Практика Shadowing: What REALLY Happens in Meta’s AI Product Sense Round - Изучайте разговорный английский по видео

Выберите режим практики

Каждый режим открывает полноэкранную комнату практики.

218 предложений

Если предложения слишком короткие или длинные, нажмите Edit, чтобы их изменить.

You've done interviews like this before.

What's different about this?

They have you use like Llama to actually vibe code a product in the AI product sense interview.

So like my first loop was a normal traditional product sense, a normal analytical.

And then there was an AI product sense after that, that was specific to using, they would give you like a normal product sense case,

and then you walk through it, use cases, segmentation solution.

And then they stop you at the solution

and tell you to move over to the meta vibe coding board and start sort of prototyping that solution.

And so I thought that was really unique.

And that was, at that time, that was really out of the ordinary for me.

And I had already practiced, like, you know, with Kershaw Cloud Code, I use a lot of those tools daily anyways.

But I'd never been asked that in an interview before.

And now it's definitely becoming kind of the norm, which is fun and cool, but also comes with its own challenges, you know.

In the AI product sense round, just start from the beginning and just kind of walk me through play-by-play yeah

so i joined the call um i was interviewing with another senior product manager

that worked on the team um and he said

that he's going to give me a prompt not really it's

going to sound like a normal traditional product sense prompt he wants me to go as far as the solution

and then stop

and use 30 minutes to flush out sort of my use cases my clarifying questions my user groups

and my um and which ones i prioritize for solution stop at the 30 minute mark

and then go over to uh he sent me a link to like a llama choppa chatbot um

and he wanted me to vibe code whatever solution i had in mind at

that point so it was like 30 minutes like sort of traditional product sense kind of

and then 30 minute prototyping what was the actual prompt prompt

was you're a product manager at meta you've been put in

charge of a brand new product for volunteering what would you do

and why so from there i kind of started with clarifying questions

I asked if there was any time or resource constraints when I think about solutioning.

He said, no. I said, will it be built as part of one of Meta's ecosystem of products or standalone?

He said, you decide.

And then I said, am I both mobile or web?

And he said, your decision.

We've built a database that houses all of the top companies

and their interview processes from anonymous reports so

that you can learn what really happens in the room in the interviews at Anthropic, OpenAI, and Fang+.

Check out our full library of interview experiences in the link below.

Submit your own whenever you have one of these

and pay it forward so that you can reap the same benefits from what you're getting from the community.

So the only real constraints were voluntary volunteers and unlimited time and resources, which is not really a constraint,

but kind of an unlimited constraint and everything else you decide.

And so they decide to have you use their LLAMA tool.

They bring you into their tool.

You use that tool.

What do you remember in your notes from how the interviewer reacted to how you used AI?

Specifically, were there moments where they like pushed back on your prompts or outputs?

So I went through goals.

I went through user groups.

I went through pain points and then solutions.

And so once I went over to the prompt, it was very like, I won't say cursor-esque because I think the cursor UX is much better.

I would even say the quad code UX is much better.

But it was very much like ask a bunch of questions.

I put in my whole prompt notes, like my assumptions, my pain points, and my user group, as well as my solution.

Once I put that in, it went to work for like, I want to say five to seven minutes.

And then from there, I started making iterations.

And he asked me questions around latency.

He also asked me questions around why I wouldn't leverage image generation.

And then he also asked me about what did I think about interference, compute, as well as retrieval.

So that was all those questions sort of came as follow ups after I had that first prototype up.

And I clicked through it and kind of showed him what I was thinking and what I wanted to change.

And he was asking me like, oh, is this the most efficient way to prompt it?

Like, aren't you using more tokens versus prompting it for this?

Like maybe, maybe like for a good example was I was trying to get to generate a chart of things

that the users apply to.

And he was asking me if I should let the LLM determine what chart to use.

And so I was prompting for that.

And he's like, oh, don't you think that'll eat more tokens?

Like maybe there'll be less latency if you use this type of chart.

And so there was follow up questions like that.

But it honestly kind of felt to me, even when I was interviewing with him, that it was his first time doing this as well, because there's a lot of times that I was typing things in,

or I was like, about to hit send.

And he was like, oh, let's rethink this.

And I was like, oh, okay, let's talk about it.

That's something that we're hearing.

And even like all across the rounds nowadays, all these Vibe Coding rounds are so new.

There's no question banks that people can rely on where they can just like, you know, so I totally hear you.

There's a real thing.

People are saying that.

I want to go back to these follow-up questions are super, super cool.

Did he ask you another question about latency besides the prompting it in a different way to use less tokens?

He asked me once I finished like all my follow-up prompts and like towards the end of the overall thing, he went back to latency again and was like,

what's a better way you could have maybe built this prototype with the consideration of CompuPower, token optimization, and latency in general.

What about your friend?

Did he give you any insights into maybe the question that he got asked?

Did he get the exact same prompt about volunteering?

His prompt was about if Meta wanted to create an app, a map app, how would you go about creating it?

And he also had 30 minutes to kind of do the traditional sense

and then bring it over to a live coding session

and he was asked mainly about retrieval like where is he getting all this data from

because they don't have a map api right now how would

you get all the various geographic data how would you get

coordinate how would you maybe built in like directional uh like features within the app within the map

and so his was a little different

and honestly it sounded like a lot of his follow-ups were

specifically on the actual product versus mine being very much about like

the LLM tooling and the tool calling and token optimization and latency.

Like, he didn't get any questions about that.

His was all based on, like, retrieval and, like, how he would build to call different APIs to retrieve data in a more efficient manner.

And yours asked about retrieval, but maybe only, like, once.

100

Yeah, dude.

101

Just, like, where would we retrieve all the—first of all, if we're pulling data from—for all these volunteering opportunities,

102

what's the best way to optimize the LLM for retrieval specifically on those?

103

And then secondly, if the user applies, what's the best way to actually know that we're sending them down the correct path?

104

What would you do differently if you were going to prepare for the round again?

105

I would probably, one, brush up really well on my technicals when it comes to just vibe coding in general.

106

Like optimizing for specific clarifying questions up front.

107

Like even the meta bot, like it didn't really like you know how like cursor

108

and cloud code

109

if you used it will ask you like a like a

110

multiple choice question almost like it was more an open-ended question

111

that meta threw back at me like hey what should we do in this specific edge case

112

or what should we do in this scenario

113

and then i had to type it out you think about it versus like take one of their pre-camp answers

114

so i probably would have one like relied on myself for

115

more product thought process than thinking the llm would work with me

116

and then two like honestly from from a technical perspective like

117

i would have brushed up more on how to actually optimize for token usage latency retrieval like

118

that was all at the time like very foreign to me

119

totally hear you one small question i'm kind of using an assumption here

120

because they just added uh ai assisted coding rounds for software engineers at meta

121

and one of the things they let them do is they let them use the exact tool

122

that they're going to use in the interview in their practice i'm not surprised

123

that they're doing

124

that they seem very ai first i've talked to dozens of

125

fang interviewers over the last couple years meta is just in a different freaking league absolutely

126

so i'm gonna guess

127

that they probably would do well to have the pms use the tool beforehand

128

so even if they're not doing that i bet they're gonna do that

129

because that's how it is with the with the software engineering

130

one that's actually a good point that's one thing i probably would have changed as well

131

because they did send me like a demo of the tool beforehand.

132

I think it was like a quick time demo of somebody using the tool and how it works.

133

And it looked very just like any other blog coding.

134

Yeah, it kind of looked a little bit like V0 that Versal makes.

135

And so I practiced a lot on V0 and maybe a little bit of lovable as well.

136

It seemed like it was gated internally besides the one instance they gave me while I was on the call.

137

Open-ended question that no one has a solution to.

138

I want to hear your opinions.

139

When that thing's loading for five to seven minutes, what are you trying to do to add value oh yeah that's a good question

140

when i was going back uh well while i was sort of loading

141

and generating my prototype i went back to my notes

142

and just like i had a lot of notes and things

143

that i would that i kind of called out of scope

144

and i was thinking of ways to include that because i thought

145

that if i had a better prototype that included maybe some additional features

146

that wouldn't have been mvp

147

but would be built into the prototype it would make me

148

look like a well more well-rounded pm that's trying to shoot for the north star um

149

but maybe iterate towards it and And so I was going back and thinking of additional features that I can add.

150

And so I was thinking of ways to kind of increase my scope while that was spinning.

151

And then he was also following up

152

and asking me like what types of things I would include in the user creation onboarding profile.

153

But those were like honestly more product sense questions versus like AI specific questions.

154

I think all the AI specific questions around latency and compute came after the first generation.

155

It was like, okay, if somebody's joining via Facebook or connecting via Instagram and we don't have much data on them, let's say it's like, for lack of a better term,

156

a burner account, how would we even prompt them to fill in some of the required fields

157

that we want versus just sending them directly to the homepage of the app?

158

How do we get more interest or more user input so

159

that we can actually fix the listings algorithm or the listings display to show them like opportunities they would like?

160

And we can't do that when they don't when we don't have the data.

161

So what ways would you incentivize a user to get meta additional data

162

so that we can leverage that to show them a better listing?

163

So cool um another one that i hear about people talking about in

164

while it's loading is like one of the common questions people are asking is like hey

165

if this were actually production app what would you do differently

166

so what do you think of that tip of saying like proactively answering

167

that and saying like just in case they're gonna ask

168

that like hey by the way

169

if i were gonna make this a production level app here's

170

what i would do differently that's i no i didn't i didn't actually consider

171

that i probably should have um at the time i wasn't

172

really i was thinking like we were really doing just a quick

173

and dirty exercise and so i didn't think of really like scalability

174

but in today's day it's obviously like the biggest problem you

175

see like you know amazon fired a bunch of people

176

and then had four step ones or five step ones in prod

177

and it's bigish scalability in general um and

178

so my take on it nowadays is like okay this is

179

very much just a clickable prototype with a lot of functionality

180

that you can see what vision I want

181

and maybe I'll even MCP to Figma to get in my design system

182

and components so that you can see exactly what I'm trying to build towards

183

but you still need to go in

184

and integrate it into the back-end system you need to integrate into the front-end system you need to make sure

185

that the code is actually like functional and it's written in a way

186

that it's not going to create additional tech debt

187

because I think that's a big problem even in my current

188

role today is like the product team will buy code a lot of things

189

and hand it to engineering and engineering will be like, I need two weeks with this because I have to fix a bunch of functions

190

or a bunch of technical things in the code

191

that was written very poorly by AI to just get you this initial prototype.

192

I love those, those two concrete examples of like the real problems people are experiencing on the job

193

and then connecting that to the interview.

194

I think that's a really, really smart tip.

195

I guess my best advice and kind of my reflection on that is like using those tools as much as possible.

196

So, you know, the best things to prompt is a big one.

197

And the second one is also like what data source you're using really goes a long way.

198

Like you can tell it's a data and it'll mock some things up, but it won't really be functional on the prototype unless you have it point somewhere.

199

So what I've started to do in my own time is like, there's like a side furniture project I'm working on.

200

I've had to start pointing to like catalogs in Ikea

201

or sort of crate and barrel or somewhere else to actually pull in real data.

202

Even though it's and it's just from the web, it's at least pulling in like real catalog things that are, that, that exist in those domains.

203

And so that's something that I probably would have done differently.

204

Like I probably would have gone to one of the volunteering boards and pulled in opportunities from there.

205

And I think if I could have somehow like prompt, like even wrote a couple of things about myself before starting it and being like, I want you to use me as sort of the test user, it might've been, it might've flown better.

206

And then, oh, one additional thing that I just, I've totally forgotten.

207

And you kind of asked how I would build that into a mobile space.

208

And if I would add that as part of the current Facebook app or standalone app, I thought that was really interesting.

209

Part of my mind went standalone

210

because it might be easier to implement in a brand new code base and a brand new repo standalone.

211

But at the same time, I think a lot of people already use Instagram, Facebook.

212

There's already a lot of daily active users, monthly active.

213

And so it might make more sense to get a wider top of funnel by using those platforms first.

214

Looking back, I would be more prepared to talk about maybe a mobile standalone product.

215

Even if I did inspect and like show to me and, you know, mobile responsive, it's not the same as a mobile app with the various components and things.

216

And so I thought that was a really interesting question that he threw at me at the end.

217

It's just so cool.

218

You did an amazing job.

Скачать приложение

Всё для свободной речи

ИИ-оценка произношенияОценка каждой фразы

Практика IPAОсвойте каждый звук

СловарьСоберите свой словарный запас

Игра со словамиУчитесь играя

Сканировать QR

Download on theApp Store Get it onGoogle Play

Тот же канал

FOR YOU

Лексика и советы по произношению к этому уроку

В этом видео 218 предложений и 2956 слов для шедоуинга. Речь длится 13:36. Говорящий говорит быстро, около 217 слов в минуту, поэтому звуки часто сливаются и сокращаются. 84% слов входят в 3000 самых частых слов английского языка; остальные стоит разобрать перед занятием.

Ключевая лексика этого видео

15 слов из видео, которые стоит выучить, с произношением и значением:

Слово	Произношение	Значение
prompt глагол	/pɹɑmpt/	побужда́ть, внуша́ть
prototype существительное	/ˈpɹəʊtətaɪp/	прототи́п, моде́ль
token существительное	/ˈtoʊkən/	знак, си́мвол
versus	/ˈvɝsəs/	про́тив
volunteer существительное	/ˌvɑlənˈtɪɹ/	доброво́лец, доброво́лица
vibe существительное	/vaɪb/	атмосфе́ра, а́ура
differently наречие	/ˈdɪf.ə.ɹənt.li/	по-друго́му
chart существительное	/t͡ʃɑɹt/	морска́я ка́рта
clarify глагол	/ˈklæɹ.ɪ.faɪ/	очища́ть, чи́стить
cloud существительное	/ˈklaʊ̯d/	о́блако, ту́ча
efficient прилагательное	/ɪˈfɪʃənt/	эффекти́вный
connect глагол	/kəˈnɛkt/	свя́зывать, связа́ть
component существительное	/kəmˈpoʊ.nənt/	компоне́нт, составна́я часть
rely глагол	/ɹɪˈlaɪ/	полага́ться, положи́ться
scope существительное	/skəʊp/	масшта́б, преде́л

Фразовые глаголы, которые вы услышите

Слово	Значение
brush up глагол	освежа́ть, освежи́ть
go back глагол	возвраща́ться, возврати́ться
check out глагол	проверя́ть, прове́рить
look back глагол	огля́дываться, огляну́ться

Произношение, на которое стоит обратить внимание

Говорящий использует 51 сокращённых и редуцированных форм, например didn't, don't, I'm. Произносите их коротко, так, как слышите.

Звуки «th»: algorithm /ˈælɡəɹɪðm̩/, rethink /ɹiːˈθɪŋk/
Звуки «sh» и «zh»: efficient /ɪˈfɪʃənt/, brush /bɹʌʃ/, assumption /əˈsʌm(p).ʃ(ə)n/, optimization /ˌɑptəmaɪˈzeɪʃən/, consideration /kənˌsɪd.əˈɹeɪ.ʃən/
Длинные слова — следите за ударением: differently /ˈdɪf.ə.ɹənt.li/, unlimited /ʌnˈlɪm.ɪ.tɪd/, optimization /ˌɑptəmaɪˈzeɪʃən/, ordinary /ˈɔɹdɪnɛɹi/, consideration /kənˌsɪd.əˈɹeɪ.ʃən/

Как заниматься с этим видео

Прослушайте всё видео один раз молча и выпишите незнакомые слова.
Начните со скорости 0,75×, повторяйте предложение за предложением и вернитесь к обычной скорости, когда станет легко.
Запишите себя и сравните с оригиналом, обращая внимание на такие слова, как prompt, prototype, token.

Грамматика в этом видео

Конструкции, которые говорящий использует чаще всего, с точными словами из видео:

Конструкция	В видео
Условные предложения if + предложение, will/would + глагол — условие и его результат	if you used it will ask
Present Perfect have/has + причастие прошедшего времени — прошлое действие, важное сейчас	You've done · you've been put · We've built
Пассивный залог be + причастие прошедшего времени — важно, что происходит, а не кто это делает	been asked · you've been put · be built
Придаточные определительные who / which + предложение — уточнение о человеке или предмете	norm, which is · resources, which is

Что такое техника Shadowing?

Shadowing — это научно обоснованная техника изучения языка, изначально разработанная для подготовки профессиональных переводчиков и популяризированная полиглотом доктором Александром Аргуэльесом. Метод прост, но эффективен: вы слушаете аудио на английском от носителей языка и немедленно повторяете вслух — как тень, следующая за говорящим с задержкой в 1–2 секунды. В отличие от пассивного прослушивания или грамматических упражнений, Shadowing заставляет мозг и мышцы рта одновременно обрабатывать и воспроизводить реальные речевые паттерны. Исследования показывают, что это значительно улучшает точность произношения, интонацию, ритм, связную речь, понимание на слух и беглость речи — что делает его одним из самых эффективных методов для подготовки к IELTS Speaking и реального общения на английском.

Техника шедоуинга: читать полное пошаговое руководство →

Практика Shadowing: What REALLY Happens in Meta’s AI Product Sense Round - Изучайте разговорный английский по видео

Выберите режим практики

Скачать приложение

Тот же канал

Uber Data Engineering Mock Interview - Ride-Sharing Data Warehouse Schema

What is an API?

System Design Mock Interview: Design Facebook Messenger

How to Answer System Design Interview Questions (Complete Guide)

System Design Mock Interview: Design WhatsApp

Senior Product Designer Mock Interview: App Critique of Yelp

Understanding TCP vs UDP

FULL Data Science Mock Interview | SQL, Statistics, Business

Anthropic Culture Fit | Behavioral Interview

Рекомендуемое видео

[2026] TOEFL iBT Speaking Take an Interview | Step-by-Step Strategies #0008

Excellent 2 Years AWS DevOps Engineer Mock Interview with Questions and Answers | Cloud | Docker

Learn how to SOLVE a data analytics case study problem

How Great Software Engineers Are Moving Faster Than Ever

Distributed Task Scheduler: System Design Interview (Stripe & Amazon Offers)

Google product manager behavioral interview (Top 5 questions)

The Skills You’ll Need to be a Senior Software Developer

API Gateways in System Design Interviews w/ Ex-Meta Staff Engineer

Mastering iOS Development: Mock Interview with Ramiro Neto

Anthropic's CCA Exam as a Field-Guide for Agentic Engineering — Frank Coyle, UC Berkeley