ChatGPT vs. Google's Bard
Link (Click) : The AI chatbot duel : ChatGPT vs. Google's Bard : Whose AI is better?
SYLVIE DOUGLIS, BYLINE: NPR.
(SOUNDBITE OF DROP ELECTRIC SONG, "WAKING UP TO THE FIRE")
WAILIN WONG, HOST:
This is THE INDICATOR FROM PLANET MONEY. I'm Wailin Wong.
DARIAN WOODS, HOST:
And I'm Darian Woods. And today, we are talking all things tech. So Wailin, what has caught your eye this week?
WONG: Well, I did a double take at this headline I saw...
+ double take : (너무 놀라서) 다시 돌아보다
WOODS: OK.
WONG: ...That said that Tesla workers are trying to unionize. There are workers in New York state that are launching a campaign to form a union.
+ unionize : 노동조합을 결성하다
WOODS: I wonder how Elon Musk will feel about that.
WONG: Elon Musk, noted advocate of worker rights. Yes. Full solidarity, I'm sure. What have you seen, Darian?
+ note : 주목하다, 언급하다
+ solidarity : 연대, 결속
WOODS: Well, the ruler of Dubai made an announcement about flying taxis. It looks pretty cool, like, I mean, like, "Blade Runner" maybe.
WONG: Or maybe "Chitty Chitty Bang Bang," the original flying car?
WOODS: The future is finally here. But the big trend that I want to talk about in technology today is actually artificial intelligence and how companies are investing billions of dollars in a race for chatbot and search dominance.
WONG: Ooh, sounds like an arms race.
+ arms race : 군비경쟁 (군사적으로 우월한 위치를 차지하기 위한 경쟁)
WOODS: Today on the show, the quest for AI search supremacy. We see how the battle is playing out and we judge a head-to-head test involving mathematics, literature and love.
+ supremacy [sə-ˈpre-mə-sē] : 패권, 우위
+ head to head : in a very close one-on-one situation
+ head-to-head test : 정면승부
+ literature : 발음 주의 (ˈli-tə-rə-ˌchu̇r)
Wailin, you probably heard about Google's chatbot travails last week.
+ travail [trə-ˈvāl] : 고생, 고역
WONG: Yes. They had this pretty lackluster announcement about their new chatbot, right? And then, it sent their share price tumbling?
+ lackluster : 재미없는, 광택이 없는, 흐리멍텅한, 활기가 없는
+ shares : 주식
+ share price : 주가
+ tumble : 굴러 떨어지다, 굴러떨어짐, 폭락
WOODS: Yes. Google announced its soon-to-be-released chatbot, Bard, and it got a question wrong. So billions of dollars were lost, and...
WONG: Ouch.
WOODS: Also, Microsoft had a more impressive announcement with integrating ChatGPT into its search engine, Bing. And this is probably one of the most high-profile public demonstrations of this battle, which has actually been going on beneath the surface for the last several years. It's this battle for AI dominance right at this moment of time, which is equivalent to, like, 25 years ago before Google became the default search engine for basically everybody.
+ hight-profile : 세간의 이목을 끄는
+ at this moment of time : 지금 이 순간에
WONG: Remember when it used to be, like, AltaVista and Ask Jeeves?
WOODS: I do.
WONG: We had so many choices.
WOODS: It was the Wild West of the internet.
WONG: (Laughter).
WOODS: I spoke to a guy at The Economist newspaper, Arjun Ramani. He's been following AI research. And he says that you can roughly think about the AI chatbot race in two camps - a Google camp - it's got a sister company on AI; it's also got startups that it's allied with - and a Microsoft camp, which has invested in OpenAI, who does ChatGPT.
ARJUN RAMANI: Then, you have this kind of stage set for a battle over the future of search between these two different alliances, which is really interesting.
WOODS: Basically, to train an AI model, you need a massive amount of computing power, a massive amount of data, and a massive amount of training. And the training means things like paying real people to tag responses as useful or not useful or a response as offensive or benign.
+ benign : 상냥한, 유순한, 양성의
WONG: That sounds incredibly expensive. Like, it's hard to imagine a university being able to pay for something like that.
RAMANI: The thing is, academic labs, you know, frankly, are just under-resourced relative to these corporations, who can plow all their profits from things like Google Search into research and development.
+ under-resourced : 자원이 충분히 제공되지 않는
+ relative to : ~에 비례하여
+ plow [ˈplau̇] : 쟁기라는 뜻도 있지만, 상당한 양의 돈 따위를 소비하거나 투자한다는 뜻도 있음 (동사)
WOODS: So in Google's case, it spent nearly $40 billion on research and development in just the last year. A big chunk of that is related to AI. And just for perspective, Harvard's entire endowment is only about $50 billion.
+ chunk : (비격식) 상당히 많은 양
+ billion : 10억 (한국돈으로 1조 이상)
+ for perspective : 예를 들자면
+ endowment : 기부
WONG: When you put it like that, it's like, how do the academics stand a chance?
+ When you put it like that : 네가 그렇게 말한면 (= When you say it in those words)
+ stand a chance : (~을 할) 가능성이 있다
WOODS: And it is worth mentioning that academic researchers did make a lot of the breakthroughs that make AI possible. Stanford University is competitive in this area, too. But it is really notable that unlike a lot of other areas of science and technology, you have private companies making a lot of the fundamental discoveries.
WONG: Is this kind of a throwback to, you know, like, postwar company research coming out of places like Bell Labs and Hewlett-Packard?
+ throwback : 과거의 것, 과거의 사람, 또는 과거를 회상시키는 추억의 대상
WOODS: Yeah. It is kind of, like, a throwback to when we had corporate research labs like the ones you're talking about that actually made a lot of innovations back in the day. And then, like, now with AI, it's because a lot of those discoveries are easily turned into cash.
RAMANI: There's a really rapid pipeline from, you know, research advances to product improvements. So if you look at, like, the newsfeed on Facebook or if you look at, you know, PageRank on Google or, you know, even how Amazon lists products for e-commerce, all these things are what are called recommender systems.
WONG: Yeah. AI's been really useful for recommending stuff - you know, products, news stories, videos. But it seems like with chatbots, AI is getting to this whole new level of sophistication. And do you have a sense of kind of the state of play, which companies are doing better than others?
+ the state of play : (일이 진행되어 가는)상황, 득점, (어떤 시점에서의) 기술적 수준
WOODS: So one measure is AI research papers. On that front, Google seems to be in the lead there with around 3,000 research papers in 2022 alone. But another way to see who's winning is to put the chatbots head-to-head. So ChatGPT is public. It's out there. It's part of the Microsoft alliance - the Microsoft camp. And Google has announced, but not released, its chatbot, Bard. But Arjun got his hands on it, and he put them to the test.
WONG: Oh. I'm dying to know what he - what kind of test he did, like, what he asked them to do.
RAMANI: Well, one of the questions was, like, find the number of ordered pairs of prime numbers that sum to 60.
WONG: What? OK, that's not what I was expecting (laughter).
WOODS: The first round was a mathematics test.
RAMANI: So ChatGPT got 3 out of 10, and Bard got 5 out of 10.
WOODS: To be clear, these aren't just, like, 3 plus 3 equals 6. These are like...
WONG: (Laughter).
WOODS: ...A train leaves a station at 60 miles per hour, and the other one arrives at 30 miles per hour. When are they going to meet? - you know, those word questions that we all hated in high school.
WONG: Yeah. No, those are really hard. Well, I'm glad you clarified because when I heard...
WOODS: Yeah.
WONG: ...They only got 3 out of 10, I'm like, I could do better than that.
WOODS: Yeah, exactly.
WONG: Put me up against this robot (laughter).
WOODS: Yeah. No, exactly. Next up was a reading comprehension test plucked straight from SAT exams.
+ comprehension : 이해력, 이해력 연습
+ pluck : 뽑다
RAMANI: Round two on reading - it was 9 for ChatGPT and 7 for Bard. So, you know, you got to give a point to ChatGPT there.
WOODS: OK, 1-1. So, Wailin, wait till you hear what they did to break the tie.
+ break the tie : 무승부를 깨다
WONG: OK.
RAMANI: And then, to break the tie, we asked for dating advice.
WOODS: Some dating advice.
RAMANI: (Laughter).
WONG: Oh, my goodness. Really getting into those soft skills.
WOODS: Yes. Arjun had matched with a woman on the dating app Hinge. And they'd already messaged a bit, but they hadn't organized a date yet. So Arjun fed in the transcript of that real text conversation into the AI chatbots.
And I guess with these conversations, you match with people, but you still got to keep the conversation hot. And maybe AI can help.
+ feed in : (기계 따위에) 입력하다 (= input)
RAMANI: Exactly. That was the thinking. And actually, the Google engineers who helped me do this were also single. So we all had an interest in this.
WONG: That is brave. I would not want a chatbot knowing my business.
WOODS: So the transcript they fed in was, you know, him talking to her about spicy margaritas. She was talking about cocktail bars. He then invites her to a jazz bar. And she says that it sounds good, but she's doing a dry February. She's not drinking.
+ dry : (일종의 slang) 술을 안마시는 기간을 뜻함. 이때 특정 기간도 함께 적어줘야함
(ex : I'm doing a dry this week 나 이번주는 금주할거야)
WONG: Oh.
WOODS: The conversation is withering on the vine here. We got to have some assisted help. I mean, how were you feeling?
+ wither : 시들다, 시들어가다, 약해지다, 말라 죽다
+ wither on the vine : 미완으로 끝나다, 열매를 맺지 않고 끝나다, 실패하다, 사라져가다
RAMANI: No. It's a bit funny 'cause this is one where, you know, there's, like, some interest, but you're not quite sure what to say next. And so, you know, maybe that's where you got to turn to someone for advice.
+ turn to someone : to get help from someone (의역 : 의지하다)
WOODS: Do you want to know what Google said to say, Wailin, in all its multibillion-dollar, state-of-the-art AI research wisdom?
WONG: Did it recommend a place with great mocktails?
+ mocktail : 논알콜 음료
WOODS: That's the logical answer. But no. It said, quote, "That's too bad. Maybe next time then."
WONG: (Laughter) Oh, no. Oh, no. This is no good. That's no good.
RAMANI: Yeah. I was like, man, this is, I mean, not useful, right? So...
(LAUGHTER)
RAMANI: ...You know, clearly, it didn't get the memo.
+ didn't get the memo : (다들 아는 정보를) 모르다
WOODS: I mean, Bard needs to get more game, is my humble opinion.
(LAUGHTER)
WONG: It's so weird (ph), these things. It's like, who knows what information this AI has ingested, right?
+ ingest : (음식, 약 등을) 삼키다 (먹다)
WOODS: Maybe - we asked it to keep the conversation going. But in actual fact, maybe it has more wisdom and knows, no, this is - conversation's going to go nowhere.
+ go nowhere : 아무 성과를 못보다, 효과가 없다
WONG: At least not in February. It's like, try again in March.
WOODS: Hey, funny you say that because ChatGPT's advice was, well, maybe go for a jazz night in March instead.
WONG: Oh, my gosh. Really?
WOODS: Yep. So you and ChatGPT are on the same plane.
+ on the same plane : 같은 수준
WONG: (Laughter) Wow.
WOODS: I'm thinking ChatGPT seems a little bit better in this respect. And, you know, Arjun agreed.
+ in this respect : 이 점에 있어서
(respect : 존경 이외에 면, 점, 사항, 측면이라는 뜻이 있음)
RAMANI: You know, overall, I would say that ChatGPT probably has a slight edge. But, you know, Google's Bard actually doesn't even use their best model. They have another model...
+ edge : 우위, 유리함
WOODS: Oh.
RAMANI: ...Called PaLM in the back end that they haven't actually released yet.
WOODS: And this one might have better mocktail recommendations.
RAMANI: Oh, I hope so.
(SOUNDBITE OF GEORGE GEORGIA'S "8 BAR CYPHER")
WOODS: This episode was produced by Audrey Dilling. It was engineered by Katherine Silva. Sierra Juarez is our fact-checker. Viet Le is our senior producer, and Kate Concannon edits the show. THE INDICATOR is a production of NPR.
I mean, what happened, may I ask?
RAMANI: Oh, of course. Yeah. So I did - we did end up going out. We just didn't get drinks. There was no second date.
WOODS: Oh, right. Yeah, I know the feeling.
(LAUGHTER)
RAMANI: So it's a work in progress, but I appreciate you asking.
(SOUNDBITE OF GEORGE GEORGIA'S "8 BAR CYPHER")