/

Elegant cliff-side villa overlooking pristine coastline --style raw
Elegant cliff-side villa overlooking pristine coastline, in the style of Vincent van Gogh --style raw

์ด๊ฒƒ์œผ๋กœ ์ด๋ฏธ์ง€ ๋งŒ๋“ค์–ด์ฃผ๋ ด.

https://rowan-sail-868.notion.site/Prompt-10e7d480b593804d9a49cdfbae95f762

 

Prompt ์—”์ง€๋‹ˆ์–ด๋ง - ์ด๋ฏธ์ง€ | Notion

01 ์ด๋ฏธ์ง€ ํ”„๋กฌํ”„ํŠธ ์—”์ง€๋‹ˆ์–ด๋ง ๊ฐœ์š”

rowan-sail-868.notion.site


https://flux-ai.io/ , https://ideogram.ai/. ImageFX


Another student)
๋‚ด๊ฐ€ ๋งŒ๋“ค๊ณ ์ž ํ•˜๋Š” ์ด๋ฏธ์ง€: ๋ฏธ๋‹ˆ๋ฉ€๋ฆฌ์ฆ˜ ์Šคํƒ€์ผ์˜ ํ”ผ์•„๋…ธ ๊ทธ๋ฆผ์„ ๊ฐ„๊ฒฐํ•˜๊ณ  ์„ธ๋ จ๋˜๊ฒŒ ํ‘œํ˜„ํ•˜๊ณ  ์‹ถ์—ˆ์Šต๋‹ˆ๋‹ค
ํ”„๋กฌํ”„ํŠธ: a one liners drawing of piano, minimalistic elements, abstract whispers, tondo, minimalistic design, kurzgesagt, calligraphic lines

https://rowan-sail-868.notion.site/AI-d47af143356d48028f66ac299344988f?pvs=74

 

์ƒ์„ฑ AI ์ดํ•ดํ•˜๊ธฐ | Notion

์ˆ˜์—… ์ค€๋น„

rowan-sail-868.notion.site

https://ldjwj.github.io/CHATGPT_AI_CLASS/LLM%EC%8B%9C%EC%9E%91%ED%95%98%EA%B8%B0_V10_wc_PDF.pdf
ํ† ํฐํ™”๋ž€ ๋ฌธ์žฅ์ด๋‚˜ ๋ฌธ์„œ๋ฅผ ์˜๋ฏธ ์žˆ๋Š” ๋‹จ์œ„๋กœ ๋ถ„ํ• ํ•˜๋Š” ๊ณผ์ •

์ž„๋ฒ ๋”ฉ: ๋‹จ์–ด๋‚˜ ๋ฌธ์žฅ๊ณผ ๊ฐ™์€ ์ด์‚ฐ์ ์ธ ๋ฐ์ดํ„ฐ๋ฅผ ์—ฐ์†์ ์ธ ๋ฒกํ„ฐ ๊ณต๊ฐ„์œผ๋กœ ๋ณ€ํ™˜ํ•˜๋Š” ๊ธฐ์ˆ  

๋Œ€๋Ÿ‰์˜ ๋ฐ์ดํ„ฐ๋ฅผ ๋ชจ๋ธ๋กœ ํ•™์Šต์‹œํ‚ค๋Š” ์‚ฌ์ „ํ•™์Šต(pre-training).

๊ณต๊ณต ์™€์ดํŒŒ์ด ์ •๋ณด ์œ ์ถœ ๊ฐ€๋Šฅ

๋…ผ๋ฌธ : Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks(2020๋…„ 10์›” ๋ฐœํ‘œ). Facebook AI Research์˜ ์—ฐ๊ตฌ์ž ๋ฉค๋ฒ„

https://arxiv.org/abs/2005.11401

 

Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

Large pre-trained language models have been shown to store factual knowledge in their parameters, and achieve state-of-the-art results when fine-tuned on downstream NLP tasks. However, their ability to access and precisely manipulate knowledge is still lim

arxiv.org

API๋Š” ํ•œ ํ”„๋กœ๊ทธ๋žจ์ด ๋‹ค๋ฅธ ํ”„๋กœ๊ทธ๋žจ์˜ ๊ธฐ๋Šฅ์„ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋„๋ก ๋„์™€์ฃผ๋Š” "๋‹ค๋ฆฌ" ์—ญํ• .

๋ฐ์ดํ„ฐ๋ฅผ ์ธํ„ฐ๋„ท์œผ๋กœ ๊ฐ€์ ธ์˜ด
์‹๋‹น ์ฃผ๋ฌธ -> ์‹๋‹น ์ „๋‹ฌ (API) -> ์ฃผ๋ฐฉ ์กฐ๋ฆฌ (Python) -> ํ…Œ์ด๋ธ”


Google Scholar: ๋‹ค์–‘ํ•œ ํ•™์ˆ  ์ž๋ฃŒ๋ฅผ ๊ฒ€์ƒ‰ํ•  ์ˆ˜ ์žˆ๋Š” ๋Œ€ํ‘œ์ ์ธ ์‚ฌ์ดํŠธ์ž…๋‹ˆ๋‹ค. ํŠน์ • ์—ฐ๊ตฌ ๋ถ„์•ผ๋‚˜ ๋…ผ๋ฌธ์„ ์ฐพ๊ธฐ์— ์œ ์šฉํ•˜๋ฉฐ, ๋‹ค์–‘ํ•œ ํ•„ํ„ฐ๋ง ๊ธฐ๋Šฅ์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.

ArXiv: AI, ๋จธ์‹ ๋Ÿฌ๋‹, ์ปดํ“จํ„ฐ ๋น„์ „, ์ž์—ฐ์–ด ์ฒ˜๋ฆฌ์™€ ๊ฐ™์€ ์ปดํ“จํ„ฐ ๊ณผํ•™ ์—ฐ๊ตฌ์— ๋Œ€ํ•œ ์ตœ์‹  ๋…ผ๋ฌธ์„ ๊ณต๊ฐœํ•˜๋Š” ์‚ฌ์ดํŠธ์ž…๋‹ˆ๋‹ค. ํŠนํžˆ, RAG์™€ ๊ฐ™์€ ์ตœ์‹  ๊ธฐ์ˆ ์— ๋Œ€ํ•œ ๋…ผ๋ฌธ์„ ์‰ฝ๊ฒŒ ์ฐพ์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

Semantic Scholar: AI ๊ธฐ๋ฐ˜ ํ•™์ˆ  ๋…ผ๋ฌธ ๊ฒ€์ƒ‰ ์—”์ง„์œผ๋กœ, ๋…ผ๋ฌธ ์ถ”์ฒœ, ์ธ์šฉ ์ˆ˜ ๋ถ„์„, ๊ด€๋ จ ์—ฐ๊ตฌ ์ž๋ฃŒ ๋“ฑ์„ ํšจ๊ณผ์ ์œผ๋กœ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.

PubMed: ์˜ํ•™๊ณผ ์ƒ๋ฌผํ•™ ์—ฐ๊ตฌ์— ์ค‘์ ์„ ๋‘๊ณ  ์žˆ์ง€๋งŒ, AI์™€ ๊ด€๋ จ๋œ ์ผ๋ถ€ ๋…ผ๋ฌธ๋„ ์ฐพ์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

728x90
๋ฐ˜์‘ํ˜•
Comments