https://www.wir1ed.com/story/deepmind-ceo-demis-hassabis-interview-artificial-intelligence-scale/?utm_brand=w1ired&mbid=social_twitter&utm_social-type=owned
์๋๋ ์ผ๋ถ AI ๋ฒ์ญ
Q
In the last few years, increasing the amount of computer power and data used in training an AI model is the thing that has driven amazing advances. Sam Altman is said to be looking to raise up to $7 trillion for more AI chips. Is vastly more computer power the thing that will unlock artificial general intelligence?
์ง๋ ๋ช ๋
๋์ AI ๋ชจ๋ธ ํ์ต์ ์ฌ์ฉ๋๋ ์ปดํจํฐ ์ฑ๋ฅ๊ณผ ๋ฐ์ดํฐ์ ์์ ๋๋ฆฌ๋ ๊ฒ์ด ๋๋ผ์ด ๋ฐ์ ์ ๊ฐ์ ธ์จ ์๋๋ ฅ์ด์์ต๋๋ค. ์ ์ํธ๋จผ์ ๋ ๋ง์ AI ์นฉ์ ์ํด ์ต๋ 7์กฐ ๋ฌ๋ฌ๊น์ง ๋ชจ๊ธํ ๊ณํ์ด๋ผ๊ณ ํฉ๋๋ค. ๋ ๋ง์ ์ปดํจํฐ ์ฑ๋ฅ์ด ์ธ๊ณต ์ผ๋ฐ ์ง๋ฅ์ ์คํํ ์ ์๋ ์ด์ ์ผ๊น์?
A
Was that a misquote? I heard someone say that maybe it was yen or something. Well, look, you do need scale; that's why Nvidia is worth what it is today. Thatโs why Sam is trying to raise whatever the real number is. But I think we're a little bit different to a lot of these other organizations in that we've always been fundamental research first. At Google Research and Brain and DeepMind, we've invented the majority of machine learning techniques we're all using today, over the last 10 years of pioneering work. So thatโs always been in our DNA, and we have quite a lot of senior research scientists that maybe other orgs don't have. These other startups and even big companies have a high proportion of engineering to research science.
์๋ชป ์ธ์ฉํ ๊ฑด๊ฐ์? ๋๊ตฐ๊ฐ ์ํ ๊ฐ์ ๊ฑฐ๋ผ๊ณ ๋งํ๋ ๊ฑธ ๋ค์์ด์. ํ์ง๋ง ๊ท๋ชจ๋ฅผ ํค์ฐ๋ ค๋ฉด ๊ท๋ชจ๋ฅผ ํค์์ผ ํ์ฃ . ๊ทธ๋์ ์ค๋๋ ์๋น๋์์ ๊ฐ์น๊ฐ ์๋ ๊ฒ๋๋ค. ๊ทธ๋์ ์์ ์ค์ ์ซ์๊ฐ ์ผ๋ง๋ ์ฌ๋ฆฌ๋ ค๊ณ ๋
ธ๋ ฅํ๊ณ ์์ต๋๋ค. ํ์ง๋ง ์ฐ๋ฆฌ๋ ํญ์ ๊ธฐ์ด ์ฐ๊ตฌ๋ฅผ ์ฐ์ ์ํด ์๋ค๋ ์ ์์ ๋ค๋ฅธ ๋ง์ ์กฐ์ง๊ณผ ์กฐ๊ธ ๋ค๋ฅด๋ค๊ณ ์๊ฐํฉ๋๋ค. ๊ตฌ๊ธ ๋ฆฌ์์น์ ๋ธ๋ ์ธ, ๋ฅ๋ง์ธ๋์์๋ ์ง๋ 10๋
๊ฐ์ ์ ๊ตฌ์ ์ธ ์ฐ๊ตฌ๋ฅผ ํตํด ์ค๋๋ ์ฐ๋ฆฌ๊ฐ ์ฌ์ฉํ๊ณ ์๋ ๋๋ถ๋ถ์ ๋จธ์ ๋ฌ๋ ๊ธฐ์ ์ ๋ฐ๋ช
ํ์ต๋๋ค. ์ด๋ ํญ์ Google์ DNA์ ๋ด์ฌ๋์ด ์์ผ๋ฉฐ, ๋ค๋ฅธ ์กฐ์ง์๋ ์๋ ๋ง์ ์ ์ ์ฐ๊ตฌ ๊ณผํ์๋ฅผ ๋ณด์ ํ๊ณ ์์ต๋๋ค. ๋ค๋ฅธ ์คํํธ์
์ด๋ ๋๊ธฐ์
์ ์ฐ๊ตฌ ๊ณผํ์์ ๊ณตํ์ด ์ฐจ์งํ๋ ๋น์ค์ด ๋์ต๋๋ค.
Q
Does that mean that the competition between AI companies going forward will increasingly be around tool use and agentsโAI that does things rather than just chats? OpenAI is reportedly working on this.
๊ทธ๋ ๋ค๋ฉด ์์ผ๋ก AI ๊ธฐ์
๊ฐ์ ๊ฒฝ์์ ์ ์ ๋ ๋๊ตฌ ์ฌ์ฉ๊ณผ ์์ด์ ํธ, ์ฆ ๋จ์ํ ์ฑํ
์ด ์๋ ์ผ์ ์ฒ๋ฆฌํ๋ AI๋ฅผ ๋๋ฌ์ผ ๊ฒฝ์์ด ๋ ๊น์? OpenAI๊ฐ ์ด๋ฅผ ์ํด ๋
ธ๋ ฅํ๊ณ ์๋ ๊ฒ์ผ๋ก ์๋ ค์ก์ต๋๋ค.
A
Probably. Weโve been on that track for a long time; thatโs our bread and butter really, agents, reinforcement learning, and planning, since the AlphaGo days. [In 2016 DeepMind developed a breakthrough algorithm capable of solving complex problems and playing sophisticated games.] Weโre dusting off a lot of ideas, thinking of some kind of combination of AlphaGo capabilities built on top of these large models. Introspection and planning capabilities will help with things like hallucination, I think.
It's sort of funny, if you say โTake more careโ or โLine out your reasoning,โ sometimes the model does better. What's going on there is you are priming it to sort of be a little bit more logical about its steps. But you'd rather that be a systematic thing that the system is doing.
This definitely is a huge area. We're investing a lot of time and energy into that area, and we think that it will be a step change in capabilities of these types of systemsโwhen they start becoming more agent-like. Weโre investing heavily in that direction, and I imagine others are as well.
์๋ง๋์. ์ ํฌ๋ ์ํ๊ณ ์์ ๋ถํฐ ์์ด์ ํธ, ๊ฐํ ํ์ต, ๊ณํ ์๋ฆฝ ๋ฑ ์ด ๋ถ์ผ๋ฅผ ์ค๋ซ๋์ ์ฐ๊ตฌํด ์์ต๋๋ค. [2016๋
๋ฅ๋ง์ธ๋๋ ๋ณต์กํ ๋ฌธ์ ๋ฅผ ํด๊ฒฐํ๊ณ ์ ๊ตํ ๊ฒ์์ ํ๋ ์ดํ ์ ์๋ ํ๊ธฐ์ ์ธ ์๊ณ ๋ฆฌ์ฆ์ ๊ฐ๋ฐํ์ต๋๋ค.] ์ ํฌ๋ ์ด๋ฌํ ๋๊ท๋ชจ ๋ชจ๋ธ ์์ ๊ตฌ์ถ๋ ์ํ๊ณ ๊ธฐ๋ฅ์ ์กฐํฉ์ ์๊ฐํ๋ฉฐ ๋ง์ ์์ด๋์ด๋ฅผ ๋ ์ฌ๋ฆฌ๊ณ ์์ต๋๋ค. ์ฑ์ฐฐ๊ณผ ๊ณํ ๋ฅ๋ ฅ์ด ํ๊ฐ๊ณผ ๊ฐ์ ๊ธฐ๋ฅ์ ๋์์ด ๋ ๊ฒ์ด๋ผ๊ณ ์๊ฐํฉ๋๋ค.
"์ข ๋ ์ฃผ์ํด" ๋๋ "์ถ๋ก ์ ์ ๋ฆฌํด"๋ผ๊ณ ๋งํ๋ฉด ๋ชจ๋ธ์ด ๋ ์ํ ๋๊ฐ ์์ต๋๋ค. ๊ทธ ๋จ๊ณ์ ๋ํด ์ข ๋ ๋
ผ๋ฆฌ์ ์ผ๋ก ์ค๋ช
ํ ์ ์๋๋ก ์ค๋น์ํค๋ ๊ฒ์ด์ฃ . ํ์ง๋ง ๊ทธ๋ณด๋ค๋ ์์คํ
์ด ์ฒด๊ณ์ ์ผ๋ก ์ฒ๋ฆฌํ๋ ๊ฒ์ด ์ข์ต๋๋ค.
์ด๊ฒ์ ํ์คํ ๊ฑฐ๋ํ ์์ญ์
๋๋ค. ์ ํฌ๋ ์ด ๋ถ์ผ์ ๋ง์ ์๊ฐ๊ณผ ์๋์ง๋ฅผ ํฌ์ํ๊ณ ์์ผ๋ฉฐ, ์ด๋ฌํ ์ ํ์ ์์คํ
์ด ์์ด์ ํธ์ฒ๋ผ ๋๊ธฐ ์์ํ๋ฉด ๊ทธ ๊ธฐ๋ฅ์ด ํ ๋จ๊ณ ๋ ๋ฐ์ ํ ๊ฒ์ด๋ผ๊ณ ์๊ฐํฉ๋๋ค. ์ ํฌ๋ ๊ทธ ๋ฐฉํฅ์ ๋ง์ ํฌ์๋ฅผ ํ๊ณ ์์ผ๋ฉฐ ๋ค๋ฅธ ๊ธฐ์
๋ค๋ ๋ง์ฐฌ๊ฐ์ง์ผ ๊ฒ์ด๋ผ๊ณ ์๊ฐํฉ๋๋ค.
์ ์ฒด 3์ค์์ฝ
์ค์ผ์ผ๋ง์ผ๋ก ์๋๊ณ ์ฐ๋ฆฌ๋ ์๊ณ ๋ฆฌ์ฆ ๊ฐ์ ๋ถ๋ถ์์ ๋ฐ์ด๋๋ค
์์ด์ ํธ AI๊ฐ ์ ๋ฐฉํฅ์ฑ์ด๊ณ ์ฐ๋ฆฌ๋ ๊ทธ๊ฑฐ ๋ค ์ค๋น๋ผ์๋ค
์์ด์ ํธ AI๋ ์ค์ค๋ก ์ธํฐ๋ท๊ณผ ํ์ค์ ์ค๊ฐ๋ฉฐ ํ์ตํ ๊ฒ์ด๋ค (AGI์ ๋ฐํ?)
๋๊ธ 0