Frontier Feud

Talk transcript

1. 0:00

   [upbeat music] Welcome to Frontier Feud.

2. 0:19

   So I'm your host, uh, Barr Yaron. I'm a partner at Amplify. We're the first investors in technical founders, and I invest in data and AI companies. I'm very excited to be here frontier feuding- [laughing] ...

3. 0:32

   in my favorite city, New York City. And today on stage, we have some amazing folks competing for prizes and eternal glory, and all that good stuff. So we're gonna introduce our teams.

4. 0:44

   Uh, do not let these smiling faces deceive you. We are here to compete. Uh, before we do intros for the teams, just an audience feeler, who here in the audience, including those on stage, have watched Family Feud before?

5. 0:57

   Quick show of hands. Okay, most people. For those of you, uh, who haven't, the premise is pretty simple. The nuances you'll learn with the game. So we surveyed 100 AI engineers on a series of questions.

6. 1:13

   Folks on stage are going to guess the answers to those questions, and the most popular answers from the survey are going to get the most points. So feel free to follow along and guess yourself in the audience as we go along.

7. 1:26

   Uh, if you don't like the answers, you can look to your left and look to your right and blame your neighbors. [laughing] No, I'm just kidding. In all seriousness, thank you so much to everyone who filled out that survey before.

8. 1:36

   And so we're gonna do some quick intros. Um, we have on my left, Mihir, you can kick it off. [laughs] Tell us who you are, what you do, and your tech hot take.

9. 1:44

   We're gonna do quick intros.

10. 1:47

    Cool. Uh, hey, everyone, my name is Mihir. Uh, I work at Anthropic. Um, my tech hot take is, I think, uh, let's say of the five-ish people training big models today, at least one of them will no longer be training AI models by the end of the year.

11. 2:04

    Mic drop, but really mic pass. [laughs] So, John. [laughs]

12. 2:08

    Hi, guys. I'm John. I work at Anthropic, and my tech hot take is, I think AIs make really good therapists. [laughs] [laughs] [laughs]

13. 2:19

    Hi, I'm Tina. Um, I work at Reflection AI. It's a startup building coding agents. And my tech hot take is that everybody working in our industry should spend, like, 20 minutes a day with just a piece of paper and a pencil, um, and think, like, non-AI interrupted thoughts. [laughs] [applause]

14. 2:35

    Thank you. [laughs]

15. 2:38

    Thanks. [laughs]

16. 2:40

    Hi, I'm Shresta. I'm the product lead for the Gemini developer APIs working with Paige. Uh, I have two. My first tech hot take is a follow-up from Tina's. I think it'll be important for people to do what Tina said because w- eventually when you have AI models representing you, they'll need some good content to be trained on. [laughs] [laughs]

17. 3:00

    I think I'll save my next one for later. [laughs]

18. 3:04

    I love it.

19. 3:05

    Excellent.

20. 3:05

    Paige, you wanna start your side?

21. 3:07

    Heck yeah. Greetings, everyone. Um, we on the stage did not get to vote on the survey, so just be aware that, uh, we did not pre-game it.

22. 3:16

    No fraud.

23. 3:17

    No fraud. Um, uh, my name is Paige. I work at Google DeepMind leading, uh, engineering for AI DevRel team. Um, uh, and I guess my tech hot take is that, uh, speed forward ahead a year and a half, I think that the majority of deployed models will be on device, um, and we'll be seeing a trend more

24. 3:40

    towards smaller models, um, that are, uh, kind of orchestrated together, maybe, uh, hyper-specialized for a specific task, um, as opposed to relying just on larger models, um, and sending data elsewhere to do interesting things.

25. 3:55

    Sweet.

26. 3:57

    Uh, my name's Colin. Uh, you might remember me from my talk earlier today. I'm a researcher at Augment Code building AI coding tools. Before that, I was at FAIR, Facebook AI Research here in New York on, working on AI for board games.

27. 4:09

    I, I just had to get out both hot takes 'cause I had two-

28. 4:11

    Mm-hmm

29. 4:11

    ... two ones. So one is that in the future, dating apps will be our AIs dating each other. Um, two is I don't think transformers are the final architecture because eventually we'll build these models using biological materials. [laughs]

30. 4:26

    Oh, snap.

31. 4:28

    Great.

32. 4:28

    Hi, my name is Petra. I'm the product manager of factuality for the AI answer at the top of the Google Search page.

33. 4:35

    Ooh.

34. 4:35

    My hot take is that, um, chatting with, uh, one bot one-on-one is boring, and in the future, most conversations will have at least two other bots in the con- in the room with you. [laughs]

35. 4:47

    I love it.

36. 4:49

    Hello. Uh, I'm Steven. Uh, I'm joining Thinking Machines soon. Um, and my hot take is that if you think we're hitting a token wall, that sounds like a skill issue to me. [laughs] [laughs]

37. 5:02

    I love it. Amazing. So with that, let's get started. I'm, I'm gonna have Paige and Mihir join me, not that you're so far from me.

38. 5:09

    Excellent.

39. 5:10

    Great.

40. 5:11

    So the objective is, like, hit this as fast as possible, right?

41. 5:15

    Yes.

42. 5:15

    Okay.

43. 5:15

    So, so, um- [laughs] [laughs] So whoever buzzes first has the first chance to answer this question. So I'm going to ask a question, and we'll take it from here. So we asked 100 AI engineers, name the most influential AI researcher.

44. 5:32

    God dang it. [laughs] I looked at the screen.

45. 5:34

    Okay, what's your, what's your guess?

46. 5:36

    Uh, Ilya Sutskever.

47. 5:38

    Ooh.

48. 5:39

    Ooh. [chime]

49. 5:40

    Okay, we have Ilya.

50. 5:43

    Noam Shazeer.

51. 5:44

    Noam Shazeer.

52. 5:45

    That's calm.

53. 5:45

    Very influential, but not on the board. Not in our-

54. 5:48

    Oh.

55. 5:48

    Oh, wow

56. 5:48

    ... not in our top of answers.

57. 5:49

    Oh.

58. 5:49

    So your team decides if to play or pass.

59. 5:52

    You guys taking this? Yeah? All right, we're gonna play. We're gonna play.

60. 5:55

    You're gonna play. You're gonna play. Okay, awesome. So John, it's, it's your turn to guess someone.

61. 5:59

    I'll guess another person.

62. 6:00

    But you can stay there. [laughs]

63. 6:02

    Okay. Um-

64. 6:02

    We're working out the kinks

65. 6:10

    Yeah. [laughs] Uh, Andrej Karpathy. Yeah.

66. 6:13

    Oh, yeah. [chime] [clapping]

67. 6:16

    Oh.

68. 6:16

    Answer. Good answer.

69. 6:17

    Thank you.

70. 6:19

    All right, Tina, how about you?

71. 6:21

    Geoffrey Hinton.

72. 6:22

    Oh.

73. 6:24

    Ooh.

74. 6:24

    An excellent one, but [buzzer] it's not on the board.

75. 6:29

    How? That's so... Geoff, how? Like-

76. 6:31

    By influencers. [laughs]

77. 6:34

    Oh.

78. 6:34

    Oh, sorry. Did you say Geoffrey Hinton?

79. 6:36

    Yeah.

80. 6:37

    Yeah. Oh. [chime] [clapping] [laughs] That's on me. Um, I thought you said something else. Great.

81. 6:47

    I... The p- the, the lead author on the transformer paper. Uh, the Attention All You Need paper. Uh-

82. 6:53

    We need, we need a specific name.

83. 6:55

    Sorry. Um. Sorry, I'm having a literal, like I know the name- [laughs] ... but I, uh, I'm having a literal, like,

84. 7:07

    uh, uh-

85. 7:08

    We're gonna have-

86. 7:10

    Options

87. 7:11

    Eh.

88. 7:11

    Yeah, I think, eh, but we, we, you already had a strike that doesn't count, so we're gonna-

89. 7:14

    Okay

90. 7:14

    ... we keep it. We're netting at one. Go, Mihir.

91. 7:19

    Okay. Uh, [clears throat] I'm gonna say Yann LeCun.

92. 7:22

    Yann LeCun. We do have Yann LeCun. [chime]

93. 7:26

    Okay. All right. [clapping] Why do we keep going? We just keep going.

94. 7:29

    Because you're winning.

95. 7:30

    Oh, until we lose? Yeah, yeah.

96. 7:31

    Yeah, until you lose.

97. 7:32

    Oh, I didn't, I, I didn't- [laughs] Uh, wait.

98. 7:36

    If you wanna lose-

99. 7:37

    Oh

100. 7:37

     ... here's your opportunity.

101. 7:37

     Ashish Vaswani?

102. 7:39

     Uh, no.

103. 7:41

     Oh, okay. [buzzer]

104. 7:41

     So not on the board.

105. 7:42

     Okay. [laughs]

106. 7:46

     Um, what was the question again? [laughs]

107. 7:49

     Good question. So name the most influential AI researcher.

108. 7:53

     Um, I'm gonna say Sam Altman. [laughs]

109. 7:55

     Sam Altman. It's a good one. It's not on the board.

110. 7:58

     Oh, that was ironic.

111. 7:58

     And so- [buzzer]

112. 7:59

     Sorry

113. 8:00

     ... we are on three strikes, so your team needs to deliberate, and if you pick an answer, you actually steal the points.

114. 8:07

     Well, why did she...

115. 8:08

     She...

116. 8:16

     We have four options on the board. And for those who are joining, the question is, we asked 100 AI engineers, name the most influential AI researcher.

117. 8:33

     All right, I'm giving you, uh, five, four-

118. 8:37

     We're good. We're ready.

119. 8:38

     You're good?

120. 8:38

     Yeah.

121. 8:38

     All right.

122. 8:39

     Okay.

123. 8:40

     Let's go. Do the same.

124. 8:41

     Um, I'll do Jeff Dean.

125. 8:45

     Um, Jeff Dean is not on the board-

126. 8:48

     Oh

127. 8:48

     ... which is-

128. 8:49

     Oh. [laughs] Do we have to go again?

129. 8:51

     ... what I had registered earlier. Um, okay, so that goes to, uh, Rocco's Basilisk.

130. 8:59

     Ooh.

131. 9:01

     And we're gonna go-

132. 9:02

     Wait, what were the rest of them?

133. 9:03

     Yeah, I'll show you-

134. 9:04

     The answers?

135. 9:04

     Yeah, I'll show you the rest of them.

136. 9:07

     This one.

137. 9:08

     So these were the ones that were guessed. We also have Andrew Ng.

138. 9:12

     Oh, yeah.

139. 9:12

     We have Yann LeCun and Uber. [laughs]

140. 9:14

     Yes.

141. 9:14

     We have Fei-Fei Li.

142. 9:15

     Fei-Fei Li. [laughs]

143. 9:16

     Yes.

144. 9:16

     And we have Yoshua Bengio.

145. 9:18

     Yoshua Bengio.

146. 9:20

     I feel like-

147. 9:20

     There is a long tail of very influential researchers too-

148. 9:23

     I was closely-

149. 9:23

     ... but they, there are only eight spots, so all great guesses. Okay, next.

150. 9:29

     All right, uh, I think John and Colin.

151. 9:34

     So what is it? Okay.

152. 9:35

     Yeah. You wanna know the question?

153. 9:37

     Yeah.

154. 9:37

     Patience. So name the top considerations when choosing a model. [buzzer]

155. 9:44

     Oh, I think there's a problem. [laughs] [laughs] I think we tap first. [laughs] [thud]

156. 9:49

     More passion.

157. 9:50

     Oh, more passion. [laughs] [laughs] [thud] Hey. [laughs]

158. 9:56

     Co- co- uh, come aggressively, this team. But we'll, yes, name the top considerations when choosing a model.

159. 10:04

     Intelligence. Nice. [laughs] [laughs]

160. 10:18

     I'm gonna be very literal.

161. 10:19

     Oh.

162. 10:20

     Very smart answer.

163. 10:21

     Very smart answer.

164. 10:22

     Mm-hmm.

165. 10:24

     You have an opportunity to, to guess.

166. 10:28

     Uh, safety.

167. 10:31

     Ooh.

168. 10:31

     Safety, uh, is actually not on here. [laughs] [laughs] And so why don't we get, uh, Tina and Petra to guess?

169. 10:42

     Petra.

170. 10:43

     Am I next?

171. 10:43

     Yeah.

172. 10:44

     Oh, um, price. [chime]

173. 10:48

     So cost is the number one answer, which actually surprised me. I don't know if, like, maybe raise of hands if you would've guessed price. Now you know, so hindsight is 20/20.

174. 10:57

     That wouldn't have been obvious to me, but this means that you guys will continue with this answer, and we'll, we'll, we'll continue from there.

175. 11:05

     Latency.

176. 11:07

     Latency. Uh, yes. [chime]

177. 11:15

     After reading those scores. [laughs]

178. 11:16

     Uh, eval/benchmark scores. [laughs]

179. 11:22

     I feel like-

180. 11:23

     Some of these things are ambiguous, but I'm gonna put it under, don't, don't kill me, accuracy performance. [chime]

181. 11:30

     Okay. Yeah.

182. 11:31

     Yeah. Yeah.

183. 11:31

     Thanks. [laughs] Absolutely. Fair. Uh, uh, hmm. That's, like, everything. [laughs] [laughs] [singing]

184. 11:46

     Okay. Um, uh, the CEO.

185. 11:50

     The CEO. [laughs] [laughs] [buzzer] Good answer. I was thinking, good answer, good answer, but

186. 11:56

     moving on.

187. 11:58

     Oh, I thought w- 'cause we missed, oh, no we still go.

188. 12:00

     No.

189. 12:00

     Okay.

190. 12:00

     Your team got it.

191. 12:01

     Um, like, where the, the model's being served, like, if it's on-prem or, you know.

192. 12:06

     Uh, no, we're not gonna count that one. [buzzer] But it's a good answer. Okay, this team has an opportunity to steal these points

193. 12:15

     I hear some folks in the audience think they're ready to steal, so. [laughs]

194. 12:20

     I love that about you. I love the confidence.

195. 12:24

     Oh. Oh my God. [laughs]

196. 12:35

     Which one did you want?

197. 12:41

     The one there.

198. 12:42

     Okay.

199. 12:43

     Oh.

200. 12:43

     We can make a guess. Oh, that was my... No, it wasn't. Uh, open source versus closed source. [chime]

201. 12:49

     Woo!

202. 12:50

     Okay. So with that, you guys get 63 points, 'cause you stole the, stole the points by getting-

203. 12:56

     Let's go. Wow.

204. 12:57

     So it's a pretty close game going into the third and final question. [laughs]

205. 13:01

     Let's get the answers.

206. 13:02

     Before that.

207. 13:02

     Yeah.

208. 13:03

     Excellent.

209. 13:04

     All right.

210. 13:06

     Before the answers. [laughs]

211. 13:08

     Sorry. [laughs] I know everyone, everyone wants the answers.

212. 13:11

     Who has, who has, like, very strong upper body strength? Like-

213. 13:14

     So, so here, here are the answers.

214. 13:19

     Oh. Capabilities.

215. 13:22

     Yeah, these were the answers.

216. 13:24

     Excellent.

217. 13:25

     Great. Thanks for keeping me honest, everyone in the audience and everyone here.

218. 13:28

     Yeah.

219. 13:29

     So, all right, next. So who do we have coming up? Petra and Tina. Incredible.

220. 13:35

     You can do it. [laughs]

221. 13:37

     So both in the front. Okay. This requires, uh, power, is what we've learned. So, um, are you ready? Ready.

222. 13:45

     Yeah.

223. 13:46

     Okay. We asked 100 AI engineers, name a buzzword- [buzzer] [laughs]

224. 13:50

     Sorry. [laughs]

225. 13:52

     It's okay, I'm not offended. But name a buzzword everyone in AI is tired of hearing.

226. 13:57

     Agents?

227. 13:58

     Agent. [laughs] What do you all think?

228. 14:02

     Yeah.

229. 14:02

     Yes?

230. 14:04

     Excellent. [chime] Oh.

231. 14:07

     Okay, but you still have an opportunity to guess, 'cause you could get the number one answer. Oh, okay. Um.

232. 14:20

     Um- But you have to guess now. I know. [laughs]

233. 14:24

     I'm giving you three, two, one. [buzzer] Sorry. All right. Um, okay. So it's gonna go to this team.

234. 14:32

     John.

235. 14:33

     And, um, uh, Shresta, it goes to you. We asked 100 AI engineers, name a buzzword everyone in AI is tired of hearing.

236. 14:48

     Oh. [laughs]

237. 14:53

     I'm gonna start my five, four-

238. 14:55

     Oh, multimodality.

239. 14:58

     Multimodality? Uh, we do not have multimodality. But I don't know where this strike came from, so it's the first strike.

240. 15:08

     Uh, let me go with copilots?

241. 15:12

     Copilots. That's a good guess. What do we think? Yes or no to copilots? That's a very good guess.

242. 15:19

     No.

243. 15:19

     No.

244. 15:19

     I think it's a good... Okay. [laughs] [laughs] Good answer. Uh, but no. [buzzer] [laughs]

245. 15:28

     AGI.

246. 15:29

     AGI.

247. 15:30

     Oh. [clapping] Good answer.

248. 15:31

     Okay, what, what does the audience think? AGI?

249. 15:34

     Yes.

250. 15:35

     All right, it seems like we think it's a good answer, and no. Just kidding, yes. [chime]

251. 15:39

     Oh. [laughs]

252. 15:39

     The number one answer, AGI. All right, Tina.

253. 15:45

     Oh, um, DeepSeek? [laughs] [laughs]

254. 15:52

     And it's a team leash. I don't... [laughs] [buzzer] No, no DeepSeek-

255. 15:55

     Oh.

256. 15:56

     So this team has an opportunity to steal.

257. 16:00

     But Deep- But you're right, maybe, maybe for, like, the week. [laughs]

258. 16:06

     Yeah.

259. 16:06

     Yeah, you're back. You're so back.

260. 16:07

     All right. So I'm gonna do, come back from the last one. Uh, safety.

261. 16:12

     Safety.

262. 16:14

     Yeah.

263. 16:14

     So unfortunately, safety is not one of the answers. [buzzer]

264. 16:16

     Oh. [clapping]

265. 16:19

     RAG?

266. 16:20

     So the, these were the additional answers.

267. 16:22

     RAG, prompt engineering.

268. 16:22

     People here are sick of RAG and prompt engineering.

269. 16:26

     Yeah.

270. 16:26

     Um, but what this means is that we have a winner, which is Rocco's Basilisk. [chime]

271. 16:31

     Woo! [clapping]

272. 16:32

     And that's a great... Uh, like, amazing work by page attention and the mixture of experts. And so we're gonna move to the Fast Money round. Do you guys have two representatives?

273. 16:44

     Coming in. [laughs]

274. 16:44

     Do you want to go?

275. 16:45

     Yeah, I'll go. All right, I'll go.

276. 16:47

     All right, you two are going, and that means, John, leave the stage. I'll call you when you're ready. [laughs]

277. 16:53

     All right, we're moving into Fast Money. And, uh, so some of you have seen Family Feud, but we're playing Fast Money. The goal is for the two of them together to get to 200 points.

278. 17:05

     Uh, you have, Mihir, 20 seconds to answer five questions. Uh, if you can't think of anything, just say pass, and we can come back to it at the end or never.

279. 17:15

     Um, and if you hear a buzzer sound, let me check that it works. [buzzer]

280. 17:20

     Beautiful. Your answer wasn't one of the surveyed questions, so you can keep asking or, or move on. All right, are you ready?

281. 17:27

     Yep.

282. 17:28

     Okay. Uh, name an AI tool that engineers love.

283. 17:33

     Cursor. [laughs]

284. 17:35

     Uh, name the m- Sorry, I didn't start the clock. [laughs] [laughs] [countdown timer]

285. 17:42

     Name the job most at risk of AI disruption.

286. 17:45

     Software engineers.

287. 17:46

     Name the most influential AI paper in history.

288. 17:49

     Attention Is All You Need.

289. 17:52

     Name the biggest nightmare for AI engineer at 2:00 AM.

290. 17:55

     Uh, hardware failure. [buzzer]

291. 17:59

     Okay, so we got through four, but they were good ones. So name an AI tool that engineers love. You said Cursor. Cursor [chime] is the number one answer.

292. 18:09

     Okay.

293. 18:09

     Excellent. [clapping]

294. 18:11

     We asked, name the job most at risk of AI disruption.

295. 18:16

     You said software engineering. That is the number three answer. [chime]

296. 18:21

     We asked name the most influential AI paper in history.

297. 18:26

     What do you think? Demolished. Demolished. [chime] Attention-

298. 18:29

     Oh, yeah

299. 18:29

     ... Is All You Need is by far the number one answer. Uh, name the biggest nightmare for AI Engine to 2:00 AM.

300. 18:36

     What do y-

301. 18:37

     Hardware failure?

302. 18:39

     Hardware failure. You know what? We'll give it. We'll give it as infra problem. [chime]

303. 18:45

     Okay. Okay. Yeah.

304. 18:46

     Um, and so you're going into the round 140 points. You wanna tap, tap John in? We're ready. We're ready.

305. 18:57

     John. [laughs]

306. 18:59

     John ran away. He got scared. No. There we go. Amazing. [claps]

307. 19:05

     Incredible. So John-

308. 19:06

     Let's go. Big, big things

309. 19:07

     ... [laughs] You have 25 seconds... And they gotta stand back there. You have 25 seconds to answer five questions. Um, and if you can't think of anything, just say pass, and we'll come back to it if we have time at the end.

310. 19:19

     Uh-huh.

311. 19:19

     Uh, if you hear a buzzer sound, it means that either your answer wasn't, isn't on the board, or Mihir has already answered it.

312. 19:26

     Mm-hmm. [clears throat]

313. 19:27

     You ready?

314. 19:27

     Yep.

315. 19:28

     Um, okay, and I'm gonna start your timer. [countdown timer]

316. 19:33

     Name an AI tool engineers love.

317. 19:36

     Cursor. [buzzer] Oh, um, uh, uh, ChatGPT.

318. 19:43

     Um, name the job most at risk of disruption.

319. 19:46

     Um, artist.

320. 19:49

     Uh, name the most influential AI paper in history.

321. 19:53

     Uh, transformers is all... Attention Is All You Need. [buzzer]

322. 19:58

     Okay. [laughs] So we have model APIs, which is [chime]

323. 20:05

     uh, the third answer on the board. We have name the job most... You said, you said artist, so I'm gonna give you content creation/writing. It's a little ambiguous. [chime]

324. 20:17

     Thank you.

325. 20:17

     And Attention Is All You Need was already, uh, selected by Mihir. But we'll- I'll let you pick the next question, which is name the biggest nightmare for AI engineer at 2:00 AM.

326. 20:27

     Someone wrote in cold email from a VC, so I will talk to you after. [laughs] [laughs]

327. 20:32

     Uh, uh, CUDA errors. [laughs]

328. 20:34

     CUDA errors. Okay. Well, you know what? You're teammates because you think alike- [laughs] ... and you think, you, you think of great answers. So let me just show you the number one answers here, [chime] which are, uh, cursor, [chime] data entry for the job most at risk of AI disruption, [chime] Attention Is All You Need, the, the top paper, so that

329. 20:50

     was a good one, just already taken. The biggest nightmare, [chime] an outage of the model or otherwise, and an industry that would benefit the most from AI, [chime] healthcare.

330. 21:01

     Uh, so with that, we have... Well, you lost the 200 points. [laughs] [fail sound] [applause]

331. 21:07

     But I think that you're still winners, and you're also winners, and we have a few fr- prizes outside that I think someone is bringing. Mm-hmm. So- [laughs] [laughs]

332. 21:19

     We have a massive llama for this team. We have rainbow llama Beanie Babies for everyone.

333. 21:24

     Oh. [laughs]

334. 21:24

     We have AI engineering books and gift cards to your favorite restaurants in New York, and maybe we can bring 'em out and leave. Uh, thank you so much for joining us for Frontier Feud.

335. 21:35

     Thank you.

336. 21:35

     And we're gonna see you next year. We're running a massive survey [laughs]

337. 21:40

     on the state of AI engineering. Ooh. Gonna be presenting it at the June AI Engineer World FAIR. So if you liked some of the teaser questions, this will get much more in depth into the tools folks are using, uh, the workflows that AI engineers have, and it's a way for the industry to be more transparent.

338. 21:56

     Uh, so thank you. Uh, you can find the QR code or the link here if you wanna participate in that survey on the state of AI engineering. [outro music] Amazing. [applause] [outro music]