Pratique du Shadowing: China just dropped a new frontier model... and nobody can explain how - Apprendre l'anglais à l'oral avec la vidéo

Choisissez un mode d'entraînement

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

82 phrases
Si les phrases sont trop courtes ou trop longues, cliquez sur Edit pour les ajuster.
1
Last week, an anonymous mystery model named Kibitz showed up randomly on Chatbot Arena, performing at the frontier level.
2
Kibitz is the Yiddish word for that guy standing behind you at a chess game, giving you tips on how to blunder your queen,
3
which to be honest is the perfect analogy for a coding agent.
4
But it was pretty obvious who was behind this mystery model from the beginning, because the first two letters of this word are a dead giveaway.
5
Everybody knew it was Kimmy, then right on cue last Friday, Moonshot dropped Kimmy K3 .1,
6
officially giving us a new 3 trillion parameter open weight model that will be fully downloadable on November 4th, or so they say.
7
And just when you thought the AI bubble was popping and the Trust Me Bro benchmarks could not possibly get any better, Kimmy made them better.
8
I can go lower.
9
In today's video, we'll find out why Kimmy K3 .1's performance should be impossible, and put it to the test on the only benchmark that matters,
10
building 3D horses with 3JS.
11
It is October 26, 2026, and you're watching the Code Report.
12
Earlier this year, Anthropic accused Moonshot AI, the company behind Kimi, alongside a bunch of other Chinese companies like DeepSeek and Minimax,
13
of running large -scale distillation attacks against Claude by sending swarms of fake users to steal Claude's output and reasoning tokens,
14
which could then be used to train these open models that they give to us for free.
15
Now, I believe distillation is a fundamental human right
16
because big AI companies have literally distilled the collective human brain by taking our internet weblogs, our githubs, our scientific papers,
17
and they even destructively scanned our books without asking.
18
Just google Project Panama.
19
But OpenAI and Anthropic don't really see distillation as a human right like I do.
20
To them, it's
21
Grand Theft Autocomplete, and they've invested millions and millions of dollars into stopping it.
22
And just a few weeks ago, Anthropic said it pretty much solved the distillation problem,
23
with 90 to 99 % of distillation attacks from China now being blocked.
24
And over at OpenAI, they've proved that their tech can even stop small -scale distilleries like PewDiePie.
25
But if that's true, that means an open Chinese model with Kimi's performance should not be possible,
26
because for years, Big AI has insinuated that the reason Chinese models get so good so fast is that they steal
27
Big AI's output.
28
But if we can trust this
29
Trust Me Bro distillation data from Anthropic, that would mean that these
30
Trust Me Bro benchmarks from Kimi should not be possible.
31
Like Kimi K3 .1 is within one point of Opus 5 .5 on Terminal Bench 4 .0.
32
That's pretty crazy, and no other open model is remotely close.
33
And we might even be able to actually trust these numbers because artificial analysis ran it independently and got 67%,
34
which is the highest score of any model they tested.
35
But one thing to keep in mind is that the harness plays a big role here.
36
Running different models with different harnesses will produce huge differences in benchmark results.
37
In addition, Kimmy also wins on SWE Marathon, which is a test for multi -hour software requirements,
38
or in other words, jobs that would take humans weeks of work to finish without AI assistance.
39
That's all pretty impressive, but there is one big caveat when it comes to coding, and specifically cybersecurity.
40
Moonshot deliberately ripped out a bunch of cybersecurity data from its training dataset,
41
where the goal was to make the model even worse at hacking than pirate software.
42
I'm a genius, dude.
43
And it worked, because internally at Moonshot, an unfiltered checkpoint solved 90 .5 % of their illegal exploit tests,
44
but the final release model only solves 81%, and that's a smart business move on their part, because just a few days ago, the US banned
45
China's DeepSeq model from government use, including all federal civilian agencies,
46
and they would just love a good reason to ban all Chinese models across the board, to hand OpenAI and Anthropic a huge win,
47
and make all the insiders even more unbelievably rich than they already are.
48
And that's why Moonshot really has no choice but to lobotomize its hacking capabilities.
49
But despite that, Kimmy still performed on top of all publicly available models on the new cyber index at Artificial Analysis,
50
although that's mostly because Claude and GPT -6 declined every prompt on safety grounds and just took the L.
51
But nowadays, the only AI capability that I really care about is how well it can generate a 3D horse with 3JS.
52
And to analyze the skill objectively and scientifically, I use a formal benchmark.
53
It's a state -of -the -art eval I've developed after thousands of hours of research, which I call Thoroughbench.
54
The way it works is I literally just go into the chat window
55
and type create a realistic interactive 3D horse with 3JS and then press enter.
56
From there, I rank them from 0 to 10 based on a proprietary loss function I keep in my brain.
57
Using this technique, Google Argon achieved a 0 .69, GPT -6 Astra achieved a 5 .21, and Opus achieved a 6 .39.
58
But now it's time to try Kimi K 3 .1.
59
I tried to be cheap and use the free tier,
60
but got denied, so now I'm paying 250 yuan per month to show you guys this demo.
61
It was worth it though, because Kimi produced what is easily the most beautiful horse in this sample.
62
Its coloring is spot on, it looks realistic, and most importantly, its copulatory organ is a reasonable size.
63
I was blown away by its beauty at first glance, but then I pushed play on the animation and found a problem.
64
The animations are absolutely terrible.
65
If we look closely, you can see it achieves the animation by simply warping the original mesh
66
which is a very unsophisticated and half -assed way to animate a 3D object.
67
If we compare that to Claude, the horse itself may not look as beautiful,
68
but it actually built a mesh with 95 ,000 vertices with each shape tied to one of 35 bones, and all those bones are animated independently.
69
Even Gemini Argon used bones, but for some reason it added an extra one.
70
Skeletal animation is more difficult, but it's why the horse doesn't look like it's being torn apart by a black hole when it walks.
71
So the final score for KimiK 3 .1 on Thoroughbench is a 4 .20, which is pretty dang impressive, but still behind the other Frontier models.
72
But if you really want to build an amazing 3D horse or user interface, you need to check out Mobbin, the sponsor of today's video.
73
I've been using Mobbin for over five years,
74
and its MCP server now lets your coding agent study its library of 600 ,000 real UI screens before it designs anything.
75
So instead of asking for a login page and getting the same purple gradient vibe slop, you start by telling ClaudeCode,
76
Cursor, or Codex to study how the best apps built that complete user flow.
77
It'll use Mobbin to gather references and build a visual report of the patterns they all share.
78
Then it'll recreate that flow and adapt it to your app.
79
Every result points back to the original screen, so you can check the source, and after you connect Mobbin,
80
everything your agent designs will feel like it has actual taste.
81
But try Mobbin for free today and get a 20 % discount at the link below.
82
This has been The Code Report, thanks for watching, and I will see you in the next one.

Télécharger l'application

Tout pour parler couramment

Prononciation IAChaque phrase notée
Phonétique (IPA)Maîtrisez chaque son
VocabulaireEnrichissez votre lexique
Jeu de vocabulaireApprenez en jouant

Même chaîne

FOR YOU

Vidéo recommandée

Vocabulaire et conseils d’expression pour cette leçon

Cette leçon d’expression orale s’appuie sur la vidéo « China just dropped a new frontier model... and nobody can explain how ». Les mots qui reviennent le plus souvent : model, horse, distillation, Kimmy, Anthropic. Cette vidéo contient 82 phrases et 1258 mots à répéter en shadowing. La partie parlée dure 6:21. Le locuteur parle vite, environ 198 mots par minute : attendez-vous à des liaisons et à des sons réduits. Seuls 76 % des mots font partie des 3 000 mots les plus courants en anglais, le vocabulaire est donc exigeant.

Vocabulaire clé de cette vidéo

15 mots de la vidéo à retenir, avec leur prononciation et leur sens :

MotPrononciationSens
benchmark nom/ˈbɛn(t)ʃmɑːk/critère, aune
animation nom/ˌæn.ɪˈmeɪ.ʃən/animation, invigoration
bone nom/boʊn/os
solve verbe/sɒlv/résoudre, régler
impressive adjectif/ɪmˈpɹɛsɪv/impressionnant, épatant
mystery nom/ˈmɪs.t(ə.)ɹi/mystère
output nom/ˈaʊtpʊt/sortie
steal verbe/stiːl/voler
artificial adjectif/ˌɑː.tɪˈfɪʃ.əl/artificiel
realistic adjectif/ˌɹɪəˈlɪs.tɪk/réaliste
capability nom/ˌkeɪ.pəˈbɪl.ɪ.ti/capacité
independently adverbe/ˌɪndɪˈpɛndəntli/indépendamment
frontier nom/ˈfɹʌntɪə/frontière
mesh nom/mɛʃ/maillage
harness verbe/ˈhɑɹ.nəs/harnacher

Les verbes à particule que vous entendrez

MotSens
blow away verbesouffler
check out verbemater, regarder
find out verbedécouvrir, savoir
rip out verbearracher
show up verbeapparaître, se montrer se pointer

La grammaire de cette vidéo

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

StructureDans la vidéo
Present perfect have/has + participe passé — une action passée qui compte encore maintenanthave literally distilled · they've invested · they've proved
Voix passive be + participe passé — l’accent est mis sur ce qui arrive, pas sur qui le faitbeing blocked · was blown · are animated

Prononciation à surveiller

Le locuteur utilise 12 contractions et formes réduites, comme I'm, I've, it'll. Prononcez-les sous leur forme courte, telles que vous les entendez.

  • Les sons « sh » et « zh »: benchmark /ˈbɛn(t)ʃmɑːk/, animation /ˌæn.ɪˈmeɪ.ʃən/, artificial /ˌɑː.tɪˈfɪʃ.əl/, mesh /mɛʃ/, evaluation /ɪˌvæljuˈeɪʃn̩/
  • Mots longs — placez bien l’accent: animation /ˌæn.ɪˈmeɪ.ʃən/, artificial /ˌɑː.tɪˈfɪʃ.əl/, capability /ˌkeɪ.pəˈbɪl.ɪ.ti/, independently /ˌɪndɪˈpɛndəntli/, fundamental /ˌfʌn.dəˈmɛn.təl/

Les sons difficiles pour les francophones :

  • /tʃ/ et /dʒ/ — à ne pas adoucir en « ch » et « j »: genius /ˈd͡ʒin.jəs/, generate /ˈd͡ʒɛn.ə.ɹeɪt/, chess /t͡ʃɛs/, analogy /əˈnæləd͡ʒi/, checkpoint /ˈt͡ʃɛkˌpɔɪnt/
  • /r/ anglais — langue recourbée, sans frotter la gorge: impressive /ɪmˈpɹɛsɪv/, mystery /ˈmɪs.t(ə.)ɹi/, realistic /ˌɹɪəˈlɪs.tɪk/, frontier /ˈfɹʌntɪə/, harness /ˈhɑɹ.nəs/

Comment s’entraîner avec cette vidéo

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

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 →