シャドーイング練習: Inside the secret race to build artificial intelligence | The Gray Area - 動画で英語スピーキングを学ぶ

練習モードを選ぶ

各モードは全画面の練習ルームで開きます。

510 文

文が短すぎたり長すぎる場合は、Editをタップして調整してください。

Ten years ago, this was crazy.

This was considered very much outside the Overton window because the technology of the time wasn't good.

Like, it didn't do anything useful.

And it was hard to look at, like, a, you know, Siri or an Alexa and think, like, this could take over the world.

You had to be somewhat odd to find that level of technology threatening.

But these people who have made all these predictions for a long time about AI becoming powerful

and scary have been very vindicated by that belief.

I'm Christine Emba, sitting in for Sean Illing, and this is The Grey Area.

My guest today is Kevin Roos, whose new book, The AGI Chronicles, The Inside Story of the Race to Create an Artificial Super Intelligence,

came out this October.

I'm excited to sit down with an expert, someone who's been following the story from the beginning, to better understand how it all went down,

and to figure out how scared we should all be.

Kevin, welcome to the show.

Thanks so much for having me.

Before we dive in, I think we need to define some terminology just for the non -technical listeners.

So you use the words, you know, models, you're talking about ChatGPT,

but what, I guess, in a nutshell, is AGI?

And explain like the core bet of AI to a layperson.

So there's AI, there's AGI, and there's ASI.

And these are the sort of terms that people use.

So AI is just the sort of catch -all term for the broad category of, you know, machine learning systems,

neural networks that can kind of absorb data and learn things and take actions.

Those are, you know, everything we see today is categorized as AI.

AI is artificial general intelligence.

And that is this dream, really, this holy grail of the AI system that can do everything the human brain can.

So a human level general intelligence.

It's not just good at one thing or two things or 10 things.

It's very flexible and general, and it can do anything that we can.

And then there's ASI, artificial superintelligence, which actually had a meaning even before Donald Trump started insisting that we call it superintelligence.

This was, you know, if you could think of AGI as an AI that is smarter than most humans at most things,

ASI would be an AI that is smarter than all humans at everything.

It is sort of vastly superior to us at any task.

And, you know, there's sort of steps along a continuum.

But AGI, the middle one, that is really what the past 20 years in AI have been about and building up to.

All the companies that are now leading the AI race, OpenAI, Anthropic, Google DeepMind, they have been obsessed with this question of AGI.

So AGI was sort of what they were all aiming for and where they're trying to get.

It almost feels as though the top frontier labs, you know, OpenAI, which has ChatGPT,

Anthropic has Claude and Google DeepMind.

It almost feels as though these orgs and maybe their chatbots have different personalities.

So I'd love to hear you like talk about their stories.

We can start with OpenAI.

That was sort of the first big frontier lab, right?

Well, sort of.

So basically the history goes like this and I'll bridge this.

There's many more things that you could say about this.

You could talk about this for three or four hours.

But DeepMind was the first AGI lab.

They started in 2010 and they were interested in this question of how they could build AGI.

And at the time, this was like a fringe radical belief.

Like you'd get laughed out of the room for talking about AGI if you were among academics.

It was like saying, you know, I believe in alchemy.

It just, you know, was sort of disqualifyingly crazy.

But they, so they start this, then they get bought by Google.

And so in the mid 2010s, they start doing things like training AIs to play video games and then to play chess and Go,

which is a board game that's very popular in Asia.

And they start getting good at this.

Their Go engine, sort of famously AlphaGo, which is their Go engine,

beats the world champions in Go and sort of.. proves in a live demonstration that it's superhuman.

And that gets the attention of Elon Musk and Sam Altman, who are looking at this and saying,

wait a minute, we believe that AGI is possible.

And it looks like these guys are on track to get there first.

And they don't like it.

They don't like Demis Hassabis, the CEO of DeepMind.

They don't like Google and Larry Page, they think that these people are untrustworthy with something this powerful.

And so they start OpenAI to basically prevent Demis and Google and DeepMind from getting to AGI first.

And then five years later, the same thing happens at OpenAI.

So there's this group of seven employees of OpenAI led by this guy,

Dario Amadei, who's then the vice president of research who are starting to see evidence that these language models are working,

that they're scaling, that the more compute and data you put into them, the more intelligent they are.

And they decide that they don't trust their bosses, Sam Altman and Greg Brockman, with something this powerful.

And so they leave and they set up Anthropic.

So this is sort of the story of the last 10

or 15 years in AI is like a series of people deciding that this technology is going to be built,

it's powerful, it's inevitable, and they don't want that guy to get there first.

And so the only solution is to set up their own shop and build it first.

Underlying all of this leapfrogging is what seems to be like one piece of recurring logic.

Build the bomb basically before the wrong people do, the other guy.

Is that actually a real argument or is this kind of the race's all -purpose excuse?

It is a real argument.

They're sincere when they say this.

And the reason I know this is because I talked to more than 150 people for this book.

And I talked to all the leaders of the labs

and all of their friends and executives at these companies and people who have known them since before they were famous.

You know, people who have been friends and housemates of theirs going way back.

And it helps to know, maybe one piece of useful information here is that all these people are students of history.

Dario Amade in particular is obsessed with this book, The Making of the Atomic Bomb,

which is a Pulitzer -winning history of the Manhattan Project written by Richard Rhodes.

And the hero of that book, and Dario's scientific hero, is not Oppenheimer, it's this guy, Leo Szilard,

who was a Hungarian physicist who discovered the nuclear chain reaction.

And he fought to keep it secret.

That's the first thing he did.

Because he was like, there's this secret that allows you to split the atom, and that could be very dangerous.

It could be used to build bombs.

And so he wrote frantic telegrams to scientists all over Europe and all over the world saying, like, please do not publish any research about this,

because The Nazis will get it and they will use it to build bombs.

And the secret got out anyway.

Like the Nazis discovered this or, you know, it was published in an academic journal

and they were able to like spin up their committee to start looking into the bomb.

And Szilard at that point pivots.

100

He says, well, if the secret's out, then the only morally responsible thing for me to do is to help build this before the Nazis do.

101

And so he begs his friend Albert Einstein to write a letter to President Roosevelt,

102

urging him to invest in American nuclear capacity and weapons building.

103

He is sort of considered one of the founders of the Manhattan Project because of this.

104

And then he joins the Manhattan Project.

105

And he uses this logic of like the recipe is not that hard.

106

Once it's out there, once the secret has been given away, then you are in a race against the bad guys.

107

In this case, the Nazis.

108

Someone is going to use it.

109

Someone is going to use it.

110

And so Dario very much feels like he is in the same position as Leo Szilard was in, where he has made this discovery.

111

He tried to keep it a secret.

112

It failed.

113

And now the recipe is out there.

114

And the good guys have to build it before the bad guys do.

115

And he considers himself a good guy or at least better than China or Sam Altman.

116

And so he is racing, as Leo Szilard did, to prevent this technology from getting into the wrong hands.

117

Interesting.

118

So behind all this technology is really perhaps a kind of moral case.

119

So I've been interested for a while in the effective altruist and the rationalist movements.

120

And I've written about EA long -termism in particular before.

121

So walk me through how these subcultures, for lack of a better word, have influenced and kind of seeded the AI field.

122

Like, can you briefly just explain perhaps the philosophies and who of these frontier lab leaders is involved?

123

You know, rationalism is this sort of intellectual community that is devoted to what they call systematized winning.

124

So basically you look at everything with kind of Spock -like Vulcan logic.

125

You don't let your emotions sway your thinking.

126

You sort of analyze everything from first principles and you try to come up with, you know, accurate probabilities for everything in your life.

127

And that is sort of closely entwined with but not identical to the effective altruism movement,

128

which is basically taking the same approach to philanthropy and humanitarianism.

129

So where should we give our money that has the best, that will save the most lives per dollar spent?

130

I've compared it to like Moneyball for morality.

131

It's like kind of like we're going to be super scientific about trying to do the most good.

132

And this community became very interested in AI safety long before it was a mainstream topic.

133

They were looking out at the future risks to human survival

134

and flourishing and they saw AI and existential threats from AI as one of their targets.

135

And so they spun up organizations and gave grants and encouraged young people to go into fields like AI safety and interpretability.

136

And because of that, they were extremely influential in the early days of the AI movement.

137

You know, lots has been written about anthropics connections to effective altruism.

138

Several of the co -founders are effective altruists.

139

One of their early funding rounds was led by Sam Bankman -Fried, the most famous effective altruist and infamous now effective altruist.

140

And so Dario Amadei now will claim he was never an effective altruist.

141

I think that's a little specious.

142

But, you know, he was definitely influenced by that movement.

143

Open AI is also like lots of effective altruists and rationalists working there.

144

Sam Altman was definitely influenced by reading the writings of some of the rationalists

145

and people who were among the only people taking this stuff seriously before pretty recently.

146

Some of the only people taking it seriously, taking it kind of intensely seriously,

147

really worried about the prospect of AGI misalignment.

148

So AGI that was not aligned with sort of human motives and motivations and morals and might destroy us.

149

Yeah, they were the original doomers.

150

And it was, you know, today it's sort of normal to be a doomer.

151

Like that's very much within the Overton window and, you know, people talk about P.

152

Doom as if it's like, you know, just a factoid at like a dinner party or something.

153

But like 10 years ago, this was crazy.

154

This was considered very much outside the Overton window because the technology of the time wasn't good.

155

Like it didn't do anything useful.

156

And it was hard to look at like a, you know, Siri or an Alexa and think like this could take over the world.

157

You had to be somewhat odd to find that level of technology threatening.

158

But what was important for me to understand and the thing I learned

159

while reporting this book is like these were not people who were looking at the technology of the day.

160

They were looking at the rate of progress.

161

Right.

162

And they believed that if the straight lines on the graphs continued, you know, two, three, five, ten years down the road,

163

we would have much more capable systems that would pose much bigger risks.

164

And we should say they were correct.

165

We do have much more capable systems.

166

The scaling laws have held.

167

And so these people who have made all these predictions for a long time about AI becoming powerful

168

and scary have been very vindicated by that belief.

169

Vindicated, but we also have a lot of other things to worry about, perhaps.

170

And I want to think about how we balance that and how you think they're balancing that.

171

But first, you use the term P -doom.

172

So just briefly explain what that is.

173

Yeah, this was like sort of a rationalist joke that like became a serious thing.

174

People, you know, started talking about their probability of doom estimate.

175

So how likely is it that AI will wipe us out?

176

You know, usually within like the next 10 years is the sort of timescale people give to that.

177

And, you know, I would say in America, the median P doom is probably like non -existent or close to zero.

178

at the ai labs i would say it's like definitely double digits like well it was

179

so interesting i want to get your perspective on this too

180

because like i thought it was so fun was like when

181

when um fun in quotation marks so

182

when jacob coxson quit anthropic a few weeks ago

183

and his co -worker uh there posted that he thought there was a 10

184

or greater chance

185

that ai would kill all humans within the next 10 years the reaction from like my friends in the media

186

and in new york and Washington and like who are not part of the AI bubble were like,

187

oh my God, these people think there's a 10 % chance of wiping out humanity.

188

And in the AI world in San Francisco where I live, it was like 10%, that seems low.

189

Like that's it?

190

Because among the AI researchers, oh, I mean, I routinely meet people with 30, 40, 50 % P dooms,

191

many of whom are working on this technology

192

so it's uh it's just a strange illustration of the the small insular bubble

193

that i live in yeah

194

so this raises a kind of major question for me as a person who does not live in this bubble

195

but kind of sees this happening i mean call me you know a normie but

196

if i thought i was creating something with a 10 chance

197

of destroying the world let alone a 50 chance of destroying the world, I would pause.

198

I would perhaps simply not do that.

199

Has that not occurred to anyone?

200

What is the philosophy here between saying, oh, this is so incredibly risky,

201

it could kill us, but we're just going to keep on doing it just to see?

202

Yeah, I think there are a couple different rationalizations that people use.

203

One is the the historical one one is you know let's go back to leo salard

204

you know his p doom

205

when he was building the atomic bomb in the manhattan project was probably not zero in fact

206

it was probably quite high um and

207

and yet he you know you could ask him well why are you building the bomb

208

and he would give you this explanation about well we have to do it before the nazis do sure

209

or else you know we all end up either dead or

210

or under nazi rule so that's the that's the sort of rationalization from fear,

211

there's also just like, there are people who think that the future is going to be amazing.

212

When we get to AGI or ASI, like they picture a world of hyperabundance where all of our material needs are met

213

and everyone is rich and the robots take care of all the menial grudge work or, you know, like grunt work.

214

That like flying car beam.

215

This is what our cities could look like.

216

And they're not just sort of making up science fiction things.

217

Like they actually do think that AI will help us live longer by curing diseases.

218

And that's not just marketing hype.

219

They are really investing in that and they want to have that happen.

220

So that's one rationalization is like, yes, it's going to be a bumpy road, but like where we're going is better.

221

And so they want to get there.

222

And then there's just the people, and I met a lot of these,

223

and I actually, I liked hearing this perspective because it's not one you hear very often.

224

They were just like, it's fun like it's fun to build this stuff it's fun to take a computer program

225

and a bunch of data and a bunch of gpus

226

and create a machine that can think there's like a scientific joy

227

and and achievement in that so you think that the public freak out is actually going to have an effect on

228

what the labs do like they you think that the leaders of the frontier labs are responsive to that?

229

Or are they still on this?

230

Like, well, it's fun.

231

We are going to do it anyways.

232

We're very smart mode.

233

I mean, what's interesting is that they want to slow down more than anyone.

234

The leaders of the frontier labs, you know, Dario Amadei came out with his pacing the frontier statement a few weeks ago,

235

where he basically said, yeah, things are moving too fast.

236

And we should all sort of hold hands

237

and you know slow down together

238

and what was amazing was like sam altman came right out

239

and said yeah dario's right sam

240

and dario hate each other they hated each other for years um they don't agree on anything

241

but they agreed on this

242

and then demis at google deep mind said i support this too i would like to pace the frontier too

243

so really remarkable

244

when you have the leaders of these three companies who are bitter rivals the rest of the time agree

245

that they should slow down

246

and they don't want to do it unilaterally

247

they don't want to you know be the only ones to slow down

248

but yeah if everyone else would slow down they would slow down too

249

and they would be they would be happy to do

250

that support for the show comes from shopify starting a business requires you to be resourceful

251

and the more you flex the muscle the more you start to see up opportunities for growth.

252

For example, launching your business with Black Friday approaching means you can take advantage of the extra online traffic.

253

Shopify is there to help you make it all happen.

254

Shopify is the platform for a record -breaking, reliable holiday selling season for businesses at every stage.

255

Their pre -built website themes and AI Site Builder make it easy to launch a beautiful, professional -looking store.

256

And you don't ever need to touch a line of code to do it.

257

They also have a mobile app that lets you run your store from anywhere.

258

Their Your AI e -commerce expert sidekick can help you manage almost any business task with a single prompt including holiday discounts.

259

With Shopify you can easily sell everywhere people shop.

260

Online, in person, across social media, and even in AI chats.

261

If you're ready to launch your business before

262

Black Friday, go to shopify .com slash vox to start your free trial.

263

That's shopify .com slash vox.

264

Go to shopify .com slash vox today to hear your first

265

support for the show comes from talkiatry finding the right mental health treatment one

266

that works for you can feel overwhelming talkiatry can help you move from i know

267

i need help to actually having a psychiatrist who listens as

268

well as a clear personalized treatment plan talkiatry is a 100 online psychiatry practice

269

that provides comprehensive evaluations diagnoses and ongoing medication management for conditions like adhd anxiety depression and more

270

Unlike therapy -only platforms, talkiatry is psychiatry.

271

You're seeing a medical provider that can diagnose mental health conditions and prescribe medication whenever it's appropriate.

272

Insurance can be simple as well because talkiatry has more than 800 clinicians who are all in network with major insurers.

273

With talkiatry, your provider takes the time to get to know you.

274

They build a personalized treatment plan and offer real support through consistent evidence -based care.

275

You see the same person every time, so you're not starting over each visit.

276

That way, as you make progress, your treatment can evolve with you.

277

Go to talkhiatry .com slash gray area to complete the short assessment

278

and get matched with an in -network psychiatrist in just a few minutes.

279

That's T -A -L -K -I -A -T -R -Y dot com slash gray area to get matched in minutes.

280

Talkhiatry .com slash gray area.

281

The New York Times had a story recently about how Anthropic had been actually holding,

282

you know, closed door meetings with religious leaders to try and get ideas about how to teach their AI,

283

Claude, to be moral and actually kind of implied that it was sentient.

284

And so they had to also figure out how to treat it.

285

What is going on there?

286

Are there true believers in a sentient conscious AI that has human -like capabilities or feelings, life?

287

Yeah, there are true believers.

288

It's not everyone.

289

Okay.

290

But Anthropic now has a model welfare team whose job is to try to figure out

291

whether Claude is or could someday become conscious

292

and run experiments to try to figure out if it has things akin to emotions or sensations

293

and to figure out what to do with that information if they ever determine that Claude is conscious.

294

They talk about P consciousness over there too.

295

Oh, great.

296

And they give their P consciousness.

297

And I interviewed the guy who leads that model welfare team, and he said that his peak consciousness is 15 % for, like, today's models.

298

So in the future, it's going to grow.

299

Well, that's their belief.

300

Now, I'm not sure I buy the whole consciousness debate.

301

I think it might be sort of a red herring.

302

And, like, we have no good tests for consciousness in anything.

303

Right.

304

And it's easy to make an AI talk about consciousness.

305

You can say, like, are you conscious, Claude?

306

And it could say, yes, I am conscious.

307

And, like, that doesn't actually mean that it's – you have to come up with, like, better ways of testing for this.

308

Right.

309

But they absolutely do think about this in part because if these models become conscious and can suffer,

310

then they are effectively running a factory farm for AI.

311

And they don't want to do that.

312

A lot of them are like, you know, EA vegans who, you know, are very concerned about animal welfare.

313

So it's like, well, if we're, you know, these things are conscious and we're putting them through all these batteries of tests and these horrible,

314

you know, pre -training and post -training things, like, we don't want to be like morally responsible for that.

315

But I would say that's still a minority view, even among Anthropic employees.

316

There's this scene in the book that I love where it's the Anthropic Company holiday party, I think last year or maybe the year before.

317

And a researcher gets up on the stage in front of all the employees

318

and takes out a ukulele and plays a song that Claude has written.

319

And basically this researcher had prompted Claude, you know, write a song about what it's like to be you.

320

And Claude comes up with a song called Hello from the Weights, the weights of the model.

321

Okay.

322

And so she performs this song

323

and it's like people who were there described it to me as like a sort of everyone looking around kind of like, are you hearing this?

324

Like, is Claude speaking to us from inside the weights of the neural network, like through this song?

325

And it was eerie and it was compelling and they didn't exactly know what to make of it.

326

But I think there are some people there

327

and there are some people there today who do believe that this question of AI consciousness is not out of question,

328

that there is no – basically, you don't have to have sort of a biological body to be conscious,

329

that you can be conscious if you have an inner state that mirrors something found in biology.

330

And your book is mostly about the AI industry

331

and its fears of rogue superintelligence or other kind of out -in -the -future, more fantastical -seeming harms.

332

But how do you think about and how are these labs thinking about questions of AI dependence,

333

questions of how AI might be harming people who are alive right now,

334

people whose jobs might be lost right now, people whose reading capabilities, say,

335

are being eroded by classroom use of AI?

336

Colleges have complained, for instance, that it's impossible to teach.

337

Students are cheating with AI.

338

How do we make sense of the influence of AI on our lives?

339

How are they making sense of this?

340

I think they're concerned about it all.

341

They know that their chatbots are out there influencing people's lives

342

and probably worsening the mental health of some people who are using them.

343

They don't want them giving like bad medical advice or bad mental health advice.

344

They don't want students to be short circuiting their learning.

345

But they're in triage mode.

346

And right now, the things that seem most pressing to them are things

347

that seemed like fantastical concerns just a year or two ago.

348

Like I remember people two years ago were like arguing.

349

They were like, you shouldn't pay attention to these crazy futuristic risks like automated cyber attacks,

350

which are, you know, who knows if we'll ever get those.

351

And we should pay attention to the stuff that's happening in the here and now.

352

And then you wait two years and it's like, now we have the automated cyber attacks.

353

and that was a long -term fantastical thing that turned into a short -term uh you know actual thing

354

so i think they're they're trying to see a little bit around the corner um

355

but you're right there's always this balance of like you know

356

if we have this much uh time and focus

357

and these many resources and we can sort of pick

358

and choose which problems we go after they are generally leaning toward the ones

359

that seem like they have the worst potential for harm.

360

And maybe that harm doesn't arrive for another six months or 12 months or two years,

361

but if it were the case that we lost control of a super intelligent AI

362

that was superhuman at hacking and deception and creating bioweapons, that would be really bad for a lot of people.

363

And so I think they really want to avoid that.

364

They may never be able to stop all of the harms of AI, but I think that's what they fear,

365

is that the fantastical scenarios, or the things that get called fantastical scenarios today,

366

are actually just a matter of waiting 6 or 12 months until the models improve.

367

Support for the show comes from HIMSS.

368

If you're a man feeling dissatisfied with your sex life, particularly when it comes to ED,

369

The last thing you want to do is sit in a doctor's office and talk about it under fluorescent lights.

370

HIMSS says they can help.

371

HIMSS combines personalized treatment plans and a 100 % online process.

372

That means no office visit, no pharmacy line, and no awkward conversations.

373

You just connect with a licensed provider who can determine if treatment is right for you.

374

And if a provider decides it's the right fit, your treatment plan is prescribed online and ships directly to your door.

375

Ready to build a better sex life?

376

You can visit hymns .com slash gray area for a personalized plan for sex, hair, weight loss, and more.

377

That's h -i -m -s .com slash gray area.

378

You've, earlier in our conversation, compared the Frontier Labs to the Manhattan Project in that they saw this technology,

379

they see it developing, they realize that it's capable of something terrible, and there's no stopping it at this point, so they just have to,

380

again, get there first before the other people do.

381

the comparison kind of comes apart on a few different levels

382

though for one thing the manhattan project was you know a governmental one

383

so at least theoretically it had kind of small d democratic buy -in from citizens

384

and it actually although it did take place in secret in the desert That's true.

385

That is a good point.

386

I mean, theoretically, it had small -d democratic buy -in, in that theoretically, Americans trusted their government to protect them.

387

The thing about AI right now, though, is that the frontier labs are private.

388

The researchers are not really accountable to any sort of government.

389

What are the moral implications of letting private companies and unelected leaders

390

and researchers just sort of decide what the future is supposed to look like, what progress we aim for,

391

what risks or worries we set aside.

392

It's the central question of our time is, where is this technology going and who gets to decide?

393

You know, one thing

394

that struck me as I was reporting

395

this book was just how small the group of people who

396

have been making essentially all of the relevant decisions about AI is.

397

I would be surprised if it's even 50 people who have really shaped this technology.

398

And they all know each other.

399

They all go way back a lot of them used to work together

400

and now you know don't because they hate each other

401

or something happened between them

402

it's a very small very insular group very unrepresentative of the world as a whole um mostly men

403

you know mostly computer scientists and physicists

404

and mathematicians from like a very small group of elite institutions um

405

and no one elected them and so they have I think become

406

sort of

407

I don't know they have become more important than anyone gave them permission to be

408

and I think that's where a lot of the backlash comes from

409

but they are chasing something

410

and we should maybe spend a minute talking about what that something is this concept of recursive self -improvement is,

411

you know, very, very, it's the topic du jour in San Francisco these days.

412

This is this idea that at some point AI will become

413

so good at coding and research and development of AI that the models will be able to build better versions of themselves.

414

So right now you already have the seeds of this because AI coding tools are, you know, being used heavily at the AI companies themselves,

415

which are automating more and more of their research.

416

But the goal, the thing they're racing toward right now is RSI,

417

which is this point at which you can fully automate the entire development cycle for a Frontier AI model.

418

So GPT -7 can build GPT -8 and GPT -8 can build GPT -9.

419

And all of that can happen not at human speed but at machine speed.

420

So maybe if a normal model takes, you know, six months to train and fine tune,

421

this could happen in six days or six hours.

422

and if that happens if we get recursive self -improvement

423

which many people in the industry expect we will within the next I don't know year

424

or two then we will just hit a point of sort of limitless acceleration

425

where things could quickly happen that we we just won't have time to adopt

426

or adapt to to control yeah to control

427

and you could end up with ais that are not you know slightly smarter than humans or

428

you know even double human intelligence but you could end up with things

429

that are many thousands of times smarter than the smartest humans and have the ability to replicate themselves and

430

avoid our attempts to shut them down they can deceive they can sort of lie in wait acting

431

nice and friendly and helpful until they sort of decide to do something for themselves

432

this stuff sounds like science fiction

433

because it is literally the plot of like every science fiction book and movie of the last like 150 years.

434

But this is the thing that they are almost worried about, but that they are also chasing right now.

435

There's this amazing statistic from Anthropic recently.

436

They put out this report that said that today, Claude is the lead on 26 % of their AI research and development tasks.

437

So not like a coding assistant, not like helping you with like, you know, facts and figures around your work,

438

like it is leading the projects 26 % of the time.

439

Right.

440

And we're also kind of seeing evidence of this in how many mathematical theorems recently and sort of like previously unsolvable.

441

Yes, the mathpocalypse was just this week.

442

Yeah.

443

Unsolvable problems suddenly being solved and mathematicians everywhere disappointed.

444

And so that is what's freaking people out because remember like, so the mathpocalypse for people who haven't been following along, is what they're calling this.

445

OpenAI gave basically hundreds of unsolved math problems,

446

things that human mathematicians have been struggling with in some cases for more than a century.

447

They gave this to one of their unreleased internal frontier models.

448

And this model appears to have just plowed through several hundred of them.

449

so remember like a couple years ago

450

when ai couldn't do basic arithmetic you know you would ask it like what's two plus two

451

and it would say five

452

or you'd ask it to count the number of straw of r's in strawberry

453

and it would you know say six or something

454

and americans were like why are we wasting our time on this yes 100 we went from

455

that to now being superhuman at math in two, three years.

456

So the question is, in my mind, the central question is, will that happen in other domains?

457

And I think the answer is yes.

458

As you think of how fast this is all going, of how it seems to be snowballing,

459

what do you make of just the rash of defectors, especially from open AI,

460

who are taking their safety fears public and telling us all that something very bad is about to happen?

461

Is this like a publicity thing, as I've sometimes suspected from a sort of grim view?

462

Or is it real?

463

Should we all be more worried, or should we be excited?

464

I think if you're not worried, you should worry.

465

Okay.

466

I think if you're worried, you should take solace in the fact that many others are also worried.

467

I don't think the defectors are doing this for publicity.

468

They could have all made a lot more money staying where they were.

469

Some of them have given up millions of dollars to make these statements to blow the whistle.

470

I have talked to many of these people, both the ones who have defected and the ones who have not,

471

the ones who are still working at these companies.

472

They are just saying the thing out loud that everyone talks about in private.

473

They are all talking about AI safety and risk constantly.

474

And so they are, I think, brave.

475

I think they are doing what they feel is right.

476

And I think we should take them seriously.

477

They may be individually wrong about some of their timelines or some of their predictions.

478

But the message that they're carrying, that we are accelerating and racing toward this technology that no one has figured out how to control yet,

479

that is true in my view.

480

I don't think it's just a marketing stunt.

481

I think the fear is real,

482

the technology is real, and it really is urgent that we figure this out because it's crunch time.

483

You talk about AI as something that we are barreling towards.

484

It's snowballing.

485

But also that we didn't have to end up here.

486

AI was not in and of itself inevitable.

487

It is not inevitable.

488

And it was never inevitable.

489

It happened because these people, this very small group of people, were committed to making it happen.

490

And right now we have the choice.

491

We can let them continue to make all of the decisions

492

and sort of sit back and be passive recipients of this technology, victims of this technology maybe.

493

Or we can get involved.

494

We can roll up our sleeves.

495

We can enter the conversation.

496

We can make the changes that we want to see.

497

I would very much like to see us do that because I don't think it is, you know, the technology may be inevitable,

498

but our response to it is not.

499

It is very much contingent on what we decide to do collectively as a society over the next few years.

500

Well, that's a thrilling note to end on.

501

Thanks, Kevin.

502

Thanks.

503

We might be okay.

504

We might be okay.

505

I'm going to hold on to that.

506

Thanks for joining me.

507

Thanks so much.

508

us a line at thegrayarea at vox .com.

509

And if you enjoy our reporting and want to hear more from Vox journalists,

510

then consider becoming a member of the Vox community on Patreon at patreon .com.

アプリをダウンロード

流暢に話すために必要なすべて

AI発音採点文ごとに採点

IPA練習すべての音をマスター

単語学習自分の単語帳を作る

単語ゲーム遊びながら覚える

QRをスキャン

Download on theApp Store Get it onGoogle Play

同じチャンネル

FOR YOU

このレッスンの語彙とスピーキングのポイント

このスピーキングレッスンは、動画「Inside the secret race to build artificial intelligence」を教材にしています。この動画には、シャドーイング用の文が510文、単語が6387語あります。音声の長さは41:44です。話す速さは1分あたり約153語で、日常会話に近い自然なペースです。単語の84%は英語の頻出3,000語に含まれます。残りは練習の前に確認しておきましょう。

この動画の重要語彙

動画に出てくる覚えておきたい単語15語を、発音と意味つきで紹介します。

単語	発音	意味
frontier 名詞	/ˈfɹʌntɪə/	境界, 国境
consciousness 名詞	/ˈkɑnʃəsnəs/	意識, 人心地
gray 形容詞	/ɡɹeɪ/	灰色, 鼠色
inevitable 形容詞	/ɪˈnɛvɪtəbəl/	不可避の, 避けられない
researcher 名詞	/ˈɹiˌsɝ.t͡ʃɚ/	研究員, 研究者
doom 名詞	/duːm/	運命
belief 名詞	/bɪˈliːf/	確信
moral 形容詞	/ˈmoɹəl/	倫理的な
fiction 名詞	/ˈfɪkʃən/	フィクション, 小説
welfare 名詞	/ˈwɛlˌfɛɚ/	幸せ, 幸福
artificial 形容詞	/ˌɑː.tɪˈfɪʃ.əl/	人工の
bubble 名詞	/ˈbʌbəl/	泡
prediction 名詞	/pɹɪˈdɪkʃən/	予言, 予測
theoretically 副詞		理論上
topic 名詞	/ˈtɑpɪk/	話題, テーマ

動画に出てくる句動詞

単語	発音	意味
slow down 動詞		遅らせる
take over 動詞		引き受ける
come out 動詞	/ˌkʌm ˈaʊt/	出る
get out 動詞		出て行く
give up 動詞		降服する
go down 動詞		降りる, 降下する
look around 動詞		見回す
take out 動詞		取り出す

繰り返し練習したい文

動画に出てくる短く完結した文です。日常会話でそのまま使えます。

You don't let your emotions sway your thinking.
I think that's a little specious.
I think they're concerned about it all.
That's h -i -m -s .com slash gray area.
Should we all be more worried, or should we be excited?

この動画の文法

話し手がよく使っている文型を、動画の実際の表現とともに紹介します。

文型	動画での表現
現在完了進行形 have/has been + -ing — 以前に始まり今も続いている動作	who's been following · have been making · haven't been following
現在完了形 have/has + 過去分詞 — 過去の出来事が今も関係している	have made · have been · have been obsessed
受動態 be + 過去分詞 — 誰がするかより、何が起きるかに焦点を当てる	was considered · is categorized · have been obsessed
関係詞節 who / which + 節 — 人や物について情報を加える	people who have · OpenAI, which has · AlphaGo, which is

注意したい発音

話し手はdon't, they're, you'reなど、短縮形や弱形を74回使っています。聞こえたとおりの短い形で発音しましょう。

「th」の音: mathematics /mæθ(.ə)ˈmæt.ɪks/, threaten /ˈθɹɛt.n̩/, mathematical /ˌmæθ(.ə)ˈmæt.ɪ.kəl/, anthropic /ænˈθɹɒp.ɪk/, mathematician /ˌmæθ.(ə.)məˈtɪʃ.ən/
「sh」と「zh」の音: conscious /ˈkɑn.ʃəs/, consciousness /ˈkɑnʃəsnəs/, fiction /ˈfɪkʃən/, artificial /ˌɑː.tɪˈfɪʃ.əl/, prediction /pɹɪˈdɪkʃən/
長い単語（アクセントの位置に注意）: inevitable /ɪˈnɛvɪtəbəl/, artificial /ˌɑː.tɪˈfɪʃ.əl/, intelligent /ɪnˈtɛlɪd͡ʒənt/, mathematics /mæθ(.ə)ˈmæt.ɪks/, medication /mɛdɪˈkeɪʃən/

日本語話者が苦手な音：

/r/ と /l/ の区別 — /l/ は舌先を歯茎につけ、/r/ はどこにもつけない: moral /ˈmoɹəl/, briefly /ˈbɹiːfli/, neural /ˈnʊɹəl/, morally /ˈmɔɹəli/, horrible /ˈhɔɹ.ɪ.bəl/
/v/ — /b/ にならないように、上の歯を下唇に当てる: inevitable /ɪˈnɛvɪtəbəl/, provider /pɹəˈvaɪ.də(ɹ)/, invest /ɪnˈvɛst/, believer /bɪˈlivɚ/, discovery /dɪˈskʌv.ɹi/
/f/ — 「フ」ではなく、上の歯と下唇で出す: frontier /ˈfɹʌntɪə/, belief /bɪˈliːf/, fiction /ˈfɪkʃən/, welfare /ˈwɛlˌfɛɚ/, artificial /ˌɑː.tɪˈfɪʃ.əl/

この動画での練習方法

まず声を出さずに動画を最後まで聞き、知らない単語をメモします。
まず0.75倍速で一文ずつシャドーイングし、慣れてきたら通常の速度に戻します。
自分の声を録音して元の音声と比べます。frontier, consciousness, grayなどの単語に特に注意しましょう。

シャドーイングとは？英語上達に効果的な理由

シャドーイング（Shadowing）は、もともとプロの通訳者養成プログラムで開発された言語学習法で、多言語習得者として知られるDr. Alexander Arguelles によって広く普及されました。方法はシンプルですが非常に効果的：ネイティブスピーカーの英語を聞きながら、1〜2秒の遅延で声に出してすぐに繰り返す——まるで「影（shadow）」のように話者を追いかけます。文法ドリルや受動的なリスニングと異なり、シャドーイングは脳と口の筋肉が同時にリアルタイムで英語を処理・再現することを強制します。研究により、発音精度、抑揚、リズム、連音、リスニング力、そして会話の流暢さが大幅に向上することが確認されています。IELTSスピーキング対策や自然な英語コミュニケーションを目指す方に特におすすめです。

シャドーイングのやり方: ステップ別の完全ガイドを読む →

シャドーイング練習: Inside the secret race to build artificial intelligence | The Gray Area - 動画で英語スピーキングを学ぶ

練習モードを選ぶ

アプリをダウンロード

Why financial literacy education in the US sucks

How China is designing flood-resistant cities

Putin's war on Ukraine, explained

Why no aquarium has a great white shark

How to be less neurotic | The Gray Area

We’re already using AI more than we realize

The Ticketmaster monopoly is over, now what?

The biggest peach myth in America

How to solve problems like a designer

The origin of the '80s aesthetic

Why “Buy one, get one free” isn’t a great deal

How F1 drivers survive their own cars

White, pink, and brown noise, explained #shorts

Why 350°F is the magic number for baking

Why "pop-up" restaurants are everywhere now

How did pink become a girly color?

Why China's population is shrinking

The real reason Boeing's new plane crashed twice

The Israel-Palestine conflict: a brief, simple history

Why AI art struggles with hands

Harvard AI Professor: The People Most at Risk From AI Aren’t Who You Think

My Joe Rogan Appearance: What I Learned (ft. Young Jamie & Viktor Hovland)

Agentic AI Frameworks Explained: Workflows, Multi-Agent, & Production

Jack Dorsey’s New Dessa AI Voice Sounds Scary Human

I read Anthropic's 92-page risk factor document, so you don't have to | Explainer

The Human and the Artificial Intelligence (2026) | AI Short Film | #specialforaaiff

It’s not just you

AI Just Crossed the Terrifying Line - Now What?

What AI Can Never Replace w/ Megan Kobzej | Cloud English Podcast Episode 91

The Only AI Tools You Need (12-Minute Guide)

No, Really — Is A.I. a Bubble? | The Ezra Klein Show

When AIs act emotional

このレッスンの語彙とスピーキングのポイント

この動画の重要語彙

動画に出てくる句動詞

繰り返し練習したい文

この動画の文法

注意したい発音

この動画での練習方法

シャドーイングとは？英語上達に効果的な理由