Shadowing Practice: An ex-OpenAI researcher just deleted language from the LLM... - Learn English Speaking with Video
בחר מצב תרגול
כל מצב פותח חדר תרגול במסך מלא.
59 משפטים
אם המשפטים קצרים או ארוכים מדי, לחץ על ערוך כדי להתאים אותם.
1
Last week, everything about AI changed forever.
2
I realize everything about AI changes forever almost every week, but this time, AI changed forever more than usual.
3
Because one of the OpenAI researchers behind the instruction following work
4
that eventually became ChatGPT just released the next big frontier model after two years in stealth development.
5
A model that can't talk, a model that can't write code, a model that can't write your college essays, and a model that will never tell you you're absolutely right.
6
Its name is Jev.
7
My name is Jev.
8
And this is a huge deal because large language models have one fatal flaw.
9
They won't shut the hell up.
10
You give Fable or Astra a simple instruction like return true or false, and it'll discover a third option after thinking for 4 ,000 tokens and then charge your credit card 11 cents.
11
Jev fixed this problem with a radical solution.
12
It deleted language from the large language model, and the result is a new type of classifier that's 200 times faster,
13
400 times cheaper, with free output tokens and zero hallucinations.
14
This sounds too good to be true, so in today's video, we'll take a look at Jev's code, it's TrustMeBroBenchmarks, and the dude who says he built an open -source Jev over a year ago.
15
It is September 21st, 2026, and you're watching the Code Report.
16
The big AI duopoly is literally shaking right now, because Jev is a cheaper, faster way to solve basically any AI problem that requires a quick gut -instinct decision.
17
It's afraid.
18
But the first thing you need to know is that Jev was created by an ex -open AI researcher, Diogo Almeida, and his company TypeSafe AI, which just raised $40 million.
19
But the company name is the first clue to what JEV really is.
20
Like a regular large language model, you send it a question and some context, like a bunch of unstructured text.
21
However, it differs because it behaves more like a TypeSafe programming language, like TypeScript.
22
The question you send to the model is a strongly typed question that must return a specific shape.
23
One of three shapes actually, a choice, a score, and a null, which is basically just a yes or no. Its schema matching is guaranteed,
24
and a type error would be mathematically impossible to produce.
25
They call Jev a System 1 model, which is a name that comes from Daniel Kahneman's Thinking Fast and Slow.
26
A System 1 model is fast and goes from gut instinct, while a System 2 model is slow and deliberate,
27
like these old antique reasoning models like GPT -6 and Claude Fable that burn 40 ,000 tokens to name a variable,
28
but the difference is huge for app developers like myself who want to integrate fast, cheap AI into their applications.
29
Like on Horse Tinder, we recently had an issue of some donkeys trying to use the app, which is strictly forbidden in the terms of service.
30
Thanks to Jev, we implemented an AI moderation step that will instaban any account that is not a horse, which is accomplished by returning a null response to is this a horse.
31
Not only is it extremely fast if we are to believe these TMBBs, but more importantly, it's off the charts cheap, like 440 times cheaper than one of the big brand models.
32
In fact, it's so fast and cheap that you can even use it for real -time applications, like developers are already using it to implement NPC behavior in video games.
33
And this guy even used it to build the world's first real -time AI calculator.
34
But just because the output is type safe, that doesn't mean it's always correct.
35
And it's not even deterministic, like you could send it the exact same question and the exact same context and get different results, just like any regular large language model.
36
But to get an idea of the response quality, it returns something called the calibrated confidence number.
37
Chap models are trained to police human raiders, and humans love confidence, which is how we got models that are wrong with the confidence of Kanye.
38
Jeb gained its confidence through a technique called RLCD, or reinforcement learning for calibrated decisions.
39
This means every response provides a confidence value, like say 60%, which means 60 % of the time, it's right every time.
40
But the big question is how does Jeb actually work?
41
Well, nobody knows for sure, because the CEO says the architecture is staying close to the chest, with a paper possibly coming in the future, maybe.
42
But Jev also has some doubters.
43
Some people say it's no different than zero -shot classifiers of the past, but the company gives no credit to the original pioneers of this technique,
44
like Jin Yang, who were building zero -shot classifiers over a decade ago.
45
In addition, this guy claims his paper he released a year ago is the exact same thing as Jev, and another developer already built OpenJev,
46
which reproduces the entire interface by reading option probabilities off a frozen Quinn 4B model in a single forward pass.
47
It requires no new training and can run on a 3090.
48
And there's even a web GPU demo you can run in your browser right now.
49
It's an awesome time to be a developer, which is why you need to check out MUX, the sponsor of today's video.
50
Their highly customizable API is by far the easiest way to
51
add video features to your application without getting jump scared by FFmpeg.
52
We've used it for years to handle all the hosting and streaming for our courses, but it does a lot more than just infrastructure.
53
When you upload a video to MUX, you automatically get transcripts, storyboards, thumbnails, and clips, along with structured data about what's actually in the video.
54
That powers MUX robots, which is their AI -hosted workflows that can translate your audio into other languages.
55
moderate content, and lots more without you needing to host a model or maintain a pipeline.
56
You can automate all this with directives where you define a workflow once and it runs on every new upload, and you only pay for the jobs that actually run.
57
Perplexity, Patreon, and many other prestigious companies all trust MUX, and their free plan includes 10 videos and 100 ,000 delivery minutes per month with no credit card required.
58
And you can get an extra $50 credit at the link below.
59
This has been The Code Report, thanks for watching, and I will see you in the next one.
הורד אפליקציה
כל מה שאתה צריך כדי לדבר בשטף
הגיית AIציון לכל משפט
תרגול IPAשלוט בכל צליל
אוצר מיליםבנה את מאגר המילים שלך
משחק אוצר מיליםלמד תוך כדי משחק
אותו ערוץ
C11
PostgreSQL in 100 Seconds
Daily ConversationWorking Adults
C11
React Native in 100 Seconds
Daily ConversationUniversity Students
C12
Did Google just dethrone OpenAI and Anthropic?
Daily ConversationUniversity StudentsWorking Adults
C13
Every app on Earth is now open source... whether it likes it or not
Daily ConversationWorking Adults
C14
Brainf**k in 100 Seconds
Daily ConversationUniversity Students
C13
OpenAI’s lead engineer was secretly a spy...
Daily ConversationUniversity StudentsWorking Adults
C19
The most interesting "hack" in history...
American AccentBusiness EnglishUniversity StudentsWorking Adults
C12
Mind-bending new programming language for GPUs just dropped...
Daily ConversationUniversity Students
C12
China just dropped a new frontier model... and nobody can explain how
Daily ConversationUniversity StudentsWorking Adults
C12
The last real programmers are dying out...
Daily ConversationUniversity StudentsWorking Adults
C11
Did the AI bubble just pop?
Daily ConversationUniversity StudentsWorking Adults
B26
Intel chips can’t possibly be this bad… 100% crash rate?
American AccentBusiness EnglishUniversity StudentsWorking Adults
B23
OpenAI’s new browser feels familiar…
American AccentBusiness EnglishUniversity StudentsWorking Adults
C130
Docker in 100 Seconds
American AccentBusiness EnglishUniversity StudentsWorking Adults
B23
Firefox and Tor hit with 9.8 critical level exploit
American AccentBusiness EnglishUniversity StudentsWorking Adults
B23
5 open source tools that replaced my $320/mo AI stack...
American AccentBusiness EnglishUniversity StudentsWorking Adults
C12
C++ in 100 Seconds
American AccentBusiness EnglishUniversity StudentsWorking Adults
B24
Linux in 100 Seconds
American AccentBusiness EnglishUniversity StudentsWorking Adults
B22
Meta is pivoting again... everything you missed from Connect 2026
American AccentBusiness EnglishUniversity StudentsWorking Adults
C116
VS Code in 100 Seconds
American AccentBusiness EnglishUniversity StudentsWorking Adults
FOR YOU
המלץ על סרטון
B22
54 משפטים
One man just liberated Fable... and now it’s illegal
American AccentBusiness EnglishUniversity StudentsWorking Adults
B22
239 משפטים
Learn 95% of ChatGPT Work in Under 20 Minutes
American AccentBusiness EnglishUniversity StudentsWorking Adults
B22
79 משפטים
AI is flooding the internet—here’s what I’m protecting
American AccentBusiness EnglishUniversity StudentsWorking Adults
B21
43 משפטים
Introducing GPT-6 Astra: the most intelligent and aligned model in the world.
Business EnglishUniversity StudentsWorking Adults
C252
81 משפטים
Large Language Models explained briefly
American AccentBusiness EnglishUniversity StudentsWorking Adults
B23
165 משפטים
How I'd Learn AI From Scratch in 2026 (skip the useless 80%)
American AccentBusiness EnglishUniversity StudentsWorking Adults
B27
179 משפטים
OpenAI: A Bubble Bigger Than Dotcom
Business EnglishDaily ConversationUniversity Students
B21
169 משפטים
Can This Vibe Coder Beat a Senior Developer?
Daily ConversationUniversity StudentsWorking Adults
B24
180 משפטים
Is AI Doom Real... or Just Good Marketing?
Business EnglishUniversity StudentsWorking Adults
B25
125 משפטים
10 Times AI Behaved In Ways That Terrified The Scientists Who Built It
Business EnglishDaily ConversationUniversity Students
B21
28 משפטים
Gemma 4 Just Released - Whats In? In 2026!
Daily ConversationUniversity StudentsWorking Adults
אוצר מילים והערות הגייה לשיעור הזה
שיעור הדיבור הזה ברמה B2 מבוסס על הסרטון “An ex-OpenAI researcher just deleted language from the LLM..”. המילים שחוזרות הכי הרבה: model, Jev, language, confidence, fast. בסרטון הזה יש 59 משפטים ו-1079 מילים לתרגול Shadowing. הדיבור נמשך 5:25. הדובר מדבר מהר, כ-199 מילים בדקה, אז צפו לצלילים מחוברים ומקוצרים. רק 80% מהמילים נמצאות בין 3,000 המילים הנפוצות ביותר באנגלית, כך שאוצר המילים מאתגר.
אוצר מילים מרכזי בסרטון הזה
15 המילים המתקדמות ביותר בסרטון, עם ההגייה והמשמעות שלהן:
| מילה | הגייה | משמעות |
|---|---|---|
| developer שם עצם | /dɪˈvɛləpɚ/ | מְפַתֵּחַ |
| token שם עצם | /ˈtoʊkən/ | אֲסִימוֹן, זִ׳יטוֹן |
| calibrate פועל | /ˈkæl.ɪ.bɹeɪ̯t/ | כייל |
| instinct שם עצם | /ˈɪn.stɪŋkt/ | אִינְסְטִינְקְט, יֵצֶר |
| instruction שם עצם | /ɪnˈstɹʌkʃən/ | הוראה |
| implement פועל | /ˈɪmpləmənt/ | יישם, מימש |
| duopoly שם עצם | /duˈɑpəli/ | דואופול |
| hallucination שם עצם | /həˌluːsɪˈneɪʃən/ | הזיה |
| moderation שם עצם | /ˌmɑdəˈɹeɪʃən/ | פיקוח |
| perplexity שם עצם | /pəɹˈplɛksəti/ | מבוכה |
| calculator שם עצם | /kæl.kjə.leɪ.tɚ/ | מַחְשְׁבוֹן, מונה |
| donkey שם עצם | /ˈdɑŋ.ki/ | חֲמוֹר, אָתוֹן |
| reproduce פועל | /ˌɹi.pɹəˈdus/ | התרבה |
| forbid פועל | /fɚˈbɪd/ | אָסַר |
| deliberate שם תואר | /dɪˈlɪb.(ə.)ɹət/ | מכוון |
Phrasal verbs שתשמעו
| מילה | משמעות |
|---|---|
| check out פועל | בדק |
דקדוק בסרטון הזה
המבנים שהדובר משתמש בהם הכי הרבה, עם המילים המדויקות מהסרטון:
| מבנה | בסרטון |
|---|---|
| משפטי זיקה (Relative clauses) who / which + clause — מידע נוסף על אדם או דבר | null, which is · model, which is · app, which is |
| סביל (Passive voice) be + past participle — הדגש הוא על מה שקורה, לא על מי שעושה את הפעולה | was created · is guaranteed · is accomplished |
הגייה ששווה לשים לב אליה
הדובר משתמש ב-10 קיצורים וצורות מקוצרות, כמו can't, you're, doesn't. אמרו אותם בצורה הקצרה, כפי שאתם שומעים אותם.
- הצלילים של “th”: thumbnail /ˈθʌm.neɪl/, stealth /stɛlθ/
- הצלילים של “sh” ו-“zh”: instruction /ɪnˈstɹʌkʃən/, hallucination /həˌluːsɪˈneɪʃən/, moderation /ˌmɑdəˈɹeɪʃən/, accomplish /əˈkɑm.plɪʃ/
- מילים ארוכות — שימו לב להטעמה: developer /dɪˈvɛləpɚ/, deterministic /dɪˌtɜː.mɪˈnɪs.tɪk/, duopoly /duˈɑpəli/, hallucination /həˌluːsɪˈneɪʃən/, moderation /ˌmɑdəˈɹeɪʃən/
צלילים שדוברי עברית מתקשים בהם:
- /æ/ — פתוח יותר מה-“אֶ” העברית, לא כמו “a” ב-“cup”: classifier /ˈklæsɪfaɪɚ/, calibrate /ˈkæl.ɪ.bɹeɪ̯t/, calculator /kæl.kjə.leɪ.tɚ/, transcript /ˈtɹæn.skɹɪpt/, translate /tɹænzˈleɪt/
- /r/ — הלשון לא נוגעת בשום מקום, לא ה-“ר” הגרונית: calibrate /ˈkæl.ɪ.bɹeɪ̯t/, researcher /ˈɹiˌsɝ.t͡ʃɚ/, instruction /ɪnˈstɹʌkʃən/, moderation /ˌmɑdəˈɹeɪʃən/, perplexity /pəɹˈplɛksəti/
- עיצור קולי בסוף מילה — /b/, /d/, /g/, /z/, /v/ בלי להפוך אותם לאטומים: upload /ˈʌpˌloʊd/, forbid /fɚˈbɪd/, directive /daɪˈɹɛk.tɪv/, behave /bɪˈheɪv/, freeze /ˈfɹiːz/
איך לתרגל עם הסרטון הזה
- הקשיבו לכל הסרטון פעם אחת בלי לדבר, ורשמו את המילים שאתם לא מכירים.
- התחילו במהירות 0.75×, תרגלו Shadowing משפט אחר משפט, וחזרו למהירות רגילה כשזה מרגיש קל.
- הקליטו את עצמכם והשוו למקור, תוך תשומת לב למילים כמו developer, token, calibrate.
מהי טכניקת השאדואינג?
שאדואינג היא טכניקת למידת שפה מבוססת מדע שפותחה במקור להכשרת מתורגמנים מקצועיים ופופולרית על ידי הפוליגלוט ד"ר אלכסנדר ארגואלס. השיטה פשוטה אך עוצמתית: אתה מקשיב לאודיו באנגלית ילידית וחוזר עליו בקול רם מיד — כמו צל העוקב אחרי הדובר עם עיכוב של 1–2 שניות בלבד. בניגוד להאזנה פסיבית או תרגילי דקדוק, שאדואינג מכריח את המוח והשרירים בפה לעבד ולשחזר דפוסי דיבור אמיתיים בו זמנית. מחקרים מראים שזה משפר משמעותית את דיוק ההגייה, האינטונציה, הקצב, הדיבור המחובר, הבנת השמיעה ושטף הדיבור — מה שהופך אותו לאחת השיטות היעילות ביותר להכנה לIELTS Speaking ולתקשורת באנגלית בעולם האמיתי.