शैडोइंग अभ्यास: Building Anthropic | A conversation with our co-founders - वीडियो के साथ अंग्रेजी बोलना सीखें

Why are we working on AI in the first place?

I'm just going to arbitrarily pick Jared.

Why are you doing AI at all?

I mean, I was working on physics for a long time, and I got bored, and I wanted to hang out with more of my friends, so.

Yeah, I thought Dario pitched you on it.

I don't think I explicitly pitched you at any point.

I just showed you results of AI models

and was trying to make the point that they're very general and they don't only apply to one thing, and then just at some point after I showed you enough of them, and you were like, oh,

yeah, that seems like it's right.

How long have you been a professor when you started?

I think like six years or so.

I think I helped recruit Sam.

I talked to you, and you were like, I think I've created a good bubble here.

My goal is to get Tom to come back, and then it worked.

And did you meet everyone through Google when you were doing the interpretability stuff, Chris?

No. So I guess I actually met a bunch of you when I was 19, and I was visiting the Bay Area for the first time.

So I guess I met Dario and Jared then, and I guess they were postdocs, which I thought was very cool at the time.

And then I was working at Google Brain and Dario joined and we sat side by side actually for a while.

We had desks beside each other and I worked with Tom there as well.

And then of course, I got to work with all of you at OpenAI when I went there.

So I guess I've known a lot of you for like more than a decade, which is kind of wild.

If I remember correctly, I met Dario in 2015

when I went to a conference you were at

and I tried to interview you

and Google PR said I would have read all of your research papers that you needed to.

Yeah, I think I was writing Concrete Problems in AI Safety when I was at Google.

I think you wrote a story about that paper.

I did.

I remember right before I started working with you, I think you invited me to the office to come chat and just tell me everything about AI,

and you explained.

I remember afterwards being like, oh, I guess this stuff is much more serious than I realized, and you were probably explaining the big blob of compute

and parameter counting and how how many neurons are in the brain and everything.

I feel like Dario often has that effect on people.

This is much more serious than I realized.

Yeah, I'm the bringer of happy tidings.

But I remember when we were at OpenAI, where there was the scaling law stuff and just making things bigger, and it started to feel like it was working.

And then it kind of kept on eerily working on a bunch of different projects, which I think is how we all ended up working closely together,

because it was first GPT -2 and then scaling laws and GPT -3, Yeah, we're the plop of people that were making things work.

Yeah, that's right.

I think we're also excited about safety because in that era, there was sort of this idea that AI would become very powerful,

but like potentially not understand human values or not even be able to communicate with us.

And so I think we were all like pretty excited about language models as a way to kind of guarantee

that AI systems would have to understand kind of implicit knowledge that...

And RL for human feedback on top of language models, which was the whole reason for scaling these models up was that, you know, we couldn't do,

we couldn't, the models weren't smart enough to do RLHSF on top of.

So that's the kind of intertwinement of safety and scaling of the models that we, you know, still believe in today.

I think there was also an element of like the scaling work was done as part of the safety team

that Dario started at OpenAI because we thought

that forecasting AI trends was important to be able to have us take them seriously and take safety seriously as a problem.

Correct.

Yeah, I mean, we took, I remember being in some airport in England sampling from GPT -2

and using it to write fake news articles and slacking Dario and being like, oh, this stuff actually works and might have like huge policy implications.

I think Dario said something like, yes.

It's a typical way.

But then we worked on that a bunch, as well as the release stuff, which was kind of wild.

Yeah, I remember the release stuff.

I think that was when we first started working together.

That was a fun time, the GPT -2 launch.

But I think it was good for us, because we did a kind of slightly strange, safety -oriented thing all together.

And then we ended up doing Anthropic, which is a much larger, slightly strange, safety -oriented thing altogether.

So I guess just, like, going back to the concrete problems, because I remember, so I joined OpenAI, like, 2016, like, one of the first 20 employees or whatever with Udario.

And I remember at that time, the, like, concrete problems in AI safety seemed like it was, like, the first, like, mainstream, like, AI safety paper.

I guess I don't really know if I ever asked you what the story was for, like, how that came about.

Chris knows the story because he was involved in it.

I think, you know, we were both at Google.

I forget what other project I was working on, but like with many things, it was my attempt to kind of procrastinate from whatever other project I was working on

that I've now completely forgotten what it was.

But I think it was like Chris and I decided to write down.

what are some open problems in terms of AI safety

and also AI safety usually talked about in this very kind of abstruse, abstract way.

Can we kind of ground it in the ML that was going on at the time?

I mean, now there's been like six, seven years of work in that vein, but there was almost a strange idea at the time.

Yeah, I think there's a way in which it was almost a kind of political project where at the time, a lot of people didn't take safety seriously.

So I think that there was sort of this goal to collate a list of problems

that sort of people agreed were reasonable, often already existed in literature,

and then get a bunch of people across different institutions who are credible to be authors.

And I remember I had this whole long period where I

just talked to 20 different researchers at Brain to build support for publishing the paper.

In some ways, if you look at it in terms of the problems and a lot of things it emphasized, I think it hasn't held up that well in that I think it's not really the right problems.

But I think if you sort of see it instead as a consensus -building exercise, that there's something here that is real and that is worth taking seriously,

then it was a pretty important moment.

I mean, you end up in this really weird sci -fi world where I remember at the start of Anthropic, we were talking about constitutional AI.

And I think Jared said, oh, we're just going to write like a constitution for a language model and that'll change all of its behavior.

And I remember that sounded like incredibly crazy at the time.

But why did you guys think that was going to work?

because I remember that was one of the first early big research ideas we had at the company.

Yeah, I think Dario and I had talked about it for a while.

I think simple things just work really, really well in AI.

I think the first versions of that were quite complicated, but then we whittled away into just use the fact

that AI systems are good at solving multiple choice exams and give them a prompt that tells them what they're looking for.

That was what we needed.

Um, and then we were able to just write down these principles.

I mean, it goes back to like the big blob of computer, the bitter lesson or the scaling hypothesis.

If you can identify, you know, something that you can give the AI data for, and that's kind of a clear target, you'll get it to do it.

So like, here's this, here's this set of instructions.

Here's this set of principles.

AI language models can like read that set of principles

and they can like compare it to the behavior that they themselves are engaging in.

So like you've got your training target there.

100

So once you know that, I think my view and Jared's view is there's a way to get it to work.

101

You just have to fiddle with enough of the details.

102

Yeah.

103

I think it was always weird for me, especially in these early eras, because I was in physics and then coming from physics.

104

And I think now we forget about this because everyone's excited about AI.

105

But I remember talking to Dario about concrete problems and other things.

106

And I just got the sense that AI researchers were very, very kind of psychologically damaged by the AI winter,

107

where they just kind of felt like having like really ambitious ideas or ambitious visions was like very disallowed.

108

And that's kind of how I imagine it was in terms of talking about safety.

109

In order to care about safety, you have to believe that AI systems could actually be really powerful and really useful.

110

And I think that like there was kind of a prohibition against being ambitious.

111

And I think one of the benefits is that physicists are very arrogant.

112

And so they're constantly doing really ambitious things and talking about things in terms of grand schemes.

113

And so, yeah.

114

I mean, I think that's definitely true.

115

Like, I remember in 2014, it was like there were just like, I don't know, there were just like some things you couldn't say, right?

116

But I actually think it was kind of an extension of problems that exist across academia, other than maybe theoretical physics,

117

where they've kind of evolved into very risk -averse institutions for a number of reasons

118

and even the industrial parts of AI had kind of transplanted or forklifted that mentality.

119

And it took a long time.

120

I think it took until 2022 to get out of that mentality.

121

There's a weird thing about what does it mean to be conservative and respectful, where you might think one version you could have is

122

that what it means to be conservative is to take the risks

123

or the potential harms of what you're doing really seriously and worry about that.

124

But another kind of conservatism is to be like, taking an idea too seriously and believing that it might succeed is sort of like scientific arrogance.

125

And so I think there's like kind of two different kinds of conservatism or caution.

126

And I think we were sort of in a regime that was very controlled by that one.

127

I mean, you see it historically, right?

128

129

if you look at the early discussions in 1939 between people

130

involved in nuclear physics about whether nuclear bombs were sort of a serious concern,

131

you see exactly the same thing with like Fermi resisting these ideas because it just seemed kind of like a crazy thing.

132

And other people like Zillard or Teller are taking the ideas seriously because they were worried about the risks.

133

Perhaps the deepest lesson that I've learned in the last 10 years, and probably all of you have learned some form of it as well,

134

is there can be this kind of seeming consensus, these things that kind of everyone knows.

135

that, I don't know, seem sort of wise, seem like they're common sense,

136

but really they're just kind of herding behavior masquerading as maturity and sophistication.

137

And when you've seen, the consensus can change overnight.

138

And when you've seen it happen a number of times, you suspected, but you didn't really bet on it.

139

And you're like, oh man, I kind of thought this, but what do I know?

140

How can I be right and all these people are wrong?

141

You see that a few times, then you just start saying, nope, this is the bet we're going to make.

142

I don't know for sure if we're right, but just ignore all this other stuff, see it happen.

143

And I don't know, even if you're right 50 % of the time, being right 50 % of the time contributes so much.

144

You're adding so much that is not being added by anyone else.

145

And it feels like that's where we are today with some safety stuff, where there's a consensus view that a lot of this safety stuff is unusual or doesn't naturally fall out of the technology.

146

And then at Anthropic, we do all of this research where weird safety misalignment problems fall out as a natural dividend of the tech we're building.

147

So it feels like we're in that counter consensus view right now.

148

But I feel like that has been shifting over the past, even just like 18.

149

We've been helping.

150

We've definitely been.

151

Publishing and doing research.

152

Constant force.

153

But I also think just like world sentiment around AI has shifted really dramatically.

154

And, you know, it's more common in the user research that we do to hear just customers,

155

regular people, say, I'm really worried about what the impact of AI on the world more broadly is going to be.

156

Sometimes that means jobs or bias or toxicity, but it also sometimes means, is this just going to mess up the world?

157

How is this going to contribute to fundamentally changing how humans work together, operate, which I wouldn't have predicted that, actually.

158

For whatever reason, it seems like people in the ML research sphere have always

159

been more pessimistic about AI becoming very powerful than the general public.

160

Maybe it's a weird form of humility or something.

161

When Daria and I went to the White House in 2023, in that meeting, Harris and Raimondo and stuff basically said,

162

paraphrase, but basically said, we've got our eye on you guys.

163

AI is going to be a really big deal and we're now actually paying attention, which is...

164

And they're right.

165

They're absolutely right.

166

But I think in 2018, you wouldn't have been like, the president will call you to the White House to tell

167

you they're paying close attention to the development of language models yeah it's just like a crazy one thing

168

that i think is interesting too just uh is like i

169

guess like all of us kind of got into this

170

when it didn't seem like there was like like we thought like we thought we thought

171

that it could happen

172

but yeah it was it was like like fair me being

173

like skeptical of the atomic bomb it was like he had he was just a good scientist

174

and like there was some evidence that it could happen

175

but there also was a lot of evidence against it happening

176

and he I guess decided that it would be worthwhile because

177

if it was true then it would be a big deal

178

and I think for all of us it was like yeah like 2015, 2016, 2017 like there was some evidence and increasing evidence

179

that this might be a big deal

180

but like I remember in 2016 like talking to all my advisors

181

and I was like I've done startup stuff like I want to help out with the with like AI safety

182

but like I'm like not great at math I like don't

183

know don't know how to don't exactly know how i can do it um

184

and i think at the time people were like either were

185

like well you need to be super good at like decision theory in order to help out

186

and i was like that's probably not gonna work

187

or they were like uh like it doesn't really seem like we're gonna get some crazy ai thing um

188

and so i had uh only a few people basically

189

that were like like yeah okay

190

that seems like a good thing to do i remember in 2014 making graphs of image net results over time

191

when i was a journalist and trying to like get stories published about them

192

and people thought i was completely mad

193

and then i remember in 2015 trying to persuade bloomberg to let me write a story about nvidia

194

because every ai research paper had started mentioning the use of gpus

195

and they said that was completely mad and then in 2016

196

when i left journalism to go into ai i had these

197

emails saying you're making like the worst mistake of your life which i now occasionally look back on

198

but it was like it was seemed crazy at the time from like any many perspectives to go and take this seriously, that scaling was going to work,

199

and something was maybe different about the technology paradigm.

200

You're like Michael Jordan and that coach that didn't believe in him in high school.

201

How did you actually make the decision though?

202

Did you feel torn or was it obvious to you?

203

I did a crazy counter bet where I said, let me become your full -time AI reporter and double my salary, which I knew that they wouldn't say yes to.

204

Then I went to sleep and then I woke up and resigned.

205

It was all fairly relaxed.

206

You're just a decisive guy.

207

In that instance, I think it's because I was like going to work, reading archive papers and then printing archive papers off and coming home and reading archive papers,

208

including like Dario's papers and from the Baidu stuff and being like something like completely crazy is happening here.

209

And at some point I thought you should bet with conviction, which I think everyone here has done in their careers is just betting with conviction that this is.

210

gonna work.

211

Yeah.

212

I definitely, I definitely was not as decisive as you.

213

I spent like six months like, like flopping, like, okay, like, should I, should I do it?

214

Like, should I try to do a startup?

215

Should I try to do this thing?

216

But I also feel like back then there were, there wasn't as much talk of engineers and the impact that an engineer can have in AI, right?

217

That feels so natural to us now.

218

And we're at the same sort of talent raise for engineers of all different types.

219

But at the time it was like, you're a researcher and that's the only people that can work on AI.

220

So I don't think it was crazy that you were spending time thinking about that.

221

Yeah, yeah.

222

And I think that that was basically the thing that got me to join OpenAI was, like, I, like, I messaged the people there and they were, like, yeah, we actually think that you can help out by doing engineering work.

223

Yeah.

224

And, like, that you can help out with AI safety in that way, which I think there hadn't really been an opportunity for that.

225

So that was what got me there.

226

That's right.

227

You were my manager at OpenAI.

228

I was.

229

That's right.

230

I think I joined after you'd been there for a while.

231

A little bit.

232

I was at Brain for a bit.

233

Yeah.

234

I don't know if I ever asked you like what it was that got you to join.

235

Yeah.

236

So I had been at Stripe for about five and a half years and I knew Greg.

237

He had been my boss.

238

He was my boss at Stripe for a while.

239

And I actually introduced him and Dario because I said when he was starting OpenAI, I was like the smartest person that I know is Dario.

240

You should, you would be really lucky to get him.

241

So Dario was at OpenAI.

242

I had a few friends from Stripe that had gone there too.

243

And I think sort of like you, I'd been thinking about what I wanted to do after Stripe.

244

I had gone there just because I wanted to get more skills after working in, you know, nonprofit and international development.

245

And I actually thought I was going to go back to doing that.

246

Like essentially, I had always been working.

247

I was like, I really want to help people that have, you know, less than I do.

248

But I didn't have the skills when I was doing it before Stripe.

249

And so I looked at going back to public health.

250

I thought about going back into politics very briefly.

251

But I was also looking around at other tech companies and other sort of ways of having impact.

252

And OpenAI at the time felt like it was a really nice intersection.

253

It was a nonprofit.

254

They were working on this really big lofty mission.

255

I really believed in sort of the AI, you know, potential because, I mean, I know Dario a little bit.

256

And so he was - They needed management help.

257

They definitely needed management help.

258

That is a fact.

259

And so there was, I think that felt, it felt very me -shaped, right? i was like oh there's this non -profit

260

and they like there's all these really great people with these like really good intentions

261

but it seems like they're a little bit of a mess um

262

and that was that was that felt really exciting to me to get to come in

263

and even you know just i was i was such a utility player right i was running like people

264

but i was also running some of the technical teams yeah the scaling

265

or i worked on the language team i took over i

266

worked on policy we worked on some policy stuff i worked with chris

267

and i felt like there was just so much goodness in

268

so many the employees there and i felt a very strong desire to come

269

and sort of just try to help make the company a

270

little more functional i remember towards the end after we've done

271

gpt free you were like have you guys heard of something called trust

272

and safety yes i said you know i used to run some trust

273

and safety teams at stripe there's a thing called trust

274

and safety that you might want to consider for a technology like this

275

and it's funny because it sort of is the intermediary step between, you know, AI safety research,

276

right, which is how do you actually make the model safe to something just much more practical.

277

I do think there was a value in saying, you know, this is going to be a big thing.

278

We also have to be doing this sort of practical work day to day to build the muscles for

279

when things are going to be a lot higher stakes.

280

That might be a good transition point to talk about things like a responsible scaling policy

281

and how we came up with that or why we came up with it and how we're using it now, especially given how much trust and safety work we do on today's models.

282

So whose idea was the RSP?

283

Yeah, it was me and Paul first talked about it in late, Paul Cristiano in late 2022.

284

First, it was like, oh, should we cap scaling at a particular point until we've discovered how to solve certain safety problems?

285

And then it was like, well, it's strange to have this one place where you cap it, and then you uncap it.

286

So let's have a bunch of thresholds.

287

And then at each threshold you have to do certain tests to see

288

if the model is capable and you have to take increasing safety

289

and security measures originally we had this idea

290

and then the thought was just look like you know this will go better

291

if you know

292

if it's done by some third party like we shouldn't we

293

shouldn't be the ones to do it right it shouldn't come from one company

294

because then other companies are less likely to adopt it so paul actually went off and designed it and and you know, many, many features of it changed.

295

And we were kind of on our side, working on how it should work.

296

And, you know, once Paul had something together, then pretty much immediately after he announced the concept, we announced ours within a month or two.

297

I mean, many of us were heavily involved in it.

298

I remember writing at least one draft of it myself, but there were like several drafts of it.

299

There were so many drafts.

300

I think it's gone through the most drafts any doc

301

which makes sense right it's like like i feel like it is in the same way that like the U .S.

302

treats, like, the Constitution as, like, the holy document, which, like, I think is just a big thing that, like, strengthens the U .S.

303

Yes.

304

And, like, we don't expect the U .S to go off the rails in part because just, like, every single person in the U .S is, like, the Constitution is a big deal, and if you tread on that, like, I'm mad.

305

Yeah.

306

Like, I think that, like, the RSP is our, like, it holds that thing.

307

It's, like, the holy document for Anthropic, so it's, like, worth doing a lot of iterations getting it right.

308

Some of what I think has been so cool to watch about the RSP, like development at Aetheropic too, is it feels like it has gone through so many different phases

309

and there's so many different skills that are needed to make it work, right?

310

There's like the big ideas, which I feel like Dario and Paul and Sam and Jared and so many others are like, what are the principles?

311

Like, what are we trying to say?

312

How do we know if we're right?

313

But there's also this very like operational approach to just iterating where we're like, well, we thought that we were going to see this at this, you know, safety level and we didn't.

314

So should we change it so that we're making sure that we're holding ourselves accountable?

315

And then there's all kinds of organizational things, right?

316

We just were like, let's change the structure of the RSP organization for clearer accountability.

317

And I think my sense is that for a document that's as important as this, right?

318

I love the constitution analogy.

319

It's like, there's all of these bodies and systems

320

that exist in the U .S to like make sure that we follow the constitution, right?

321

There's the courts, there's the Supreme Court, there's the presidency, there's the you know the both houses of of you know congress

322

and they do all kinds of other things of course

323

but there's like all of this infrastructure

324

that you need around this like one document

325

and i feel like we're also learning

326

that lesson here i think it sort of reflects a view we a lot of us have about safety which is

327

that it's it's a solvable problem it's just a very very hard problem that's going to take tons

328

and tons of work yeah all these institutions

329

that we need to build up like there's there's all sorts

330

of institutions built up around like automotive safety build up over many, many years.

331

But we're like, do we have the time to do that?

332

We've got to go as fast as we can to figure

333

out what the institutions we need for AI safety are

334

and build those and try to build them here first but make it exportable.

335

It forces unity also because if any part of the org is not kind of in line with our safety values,

336

it shows up through kind of the RSP, right?

337

The RSP is going to block them from doing what they want to do.

338

And so it's a way to remind everyone over and over again, basically to make safety a product requirement, part of the product planning process.

339

And so, like, it's not just a bunch of kind of like bromides that we repeat.

340

It's something that you actually, if you show up here and you're not aligned, you actually run into it.

341

And, like, you either have to learn to get with the program or it doesn't work out.

342

The RSP has become kind of funny over time because we spend thousands of hours of work on it.

343

And then I go and talk to senators and I explain the RSP and I'm like, we have some stuff that means it's hard to steal what we make and also that it's safe.

344

And they're like, yes, that's a completely normal thing to do.

345

Are you telling me not everyone does this?

346

And you're like, oh, okay, yeah.

347

It's the exact truth that not everyone does this.

348

But that kind of, it's amazing.

349

We've and so much effort on it here.

350

And when you boil it down, they're like, yes, that sounds like a normal way.

351

Yeah, that sounds good.

352

That's been the goal.

353

Like Daniela was saying, like, let's make this as boring and normal.

354

Like, let's make this a finance thing.

355

Yeah, imagine it's like an audit.

356

Yeah, yeah.

357

No, boring and normal is what we want.

358

Certainly in retrospect.

359

Well, also, Dario, I think in addition to driving alignment, it also drives clarity because it's really, it's written down what we're trying to do

360

and it's legible to everyone in the company

361

and it's legible externally what we think we're supposed to be aiming towards from a safety perspective.

362

It's not perfect.

363

We're iterating on it.

364

We're making it better.

365

But I think there's some value in saying like, this is what we're worried about.

366

This thing over here, like you can't just use this word to sort of derail something in either direction, right?

367

To say, oh, because of safety, we can't do X or because of safety, we have to do X.

368

We're really trying to make it clearer what we mean.

369

Yeah.

370

You can't, it prevents you from worrying about every last little thing under the sun, right?

371

Because it's actually the fire drills that damage the cause of safety in the long run.

372

I've said, like, if there's a building and, you know, the fire alarm goes off every week, like, that's a really unsafe building.

373

Because when there's, like, actually a fire, you're just going to be like, oh, it just goes off all the time.

374

So it's very important to be calibrated.

375

Yeah.

376

A slightly different frame that I find kind of clarifying is

377

that I think the RSP creates healthy incentives at a lot of levels.

378

So I think internally, it aligns the the incentives of every team with safety, because it means that if we don't make progress on safety, we're going to block.

379

I also think that externally, it creates a lot of healthier incentives than other possibilities, at least that I see, because it means that, you know, if we at some point have to take some kind of dramatic action,

380

like if at some point we have to say, you know, our model, we've reached some point and we can't yet make a model safe, it aligns that with sort of the point where there's evidence that supports that decision.

381

And there's sort of a pre -existing framework for thinking about it and it's legible.

382

And so I think there's a lot of levels of which the RSP, I think in ways that maybe I didn't initially understand when we were talking about the early versions of it.

383

It creates a better framework than any of the other ones that I've thought about.

384

I think this is all true, but I feel like it undersells how challenging it's been to figure out what the right policies

385

and evaluations and what the lines should be.

386

I think that we have and continue to iterate a lot on that.

387

I think there is a question also that's difficult.

388

You can be at a point where it's very clear something's dangerous or very clear that something's safe, But with some technology that's so new, there's actually like a big gray area.

389

And so I think that has been like all the things that we're saying were things that made me really, really excited about the RSP at the beginning and still do.

390

But also, I think, enacting this in a clear way

391

and making it work has been much harder and more complicated than I anticipated.

392

Yeah, I think this is exactly the point.

393

The gray areas are impossible to predict there's

394

so many of them like until you actually try to implement

395

everything you like don't know what's going to go wrong

396

so what we're trying to do is go and implement everything

397

so we can see as early as possible what's going to go wrong

398

so yeah you have to you have to do three

399

or four passes before before you really you really get it right like iteration is just very powerful

400

and you know you're not going to get it you're not going to get it right on the first time

401

and so you know

402

if the stakes are increasing you want to get your iterations

403

in early you don't want to get them in light you're also building the for like internal institutions

404

and processes so the specifics might change a lot

405

but building the muscle of just doing it is the really

406

valuable thing i'm i'm responsible for like compute at anthropic that's important

407

so thank you so like i think

408

that like for me for me like i guess we have

409

to deal we have to deal with external folks yeah

410

and different external folks kind of are on different spectrums of

411

the like how fast do they think stuff is going again

412

and like I think that's also been a thing where like I started out like not thinking stuff would be

413

that fast and have changed over time and

414

so I have some I have sympathy for that and

415

so I think the RSP has been extremely useful for me in communicating with people who think

416

that things might take longer

417

because then we have a thing where it's like we don't

418

need to do extreme safety measures until stuff gets really intense

419

and then we can be like

420

that they might be like I don't think stuff will get intense for a long time

421

and then I'll be like okay yeah we don't have to do extreme safety measures, and so that makes it a lot easier to communicate with other folks externally.

422

Yeah, yeah, it makes it like a normal thing you can talk about, rather than something really strange.

423

Yeah.

424

How else is it, like, showing up for people?

425

Evals, evals, evals.

426

Good.

427

It's all about evals.

428

Everyone's doing evals.

429

Like, the training team is doing evals all the time.

430

We're trying to figure out, like, has this model gotten enough better that it has the potential to be dangerous?

431

So, how many teams do we have that are evals teams?

432

Frontier red team, there must be, I mean, there's a lot of people.

433

Every team produces evals, basically.

434

And that means you're just measuring against the RSP, like measuring for certain signs of things that would concern you or not concern you.

435

Exactly.

436

Like it's not, it's easy to lower bound the abilities of a model, but it's hard to upper bound.

437

So we just put tons and tons of research effort into saying, can this model do this dangerous thing or not?

438

Maybe there's some trick that we haven't thought of, like chain of thought or best event

439

or some kind of tool use that's going to make it so it can help you like do something very dangerous?

440

It's been really useful in policy

441

because it's been a really abstract concept what safety is and

442

when I'm like we have an eval which changes whether we deploy the model

443

or not and then you can go and calibrate with policymakers

444

or experts in national security or some of these CBRN areas

445

that we do to actually help us build evals that are well calibrated and that counterfactually just wouldn't have happened otherwise.

446

But once you've got the specific thing, people are a lot more motivated to help you make it accurate.

447

So it's been useful for that.

448

The RRSP shows up for me for sure.

449

Often.

450

I actually think the, weirdly, the way that I think about the RRSP the most is like what it sounds like.

451

Just like the tone.

452

I think we just did a big rewrite of the tone of the RRSP

453

because it felt overly like technocratic and even a little bit adversarial.

454

I spent a lot of time thinking about how do you build a system

455

that people just want to be a part of, right?

456

It's so much better if the RSP is something that everyone in the company can walk around and tell you, just like with OKRs like we do right now, what are the top goals of the RSP?

457

How do we know if we're meeting them?

458

What AI safety level are we at right now?

459

Are we at ASL 2?

460

Are we at ASL 3?

461

That people know what to look for because

462

that is how you're going to have good common knowledge of if something's going wrong, right?

463

If it's overly technocratic and it's something that only particular people in the company feel is accessible to them.

464

It's just like not as productive, right?

465

And I think it's been really cool to watch it sort of transition into this document where I actually think most, if not everybody at the company,

466

regardless of their role, could read it and say, this feels really reasonable.

467

I want to make sure that.. we're building AI in the following ways

468

and I see why I would be worried about these things.

469

And I also kind of know what to look for if I bump into something, right?

470

It's almost like, make it simple enough that if you are working at a manufacturing plant and you're like, huh, it looks like the safety seatbelt on this should connect this way,

471

but it doesn't connect that you can spot it.

472

And that there's just like healthy feedback flow between leadership

473

and the board and the rest of the company and the people that are actually building it.

474

Because I actually think the way this stuff goes wrong in most cases is just like the wires don't connect

475

or like they get crossed.

476

And that would just be like a really sad way for things to go wrong, right?

477

It's just all about operationalizing it, making it easy for people to understand.

478

Yeah, the thing I would say is none of us wanted to found a company.

479

We felt like it was our duty, right?

480

I felt like we had to.

481

Like we have to do this thing.

482

This is the way we're going to make things go better for the AI.

483

Right.

484

Like, that's also why we did the pledge.

485

Right.

486

Yeah.

487

Because we're like, the reason we're doing this is feels like our duty.

488

I wanted to invent and discover things in some kind of beneficial way.

489

That was that was how I came to it.

490

And that led to working on AI and AI required a lot of engineering and eventually AI required a lot of capital.

491

But what I found was that if you don't do this in a way where you're setting the environment,

492

where you set up the company, then a lot of it gets done.

493

A lot of it repeats the same mistakes that I found so alienating about the tech community.

494

It's the same people.

495

It's the same attitude.

496

It's the same pattern matching.

497

And so at some point, it just seemed inevitable that we need to do it in a different way.

498

When we were hanging out in graduate school, I remember you had kind of this whole like program of

499

trying to figure out how to do science in a way that would sort of advance the public good.

500

And I think I think that's like pretty similar to how we we think about this.

501

Maybe you had like this like Project Vannevar or something to do that.

502

I was a professor.

503

I think basically I just looked at the situation and I was convinced that AI was on a very,

504

very, very steep trajectory in terms of impact didn't seem like because of the necessity for capital that as a physics professor, I could continue doing that.

505

I wanted to work with people that I trusted in building an institution to try to make AI go well.

506

But yeah, I would never recommend founding a company or really want to do it.

507

I mean, yeah, I think it's just a means to an end.

508

I think that's usually how things go well, though.

509

If you're doing something just to enrich yourself or gain power, you have to to actually care about accomplishing a real goal in the world,

510

and then you find whatever means you have to.

511

Well, something I think about a lot is just a strategic advantage for us.

512

It sounds really funny to say, but just how much trust there is at this table, right?

513

I think that's not—Tom, you were at other startups.

514

I was never a founder before, but it's actually really hard to get a big group of people to have the same mission.

515

Right.

516

And I think the thing that I feel like the happiest about

517

when I come into work and probably the most proud of at Anthropic is how well

518

that has scaled to a lot of people.

519

It feels to me like in this group and with the rest of leadership, everyone is here for the mission and our mission is really clear.

520

And it's it's very pure.

521

Right.

522

And I think that is something that I don't see as often to Dario's point in sort of the tech industry.

523

It feels like there's just a wholesomeness to what we're trying to do.

524

I agree.

525

None of us were like, let's just go found a company.

526

I felt like we had to do it.

527

It just felt like we couldn't keep doing what we were doing the place we were doing it.

528

We had to do it by ourselves.

529

It felt like with GPT -3, which all of us had touched or worked on, and scaling laws and everything else, we could see it in front of us in 2020.

530

It felt like, well, if we don't do something soon altogether, you're going to hit the point of no return.

531

You have to do something to have any ability to change

532

the environment i i think building up danielle i do i do think

533

that um there's just like a lot of trust in this group i think like each of us knows

534

that we we got into this

535

because we want to help out with the world yeah um

536

we did the like 80 like the 80 pledge thing and

537

that was like i think

538

that everybody everybody was just like yes obviously we're going to do this it was it was

539

and yeah i i yeah i i do think that the

540

that the the trust thing is a special thing that's that's

541

extremely rare yeah i credit daniella with like keeping the keeping

542

the bar high i i credit you with the fact

543

that we scaled people say how nice people are here

544

which is actually uh a wildly important thing i think anthropic is really low politics

545

and of course we all have a different vantage point than average

546

and i try to remember it's because of low ego but But it's low ego, and I do think our interview process and just the type of people who work here, there's almost an allergic reaction.

547

to politics and unity unity unity is so important the idea

548

that the product team the research team the trust

549

and safety team you know the go -to -market team the policy team like

550

that this the safety folks they're all trying to contribute to

551

kind of the same goal the same mission of the company right i think it's i think it's dysfunctional

552

when different parts of the company are are think they're trying to accomplish different things think the company is about different things,

553

or think that other parts of the company are trying to undermine what they're doing.

554

And I think the most important thing we've managed to preserve is, and again, things like the RSP drive it, this idea that it's not,

555

you know, there's some parts of the company causing damage and other parts of the company trying to repair it, but that there are different parts of the company doing different functions,

556

and they all function under a single theory of change.

557

Extreme pragmatism, right?

558

But, you know, the reason I went to OpenAI in the first place, you know, it was a nonprofit.

559

It was a place where I could go and focus on safety.

560

And I think over time, you know, that maybe wasn't as good a fit.

561

And there were some difficult decisions.

562

And I think in a lot of ways, I really trusted Dario and Danielle on that.

563

But I didn't want to leave.

564

That was like something that I think I was actually like pretty reluctant to go along with.

565

Because I think, for one thing, I didn't know that it was good for the world to have more AI labs.

566

And I think it was something that I was pretty reluctant for.

567

And I think, as well, when we did leave, I think I was reluctant to start a company.

568

I think I was arguing for a long time that we should do a nonprofit instead and just focus on safety research.

569

And I think it really took pragmatism and confronting the constraints

570

and just being honest about what the constraints implied for accomplishing that mission that led to Anthropic.

571

I think just a really important lesson

572

that we were good about early on is like make less promises and keep more of them, right?

573

Like try to be calibrated,

574

be realistic, confront the trade -offs because trust and credibility are more important than any particular policy.

575

It is so unusual to have what we have and watching like Mike Krieger defend safety things of,

576

like, reasons why we shouldn't ship a product yet, but also then to watch Vinay sort of say, like, okay, we have to do the right thing for the business.

577

Like, how do we get this across the finish line?

578

And to hear, you know, people, like, deep in the technical safety org talking about how it's also important

579

that we build things that are practical for people and hearing, you know, engineers on inference talk about safety, that's amazing.

580

Like, I think that is, again, one of the most special things about working here is everybody with that unity is prioritizing the pragmatism, the safety, the business.

581

That's wild.

582

I think about it as spreading the trade -offs from just the leadership of the company to everyone, right?

583

I think the dysfunctional world is like, you have a bunch of people who only see a big, you know, safety is like, we always have to do this.

584

And product is like, we always have to do this.

585

And research is like, you know, this is the only thing we care about.

586

And then you're stuck at the top.

587

You're stuck at the top.

588

You have to decide between.

589

You don't have as much information as either of them.

590

That's the dysfunctional world.

591

The functional world is when you're able to communicate to everyone.

592

There are these trade -offs we're all facing together.

593

The world is a far from perfect place.

594

There's trade -offs.

595

Everything you do is going to be suboptimal.

596

Everything you do is going to be some attempt to get the best of both worlds that that, you know, doesn't work out as well as you thought it was.

597

And everyone is on the same page about confronting those trade -offs together.

598

And they just feel like they're confronting them from a particular post, from a particular job as part of the overall job of confronting all the trade -offs.

599

It's a bet on race to the top, right?

600

It's a bet on race to the top.

601

Like it's not a pure upside bet.

602

Things could go wrong.

603

But like, we're all aligned on like, this is the bet that we're making.

604

And markets are pragmatic.

605

So if the more successful and profit becomes as a company, the more incentive there is for people to copy the things that make us successful.

606

And the more that success is tied to actual safety stuff we do,

607

the more it just creates a like gravitational force in the industry that will actually get the rest of industry to compete.

608

And it's like, sure, we'll like build seatbelts and everyone else can copy them.

609

That's good.

610

It's like good world.

611

Yeah, this is the race to the top, right?

612

But if you're saying, well, we're not going to build the technology, you're not going to build it better than someone else,

613

that in the end, that just doesn't work because you're not proving that it's possible to get from here to there.

614

Where the world needs to get, never mind the industry, never mind one company, is it needs to get us successfully through from this technology does, doesn't exist.

615

to the technology exists in a very powerful way and society has actually managed it.

616

And I think the only way that's going to happen is

617

that if you have at the level of a single company and eventually at the level of the industry, you're actually confronting those trade -offs.

618

You have to find a way to actually be competitive, to actually lead the industry in some cases, and yet manage to do things safely.

619

And if you can do that, the gravitational pull you exert is so great.

620

There are

621

so many factors from the regulatory environment to the kinds of

622

people who want to work at different places to even sometimes the views of customers

623

that drive in the direction of if you can show that you can do well on safety without sacrificing competitiveness, right?

624

If you can find these kind of win -wins, then others are incentivized to do the same thing.

625

I think that's why getting things like the RSP right is so important, because I think that we ourselves,

626

seeing where the technology is headed, have often thought, oh, wow, we need to be really careful of this thing.

627

But at the same time, we have to be even more careful not to be crying wolf, saying that innovation needs to stop here.

628

We need to find a way to make AI useful,

629

innovative, delightful for customers, but also figure out what the constraints really have to be

630

that we can stand behind that make systems safe, so that it's possible for others to think that they can do that too, and they can succeed, they can compete with us.

631

We're not doomers, right?

632

We want to build the positive thing.

633

We want to build the good thing.

634

And we've seen it happen in practice.

635

A few months after we came out with our RRSP, the three most prominent AI companies had one, right? interpretability research, that's another area we've done it.

636

Just the focus on safety overall, like collaboration with the AI safety institutes, other areas.

637

Yeah, the Frontier Red team got cloned almost immediately, which is good.

638

You want all the labs to be testing for very, very scary risks.

639

Export the seatbelts.

640

Well, Jack also mentioned it earlier, but customers also really care about safety, right?

641

Customers don't want models that are hallucinating.

642

They don't want models that are easy to jailbreak.

643

They want models that are helpful and harmless, right?

644

And so a lot of the time what we hear in customer calls is just we're going with Claude

645

because we know it's safer.

646

I think that is also a huge market impact, right?

647

Because our ability to have models that are trustworthy and reliable, that matters for the market pressure that it puts on competitors, too.

648

Maybe to unpack something that Dario said a little bit more.

649

I think there's this narrative or this idea that maybe the virtuous thing is to almost like nobly fail, right?

650

It's like you should go and like put safety, you should go and put things, you should sort of demonstrate like in an impragmatic way so

651

that you can sort of demonstrate your purity to the cause or something like this.

652

And I think if you do that, it's actually very self -defeating.

653

For one thing, it means

654

that you're going to have the people who are making decisions be self -selected for being people who don't care

655

and for people who aren't prioritizing safety and who don't care about it.

656

And I think on the other hand, if you try really hard to find the way to align the incentives

657

and make it so that if there are hard decisions, they happen at the points where there is the most force to go

658

and support making the correct hard decisions and where there's the most evidence, then you can sort of start to trigger this race to the top that Dario is describing,

659

where instead of going and having the people who care get pushed out of influence, you instead pull other people to have to go and follow.

660

So what are you all excited about when it comes to the next things we'll be working on?

661

I think there's a bunch of reasons you can be excited about Interpreal.

662

One is obviously safety.

663

But there's another one that I think I find at an emotional level, you know, equally exciting or equally meaningful to me,

664

which is just that I think neural networks are beautiful.

665

And I think that there's a lot of beauty in them that we don't see.

666

We treat them like these black boxes that we're not particularly interested in the internal stuff.

667

But when you start to go and look inside them, they're just full of amazing, beautiful structure.

668

You know, it's sort of like if people looked at biology and they were like, you know, like evolution is really boring.

669

It just like it's a simple thing that goes and like runs for a long time and then it makes animals.

670

And like instead, it's like actually, you know, each one of those animals that evolution produces

671

and I think that it's an optimization process like training a neural network.

672

They're full of incredible complexity and structure and we have an entire sort of artificial biology inside of neural networks.

673

If you're just willing to look inside them, there's all of this amazing stuff and I think

674

that we're just starting to slowly unpack it and it's incredible and there's so much there.

675

But it's just.. so much to be discovered there.

676

We're just starting to crack it open.

677

And I think it's going to be amazing and beautiful.

678

And sometimes I imagine, you know, like a decade in the future, walking into a bookstore and buying, you know, the textbook on neural network interpolability, or really like on the biology of neural networks

679

and just the kind of wild things that are going to be inside of it.

680

And I think that in the next decade, we're going to, in the next couple of years even, we're going to go and start to go in and really discover all of those things.

681

And it's to be wild and incredible.

682

It's also going to be great that you get to buy your own textbook.

683

I'm excited that a few years ago, if you had said governments will set up new bodies to test

684

and evaluate AI systems and they will actually be competent and good, you would have not thought that was going to be the case, but it's happened.

685

It's like governments have built these new new embassies almost to deal with this new kind of class of technology

686

or like thing that Chris studies.

687

And I'm just very excited to see where that goes.

688

I think it actually means that we have state capacity to deal with this kind of societal transition.

689

So it's not just companies.

690

And I'm excited to help with that.

691

I'm already excited about this to, you know, to a certain extent today.

692

But I think just imagining the future world of what AI is going to be able to do for people is, it's impossible to not feel excited about that.

693

Dario talks about this a lot, but I think even just the sort of glimmers of Claude being able to help with,

694

you know, vaccine development and cancer research and biological research is crazy, like just to be able to watch what it can do now.

695

But when I fast forward, you know, three years in the future or five years in the future, imagining that Claude could actually solve so many of the fundamental problems

696

that we just face as as humans just even just from a health perspective alone even

697

if you sort of take everything else out feels really exciting

698

to me just like thinking back to my international development times it would be amazing

699

if Claude was responsible for helping to do a lot of the work

700

that I was trying to do a lot less effectively when I was like 25.

701

I mean I guess similarly I'm excited to build Claude for

702

work like I'm excited to build like I'm excited to build Claude into the company

703

and into companies all over the world i guess i'm excited

704

just for uh i guess like personally like i like using

705

claude a lot like i definitely uh there's been increasing amounts

706

of like home times with like me just like chatting with

707

john claude about stuff um i think the biggest recent thing

708

has been um uh code where like six months ago like

709

i didn't use claude to do any coding work like our teams didn't really use Claude

710

that much for for coding

711

and now it's like just face difference like i i gave a talk at yc like uh week before last

712

and at the beginning i just asked like okay

713

so like how many how many folks here use Claude for for coding now

714

and literally 95 of hands like all the hands in the room

715

which just like is totally different than how it was four months ago

716

or whatever yeah so when i think about what i'm excited about i think about places where, like I said before, where there's this consensus that,

717

again, seems like consensus, seems like what everyone wise thinks, and then it just breaks.

718

Places where I think that's about to happen and it hasn't happened yet.

719

One of them is interpretability.

720

I think interpretability is both the key to steering and making safe AI systems, and we're about to understand.

721

And interpretability contains insights about intelligent optimization problems and about how the human brain works.

722

I've said, and I'm really not joking, Chris Ola is going to be a future Nobel medicine laureate.

723

I'm serious.

724

I'm serious because a lot of these, I used to be a neuroscientist, a lot of these mental illnesses, the ones we haven't figured out, right?

725

Schizophrenia or the mood disorders, I suspect there's some higher level system thing going on and

726

that it's It's hard to make sense of those with brains

727

because brains are so mushy and hard to open up and interact with.

728

Neural nets are not like this.

729

They're not a perfect analogy, but as time goes on, they will be a better analogy.

730

That's one area.

731

Second is related to that, I think, just the use of AI for biology.

732

Biology is an incredibly difficult problem.

733

People continue to be skeptical for a number of reasons.

734

I think that consensus is starting to break.

735

We saw a Nobel Prize in chemistry awarded for alpha fold, remarkable accomplishment.

736

We should be trying to build things that can help us create 100 alpha folds.

737

And then finally, using AI to enhance democracy.

738

We worry about if AI is built in the wrong way, it can be a tool for authoritarianism how can AI be a tool for freedom and self -determination.

739

I think that one is earlier than the other two, but it's going to be just as important.

740

Yeah.

741

I mean, I guess two things that at least connect to what you were saying earlier.

742

I mean, one is I feel like people frequently

743

join Anthropic because they're sort of scientifically really curious about AI

744

and then kind of get convinced by AI progress to sort

745

of share the vision of the need not just to advance the technology, but to understand it more deeply and to make sure that it's safe.

746

And I feel like it's actually just sort of exciting to have people that you're working with,

747

like kind of more and more united in their vision for both what AI development looks like

748

and the sort of sense of responsibility associated with it.

749

And I feel like that's been happening a lot due to a lot of advances that have happened in the last year, like what Tom talked about.

750

Another is that, going back really to concrete problems, I feel like we've done a lot of work on AI safety up until this point, a lot of it's really important.

751

But I think we're now, with some recent developments, really getting a glimmer of what kinds of risks might literally come about from systems that are very,

752

very advanced so that we can investigate and study them directly with interpretability, with other kinds of safety mechanisms,

753

and really understand, like, what the risks from very advanced AI might look like.

754

And I think that that's something that is really going to allow us to sort of further the mission in a, like, really deeply scientific empirical way. And

755

so I'm excited about sort of the next six months of

756

how we use our understanding of what can go wrong with advanced systems to characterize

757

that and figure out how to avoid those pitfalls.

758

Perfect.

759

Finn.

760

I did it.

761

Good job.

762

This is the only time we ever get to catch up.

📺 उसी चैनल से

✨ अनुशंसित वीडियो

179 वाक्य

Understand AI in 14 minutes – with Anthropic's Chloe Lubinski [ARC 2026]

Business EnglishDaily ConversationUniversity Students

55 वाक्य

OpenAI’s lead engineer was secretly a spy...

Daily ConversationUniversity StudentsWorking Adults

346 वाक्य

Trust the Process: How to Design Teams That Actually Work

Daily ConversationUniversity StudentsWorking Adults

201 वाक्य

Cleo Abram Answers The Web's Most Searched Questions | WIRED

Business EnglishDaily ConversationUniversity Students

182 वाक्य

We Let $200 GPT-Pro Think for 7 Days on Unsolved Math

Daily ConversationUniversity StudentsWorking Adults

215 वाक्य

Creator of C# and TypeScript: "AI will NEVER Replace Coders, Here's Why" | Anders Hejlsberg

Daily ConversationUniversity StudentsWorking Adults

671 वाक्य

GS.TS. Po-Shen Loh, ĐH Carnegie Mellon | Rèn tư duy ngược để không bị AI bào mòn | TQKS #129

Daily ConversationUniversity StudentsWorking Adults

74 वाक्य

AI & Inequality | Daron Acemoglu: Big Tech poses risks as AI reshapes society

Daily ConversationUniversity StudentsWorking Adults

182 वाक्य

How to Become an AI Product Manager in 2026 | Ex-Google, Microsoft

Daily ConversationUniversity StudentsWorking Adults

638 वाक्य

Jensen Huang: The Mindset That Built NVIDIA

Daily ConversationUniversity StudentsWorking Adults

106 वाक्य

Most devs don't understand what agents are

Daily ConversationUniversity StudentsWorking Adults

54 वाक्य

Did Google just kickstart the intelligence explosion?

Business EnglishDaily ConversationUniversity Students

इस पाठ की शब्दावली और बोलने के सुझाव

यह C1 स्तर का स्पीकिंग पाठ वीडियो “Building Anthropic” पर आधारित है। इस वीडियो में शैडोइंग के लिए 762 वाक्य और 10613 शब्द हैं। बोला गया हिस्सा 51:48 लंबा है। वक्ता तेज़ बोलता है, लगभग 205 शब्द प्रति मिनट, इसलिए जुड़ी हुई और कमज़ोर ध्वनियाँ ज़्यादा मिलेंगी। 86% शब्द अंग्रेज़ी के 3,000 सबसे आम शब्दों में से हैं; बाकी शब्द अभ्यास से पहले देख लेना अच्छा रहेगा।

इस वीडियो की मुख्य शब्दावली

वीडियो के 9 सबसे कठिन शब्द, उच्चारण और अर्थ के साथ:

शब्द	उच्चारण	अर्थ
neural विशेषण	/ˈnʊɹəl/	तंत्रिकीय
pragmatism संज्ञा	/ˈpɹæɡmətɪzəm/	अर्थक्रियावाद, व्यावहारिकतावाद
archive संज्ञा	/ˈɑɹˌkaɪv/	पुरालेख
conservatism संज्ञा	/kənˈsɜːvətɪzəm/	रूढ़िवाद
mentality संज्ञा	/mɛnˈtæləti/	मानसिकता
textbook संज्ञा	/ˈtɛkst.bʊk/	पाठ्यपुस्तक, टेक्स्ट-बुक
decisive विशेषण	/dɪˈsaɪsɪv/	निश्चयार्थक, निर्णायक
researcher संज्ञा	/ˈɹiˌsɝ.t͡ʃɚ/	शोधकर्ता, शोधक
accomplish क्रिया	/əˈkɑm.plɪʃ/	सफलता, उपलब्धि

वीडियो में आने वाली फ़्रेज़ल वर्ब

शब्द	अर्थ
come back क्रिया	वापस आना

दोहराने लायक वाक्य

वीडियो के छोटे, पूरे वाक्य जिन्हें आप रोज़ की बातचीत में इस्तेमाल कर सकते हैं:

How did you actually make the decision though?
Should I try to do this thing?
How do we know if we're right?
Are you telling me not everyone does this?
How do we know if we're meeting them?

इस वीडियो का व्याकरण

वक्ता जिन संरचनाओं का सबसे अधिक प्रयोग करता है, वीडियो के असली शब्दों के साथ:

संरचना	वीडियो में
Present perfect continuous have/has been + -ing — पहले शुरू हुआ काम जो अब भी जारी है	has been shifting · We've been helping · there's been increasing
शर्त वाले वाक्य if + उपवाक्य, will/would + क्रिया — शर्त और उसका नतीजा	if it was true then it would be · if you had said governments will set
“used to” used to + क्रिया — बीती आदत या स्थिति जो अब नहीं है	used to run · used to be
Present perfect have/has + past participle — बीता हुआ काम जिसका असर अभी भी है	have you been · I've created · I've known

ध्यान देने योग्य उच्चारण

वक्ता 226 बार संक्षिप्त रूपों का प्रयोग करता है, जैसे we're, you're, don't। इन्हें वैसे ही छोटा बोलें जैसा आप सुनते हैं।

“th” की ध्वनियाँ: anthropic /ænˈθɹɒp.ɪk/, altogether /ɔl.tuˈɡɛð.ɚ/
“sh” और “zh” की ध्वनियाँ: ambitious /æmˈbɪʃ.əs/, gravitational /ˌɡɹævɪˈteɪʃənəl/, iteration /ˌɪt.əˈɹeɪ.ʃən/, optimization /ˌɑptəmaɪˈzeɪʃən/, accomplish /əˈkɑm.plɪʃ/
लंबे शब्द — बलाघात सही रखें: pragmatism /ˈpɹæɡmətɪzəm/, analogy /əˈnæləd͡ʒi/, conservatism /kənˈsɜːvətɪzəm/, gravitational /ˌɡɹævɪˈteɪʃənəl/, iteration /ˌɪt.əˈɹeɪ.ʃən/

हिंदी भाषियों के लिए मुश्किल ध्वनियाँ:

/v/ और /w/ का फ़र्क़ — /v/ में ऊपर के दाँत निचले होंठ को छूते हैं, /w/ में होंठ गोल होते हैं: eval /ˈiː.væl/, incentive /ɪnˈsɛntɪv/, archive /ˈɑɹˌkaɪv/, conservatism /kənˈsɜːvətɪzəm/, gravitational /ˌɡɹævɪˈteɪʃənəl/
/z/ — इसे “ज” (/dʒ/) न बनाएँ: pragmatism /ˈpɹæɡmətɪzəm/, conservatism /kənˈsɜːvətɪzəm/, prioritize /pɹaɪˈɒɹ.ə.taɪz/, optimization /ˌɑptəmaɪˈzeɪʃən/
शब्द की शुरुआत में s + व्यंजन — पहले “इ” न जोड़ें: stripe /stɹaɪp/, skeptical /ˈskɛptɪkəl/, startup /ˈstɑɹtˌʌp/

इस वीडियो से अभ्यास कैसे करें

पहले बिना बोले पूरा वीडियो एक बार सुनें और जो शब्द नहीं आते उन्हें लिख लें।
0.75× गति से शुरू करें, एक-एक वाक्य की शैडोइंग करें, और सहज लगने पर सामान्य गति पर लौटें।
अपनी आवाज़ रिकॉर्ड करें और मूल से तुलना करें, ख़ासकर neural, pragmatism, archive जैसे शब्दों पर ध्यान दें।

शैडोइंग तकनीक क्या है?

शैडोइंग (Shadowing) एक विज्ञान-समर्थित भाषा सीखने की तकनीक है जो मूल रूप से पेशेवर दुभाषिया प्रशिक्षण के लिए विकसित की गई थी। विधि सरल लेकिन शक्तिशाली है: आप मूल अंग्रेज़ी ऑडियो सुनते हैं और तुरंत इसे ज़ोर से दोहराते हैं — जैसे वक्ता की छाया 1-2 सेकंड की देरी से। शोध से पता चलता है कि यह उच्चारण सटीकता, स्वर, लय, जुड़ी हुई ध्वनियाँ, सुनने की समझ और बोलने की प्रवाहशीलता में काफ़ी सुधार करता है।

शैडोइंग तकनीक: पूरी चरण-दर-चरण गाइड पढ़ें →

शैडोइंग अभ्यास: Building Anthropic | A conversation with our co-founders - वीडियो के साथ अंग्रेजी बोलना सीखें

Translating Claude’s thoughts into language

When AIs act emotional

Introducing Claude Fable 5

The different levels of how Claude thinks

Lesson 3: AI as a learning partner | AI Fluency for students

Understand AI in 14 minutes – with Anthropic's Chloe Lubinski [ARC 2026]

OpenAI’s lead engineer was secretly a spy...

Trust the Process: How to Design Teams That Actually Work

Cleo Abram Answers The Web's Most Searched Questions | WIRED

We Let $200 GPT-Pro Think for 7 Days on Unsolved Math

Creator of C# and TypeScript: "AI will NEVER Replace Coders, Here's Why" | Anders Hejlsberg

GS.TS. Po-Shen Loh, ĐH Carnegie Mellon | Rèn tư duy ngược để không bị AI bào mòn | TQKS #129

AI & Inequality | Daron Acemoglu: Big Tech poses risks as AI reshapes society

How to Become an AI Product Manager in 2026 | Ex-Google, Microsoft

Jensen Huang: The Mindset That Built NVIDIA

Most devs don't understand what agents are

Did Google just kickstart the intelligence explosion?

इस पाठ की शब्दावली और बोलने के सुझाव

इस वीडियो की मुख्य शब्दावली

वीडियो में आने वाली फ़्रेज़ल वर्ब

दोहराने लायक वाक्य

इस वीडियो का व्याकरण

ध्यान देने योग्य उच्चारण

इस वीडियो से अभ्यास कैसे करें

शैडोइंग तकनीक क्या है?