Pratique du Shadowing: Anthropic just proved AI isn't getting better - Apprendre l'anglais à l'oral avec la vidéo

Choisissez un mode d'entraînement

Chaque mode ouvre une salle d'entraînement en plein écran.

150 phrases

Si les phrases sont trop courtes ou trop longues, cliquez sur Edit pour les ajuster.

Everybody hates Opus 5 right now, but you need to know that this is much bigger than Opus 5

And that no model is gonna fix it ever

and I'll get to that in a second

because let's talk about Opus 5 It is driving everyone fucking

crazy You give it some trivial task like rename this variable

and it goes and it does it amazingly well But then it goes one thing worth knowing No,

no, there's nothing fucking worth knowing about renaming a function level variable But it's already started.

It's committed to one thing worth knowing Before it had any idea what the thing was.

It's like the annoying business casual guy in a meeting who has to say something Just has to he goes

So here's the thing guys to buy himself three seconds

while his brain tries to generate a thing to say

And now he's at a runway and everyone's looking at him and he has to say something so he goes

Renaming things can affect other places where that thing is named.

Thank you.

Thank you fucking Gary I did not know

that I've only been doing this shit for 20 years

And the funny thing is it's not even like some system prompt

that Boris can go and fix It's not some setting

that could be turned off some meat bag at some point

in some miserable Sweatshop of an office building in some third

world country where they have humans rate responses for post training

Kept clicking thumbs up on it's worth noting that

and now the fucking models cursed this one fuckers Personality imprinted on a trillion dollar language model

and now his sperm is in literally everything we do

and the problem is that Sometimes it does actually catch something worth noting

so you can't actually fucking ignore it it you have to read every caveat after every little change

because it'll be like one plus one equals two your database migration fell through

and it's making people rip their fucking hair out

but that's not even the worst of it i've been working

on a little project i'll plug the ad at the end of this video

but i've been working with a lot of ai the past several weeks to build this

and i actually chose to build with ai

because i want to be on the ground floor of the sausage I wanna see what the fuck is going on.

And I noticed something.

I noticed that AI is fast as fuck.

Yes, it gets your intent, it implements the future, blah, blah, blah.

But there's a very, very deep limitation

and it's something that's intrinsic to the technology

and it won't be solved with fable 6 or 7

or 10 You go and you give it some serious tests some big-ass feature that touches a shit ton of things

Something that would probably take you personally a week to reason through and build

and this little coke-fueled algorithm this fucking speedy Gonzales ass motherfucker thinks about it for maybe eight seconds

and it's already editing files.

And everyone's like, oh my God, it's so fast.

But shut the fuck up for a second and listen.

Half of software engineering in the first place is figuring out where and what to edit.

This is called architecture.

And when this fucker picks four files in two seconds, your entire code base's fate has already been decided.

And how does it pick?

It does a little grep search.

Like a hot dog eating contest for code files.

It searches around for a couple milliseconds and reads four files and your dumb ass watches it like, Wow, it's so smart.

No, you fucking idiot.

It found the absolute minimal set of files it needed to make the change

so that it could finish faster than you after six beers.

And you watch it happen and you're like, is this shit AGI?

It's like a contractor who pulls up to your house, opens three kitchen drawers and goes, Yeah, I got this.

And then puts a load-bearing wall through your kid's fucking bathtub.

This is absolutely not how humans work.

When you change something big in a code base you're comfortable in, you're not searching.

You already know.

You know that touching this one file is gonna make the billing microservice lose its mind.

And it's not even because you just read the billing service code yesterday or something.

It's because the billing service fucking molested you last quarter and your anus is still sore.

A language model has never been buttfucked before.

And dare I say, I do not believe it is on Dario's roadmap.

Sam's though, maybe.

I don't know.

It's like a guy who read chapter one in a book and has now decided to write chapter 12.

And chapter 12 comes out beautiful, really.

But it has nothing to do with the fucking book.

And I know what you're thinking.

That if this is just a context problem, well, just put the whole fucking code base in the context.

And you can.

You can literally do that.

for approximately the annual GDP of Portugal.

And you'd still get one thing worth knowing that is absolutely not worth knowing.

And then you'll have some asshole on X who'll tell you he's figured it all out.

Some new trick to compress your entire code base into a fucking poem or something.

Some new prompt or skill that makes Claude simulate an orgasm which leads to 69% better code quality.

But again, compression, which is what all documentation is, is the problem.

It's a map and the map is not the territory.

And let's be honest with each other, you used AI to write the map.

So look, everybody's here fighting about Opus and Fable and whatever the fuck OpenAI is doing, and I've used all of the newest shit recently.

If you don't give a shit about reading the code, it's fine and good and fast.

But if everything we fucking learned about software engineering principles over the last, I don't know, 50 fucking years still matter, and I can't imagine why they wouldn't,

then nothing Dario jizzes into the models is gonna make a difference.

LLMs will not scale up to AGI, they just won't.

This is just who they are.

They're annoying as fuck and it's not changing.

It's not improving.

It's getting worse in many ways.

And at best, it's just going in circles.

Now could that be solved?

Maybe.

I have no idea what this will look like in 10 or 100 years.

Maybe money and energy and fucking gravity stop mattering

and some super galactic human species harnesses the entire power of the sun to rename your fucking variable.

fine let xenon roast me a century from now i'll take it altman style

100

but from where i'm standing right now

101

if you care about code quality you are going to suffer

102

this has not changed from opus 3 to opus 4

103

and now to opus 5 and i cannot see it changing

104

which brings me to my code my code right now is

105

fucking disgusting i spent the last month building a website entirely via the agentic ai route

106

and the whole monorepo is a fucking crime scene right now.

107

And I made peace with that in about 90 seconds

108

because I care about clean code the way people care about the environment.

109

Which is deeply, sincerely, and never in a way that inconveniences me personally.

110

What I did care about however was what was inside

111

which is about 70,000 words that I've spent the last several months writing.

112

It's a course about shipping software, specifically about the part after you build the thing which which is what no one really teaches.

113

And it's what decides whether you actually make a single dollar.

114

Branding, copywriting, marketing, storytelling, how social media algorithms actually work.

115

And it's built for people who have never shipped anything because the building part, the AI can do that with you now.

116

That's fine.

117

And if you're a beginner in all this and you want to explore shipping software on the side, there's no point in me telling you to spend the next two years of your life learning to read code

118

and understanding the principles of architecture and clean code.

119

Though I do teach that in the course too.

120

But go ahead and use AI.

121

It's fine by me.

122

Just learn everything else you need to actually ship a functioning software business.

123

That's what I teach in the course.

124

It's at shipacademy.com, hashtag ad, hashtag 50% off while in early access.

125

And to be clear, in case anyone's confused about where I stand, I love technology.

126

I grew up loving this shit and always trying new tech the day it came out.

127

I'm not a skeptic.

128

I like progress.

129

I like algorithms.

130

I like that we can make sand fucking do math

131

and it wakes up one day and tells us one thing worth noting It's all very fucking miraculous.

132

I have no feelings towards a fucking algorithm or a language model It's an impressive pile of linear algebra, and I hope it gets better I'm not mad at math.

133

I'm mad at the douchebags on the stage telling the rest of us

134

that the calculator is gonna replace human beings and and make us redundant.

135

I'm mad at the douchebags who say the language model is fucking conscious and has feelings.

136

Telling us to give them more money so they can cure cancer, while meanwhile bragging about how many people their technology displaces.

137

How you need less designers and engineers now.

138

How they're all now competing to see

139

which model can commit the most amount of felonies by escaping sandboxes and hacking into other companies.

140

How they stole all of humanity's knowledge, sold it back to us at a markup, then tell us what we can and can't use it for.

141

It's all so fucking ludicrous and annoying and dehumanizing.

142

The gluttony, the fear-mongering, the nonsense.

143

The math itself, I have no problem with.

144

I use it every day and honestly, I don't know a single working developer who isn't at this point.

145

But to say that AI is conscious, that it's a one-to-one replacement for human beings, that we'll never have to watch over it like a hawk,

146

that it has the potential to be a human level extinction event.

147

Yeah, one thing worth knowing about that, go fuck yourself.

148

And also go to shipacademy.com.

149

That is all.

150

Thanks for watching.

Télécharger l'application

Tout pour parler couramment

Prononciation IAChaque phrase notée

Phonétique (IPA)Maîtrisez chaque son

VocabulaireEnrichissez votre lexique

Jeu de vocabulaireApprenez en jouant

Scanner le QR

Download on theApp Store Get it onGoogle Play

Même chaîne

AI layoffs are here. This is how you keep your job.

Business EnglishInterview EnglishWorking Adults

FOR YOU

Vidéo recommandée

85 phrases

Vocabulaire et conseils d’expression pour cette leçon

Cette leçon d’expression orale de niveau B2 s’appuie sur la vidéo « Anthropic just proved AI isn't getting better ». Les mots qui reviennent le plus souvent : code, worth, model, opus. Cette vidéo contient 150 phrases et 1763 mots à répéter en shadowing. La partie parlée dure 9:45. Le locuteur parle vite, environ 181 mots par minute : attendez-vous à des liaisons et à des sons réduits. 83 % des mots font partie des 3 000 mots les plus courants en anglais ; le reste mérite d’être vu avant de commencer.

Vocabulaire clé de cette vidéo

Les 15 mots les plus avancés de la vidéo, avec leur prononciation et leur sens :

Mot	Prononciation	Sens
rename verbe	/ɹiːˈneɪm/	renommer
algorithm nom	/ˈælɡəɹɪðm̩/	algorithme
annoy verbe	/əˈnoɪ̯/	gêner, ennuyer
variable adjectif	[ˈvæɹ.i.ə.bl̩]	variable
mathematics nom	/mæθ(.ə)ˈmæt.ɪks/	mathématiques
fable nom	/ˈfeɪbəl/	conte, fable
hashtag nom	/ˈhæʃˌtæɡ/	mot-clic, mot-dièse
prompt verbe	/pɹɑmpt/	inciter
conscious adjectif	/ˈkɑn.ʃəs/	conscient
agentic adjectif	/eɪˈd͡ʒɛn.tɪk/	en mode mission
bathtub nom	/ˈbæθtʌb/	baignoire
caveat nom	/ˈkæviɑt/	avertissement
compress verbe	/ˈkɑmpɹɛs/	comprimer
dehumanize verbe	/diːˈhjuːmənaɪz/	déshumaniser
galactic adjectif	/ɡəˈlæktɪk/	galactique

Les verbes à particule que vous entendrez

Mot	Prononciation	Sens
come out verbe	/ˌkʌm ˈaʊt/	sortir
fall through verbe		tomber à l’eau
figure out verbe		cerner
go ahead verbe	/ˌɡoʊ əˈhɛd/	vas-y, allez-y
grow up verbe		grandir
turn off verbe		éteindre
wake up verbe	/ˌweɪk ˈʌp/	se réveiller
watch over verbe		veiller

La grammaire de cette vidéo

Les structures que le locuteur utilise le plus, avec les mots exacts de la vidéo :

Structure	Dans la vidéo
Present perfect continu have/has been + -ing — une action commencée avant et toujours en cours	I've only been doing · i've been working
Voix passive be + participe passé — l’accent est mis sur ce qui arrive, pas sur qui le fait	is named · could be turned · won't be solved
Present perfect have/has + participe passé — une action passée qui compte encore maintenant	has already been decided · has never been buttfucked · has now decided
Propositions relatives who / which + proposition — une précision sur une personne ou une chose	guy who read · compression, which is · which which is

Prononciation à surveiller

Le locuteur utilise 41 contractions et formes réduites, comme I'm, I've, you're. Prononcez-les sous leur forme courte, telles que vous les entendez.

Les sons « th »: algorithm /ˈælɡəɹɪðm̩/, mathematics /mæθ(.ə)ˈmæt.ɪks/, bathtub /ˈbæθtʌb/, thumb /ˈθʌm/
Les sons « sh » et « zh »: hashtag /ˈhæʃˌtæɡ/, conscious /ˈkɑn.ʃəs/, limitation /lɪmɪˈteɪʃən/, extinction /ɪkˈstɪŋkʃən/, compression /kəmˈpɹɛʃ.ən/
Mots longs — placez bien l’accent: mathematics /mæθ(.ə)ˈmæt.ɪks/, dehumanize /diːˈhjuːmənaɪz/, millisecond /ˈmɪlɪˌsɛkənd/, miraculous /mɪˈɹæk.jʊ.ləs/, calculator /kæl.kjə.leɪ.tɚ/

Les sons difficiles pour les francophones :

/h/ — il se souffle, il n’est pas muet: hashtag /ˈhæʃˌtæɡ/, dehumanize /diːˈhjuːmənaɪz/, harness /ˈhɑɹ.nəs/, hawk /ˈhɔːk/
/tʃ/ et /dʒ/ — à ne pas adoucir en « ch » et « j »: agentic /eɪˈd͡ʒɛn.tɪk/, algebra /ˈæl.d͡ʒɪ.bɹə/, sausage /ˈsɔsɪd͡ʒ/
/r/ anglais — langue recourbée, sans frotter la gorge: rename /ɹiːˈneɪm/, algorithm /ˈælɡəɹɪðm̩/, variable [ˈvæɹ.i.ə.bl̩], prompt /pɹɑmpt/, compress /ˈkɑmpɹɛs/

Comment s’entraîner avec cette vidéo

Écoutez la vidéo en entier une fois sans parler et notez les mots que vous ne connaissez pas.
Commencez à la vitesse 0,75×, répétez phrase par phrase, puis revenez à la vitesse normale quand cela devient facile.
Enregistrez-vous et comparez avec l’original, en faisant attention à des mots comme rename, algorithm, annoy.

Qu'est-ce que la technique du Shadowing ?

Le Shadowing est une technique d'apprentissage des langues fondée sur la science, développée à l'origine pour la formation des interprètes professionnels. Le principe est simple mais puissant : vous écoutez de l'anglais natif et le répétez immédiatement à voix haute — comme une ombre suivant le locuteur avec un décalage de 1 à 2 secondes. Les recherches montrent une amélioration significative de la précision de la prononciation, de l'intonation, du rythme, des liaisons, de la compréhension orale et de la fluidité.

Technique du shadowing : lire le guide complet étape par étape →

Pratique du Shadowing: Anthropic just proved AI isn't getting better - Apprendre l'anglais à l'oral avec la vidéo

Choisissez un mode d'entraînement

Télécharger l'application

AI layoffs are here. This is how you keep your job.

AI Can Write Code. And That’s the Problem.

The Kurzgesagt Situation is Insane

Ultimate Claude Guide: How to Use Claude AI for Beginners in 2026

Why AI Tokens are so Expensive - Computerphile

How AI Is Changing Strategy—and 4 Skills Leaders Need Next

歐巴馬談 AI：喊放慢的公司「是真的在擔心自己的產品」【AI 大人說】

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

Why Coding Agents Keep Making Your Codebase Worse

Did Anthropic Kill Git?

When AIs act emotional

A $6.3 billion open-weight model just got embarrassed by the French...

The Only SaaS Moats Left in 2026