AI sohasida ishlash uchun ingliz tili ish vositasining o‘zi: hujjatlar, kutubxonalar, xato xabarlari, kurslar va ilmiy maqolalarning deyarli hammasi inglizcha. Eng avval texnik matnni o‘qish kerak bo‘ladi, keyin qisqa va aniq yozish (xabar, hisobot, bug report), undan keyin gapirish (stand-up, demo, intervyu). Qaysi ko‘nikma birinchi o‘rinda turishi kasbga bog‘liq; quyidagi jadval shuni ko‘rsatadi.
Bu sahifada maosh raqamlari yo‘q: ular kompaniya, shahar va tajribaga qarab juda farq qiladi. Aniq talablarni har doim ish e’lonining o‘zidan o‘qing.
Yetti kasb va ulardagi til vazifalari
| Kasb | Nima ish qiladi | Ingliz tilida eng ko‘p nima qiladi |
|---|---|---|
| Data annotator / AI trainer | Ma’lumotlarni belgilaydi, model javoblarini baholaydi | Uzun ko‘rsatmalarni (guidelines) aniq o‘qiydi, qisqa izoh yozadi |
| Data analyst | Ma’lumotdan xulosa chiqaradi | Hisobot yozadi, grafikni tushuntiradi, savol beradi |
| Machine learning engineer | Modelni o‘qitadi, sinaydi, ishga tushiradi | Hujjat va maqola o‘qiydi, kodga izoh va PR tavsifi yozadi |
| LLM ilovalari dasturchisi | Tayyor modellar ustida mahsulot quradi | API hujjatlarini o‘qiydi, promt yozadi, natijani baholaydi |
| MLOps engineer | Model ishlashini, serverlarni kuzatadi | Xato loglarini o‘qiydi, incident hisobotini yozadi |
| AI product manager | Mahsulot yo‘nalishini belgilaydi | Ko‘p gapiradi: yig‘ilish, taqdimot, mijoz bilan suhbat |
| Technical writer | Hujjat va qo‘llanma yozadi | Aniq, sodda inglizcha yozadi; bu ishning o‘zi |
Ko‘rinib turibdiki, bitta “AI ingliz tili” yo‘q. Annotator uchun o‘qishdagi aniqlik muhim: ko‘rsatmadagi unless yoki only if so‘zini o‘tkazib yuborsangiz, yuzlab belgilash noto‘g‘ri chiqadi. Product manager uchun esa ravon gapirish birinchi o‘rinda. Dasturlash yo‘lidan borishni rejalashtirayotgan bo‘lsangiz, bosqichlar va har bosqichdagi til darajasi sun’iy intellektni o‘rganish yo‘l xaritasi sahifasida.
Ish kunida uchraydigan 20 ta atama
Har bir so‘zni misol gapi bilan yodlang: AI jamoasida bu so‘zlar aynan shunday birikmalarda ishlatiladi.
| Atama | O‘zbekcha | Misol |
|---|---|---|
| dataset | ma’lumotlar to‘plami | The dataset has 20,000 labelled reviews. |
| label (v, n) | belgilamoq; belgi | Please label each image as “cat” or “not cat”. |
| train a model | modelni o‘qitmoq | We trained the model on Uzbek news texts. |
| fine-tune | tayyor modelni moslashtirmoq | We fine-tuned a small model for customer support. |
| evaluate | baholamoq | Let’s evaluate the new version before we release it. |
| accuracy | aniqlik | Accuracy dropped after the last update. |
| overfitting | haddan tashqari moslashish | The model is overfitting: it remembers the training data. |
| bias | tarafkashlik | We found a gender bias in the answers. |
| inference | tayyor modeldan javob olish | Inference is too slow on this server. |
| latency | kechikish vaqti | We need to reduce latency to under one second. |
| deploy | ishga tushirmoq | We will deploy the fix on Friday. |
| pipeline | ketma-ket ishlov bosqichlari | The data pipeline broke last night. |
| benchmark | taqqoslash testi | Our model scored higher on this benchmark. |
| edge case | kam uchraydigan holat | Mixed Uzbek-Russian messages are an edge case. |
| reproduce | xatoni qayta chiqarmoq | I can’t reproduce the bug on my laptop. |
| prompt | modelga ko‘rsatma | Can you share the prompt you used? |
| token | matn bo‘lagi | This request uses too many tokens. |
| guidelines | ko‘rsatmalar | Read the guidelines before you start labelling. |
| deadline | muddat | The deadline for the report is Thursday. |
| blocker | ishni to‘xtatib turgan muammo | My blocker is access to the database. |
To‘liqroq ro‘yxat, o‘zbekcha izohlar bilan, sun’iy intellekt so‘zlari inglizcha sahifasida.
To‘rtta namunaviy vazifa
Quyidagi vazifalar AI jamoalaridagi odatiy ishlar asosida mashq uchun yozilgan. Har birini o‘zingiz bajarib ko‘ring.
1. Ko‘rsatmani o‘qish (annotator)
Label the review as positive, negative or neutral. If the review contains both praise and a complaint, choose mixed. Do not label reviews that are not about the product; mark them as off-topic.
Savol: “The delivery was late, but the phone works perfectly” qaysi belgi oladi? Javob: mixed, chunki bitta sharh ichida ham maqtov, ham shikoyat bor. Ko‘p boshlovchilar but dan keyingi qismga qarab positive qo‘yadi. Ko‘rsatmada har bir if, unless, do not ni alohida belgilab o‘qing.
2. Stand-up yangiligi (har qanday texnik rol)
Kunlik yig‘ilishda 30–60 soniya gapirasiz. Qolip:
Yesterday I cleaned the new dataset and removed duplicates. Today I’m going to train the first version of the model. One blocker: I still don’t have access to the GPU server.
Ko‘p uchraydigan xato: I am cleaning the dataset yesterday. O‘tgan kun uchun Past Simple kerak: Yesterday I cleaned… Bugungi reja uchun I’m going to yoki I’ll. Stand-up iboralari batafsil IT stand-up inglizcha sahifasida.
3. Bug report (ML/MLOps engineer)
- Title: Model returns empty answer for Uzbek Cyrillic input
- Steps to reproduce: 1) Open the demo. 2) Type “Салом, қалайсиз?” 3) Press Send.
- Expected: A greeting in reply.
- Actual: An empty response, no error message.
Bu formatda sarlavha muammoni bitta gapda aytadi, Expected va Actual esa farqni ko‘rsatadi. “It doesn’t work” degan sarlavha hech kimga yordam bermaydi.
4. Maqola xulosasini o‘qish (tadqiqotchi, ML engineer)
Ilmiy maqolaning abstract qismini to‘liq tarjima qilmang. Uchta savolga javob toping: What problem? What method? What result? Bu usul bilan bir abstraktni 5 daqiqada tushunasiz. Qiyin jumlani AI’ga soddalashtirishni so‘rash mumkin, lekin raqam va natijalarni asl matndan tekshiring.
Til xatolari: jamoa chatida nima g‘alati eshitiladi
| Noto‘g‘ri | To‘g‘ri | Sabab |
|---|---|---|
| I did a research about bias. | I did some research on bias. | Research sanalmaydi, a olmaydi |
| Can you give me informations? | Can you give me some information? | Information ko‘plik olmaydi |
| I have a question to you. | I have a question for you. | Ruscha “вопрос к вам” kalkasi |
| I have finished it yesterday. | I finished it yesterday. | Aniq o‘tgan vaqt bilan Present Perfect ishlatilmaydi |
| Please, check my PR. | Could you review my PR, please? | Buyruq ohangi qattiq eshitiladi |
| The model is learning now. | The model is training now. | Modelni “o‘qitish” — train |
Oxirgi qatorga e’tibor bering: o‘zbekcha “o‘qimoq” ham read, ham study bo‘lishi mumkin, “o‘qitmoq” esa AI kontekstida train. Bu farq texnik suhbatda tez-tez chiqadi.
O‘rganish yo‘li: uch bosqich
- 1-bosqich (1–2 oy): texnik o‘qish. Har kuni 20 daqiqa bitta kutubxona hujjatidan (masalan, Python yoki scikit-learn Getting started sahifasi) o‘qing. Notanish so‘zlarni jadvaldagi kabi misol bilan yozib boring.
- 2-bosqich (2–3 oy): qisqa yozish. Har bir kichik loyihangiz uchun inglizcha README yozing: nima qiladi, qanday ishga tushiriladi, qanday cheklovi bor. Kodga izohlarni ham inglizcha yozing.
- 3-bosqich (doimiy): gapirish. Haftada ikki marta loyihangizni 2 daqiqada tushuntirishni mashq qiling va o‘zingizni yozib oling. Ish suhbati savollari uchun ish suhbati savollari sahifasidan foydalaning.
Bepul kurslar ham inglizcha: ularni o‘tish uchun qanday daraja kerakligi sun’iy intellekt bo‘yicha bepul kurslar sahifasida ko‘rib chiqilgan.
Mashq: jamoa chatidagi gaplarni tuzating
Har bir gapda bitta xato bor. Tuzating.
- 1. Yesterday I am training the model on the new data.
- 2. We need more datas for the edge cases.
- 3. I can’t reproduce this bug, can you send me the steps for reproduce?
- 4. I think the model has overfitting.
- 5. I will deploy it tomorrow, if I will get access.
- 6. Could you explain me how the pipeline works?
- 7. We made an evaluation of three models and the second is better.
- 8. Please read guidelines before labelling.
Javoblar
- 1. Yesterday I trained the model on the new data. O‘tgan vaqt — Past Simple.
- 2. We need more data for the edge cases. Data ko‘plik qo‘shimchasini olmaydi.
- 3. …the steps to reproduce it? Maqsad uchun to + fe’l.
- 4. I think the model is overfitting. Overfit — fe’l; “modelda overfitting bor” deb tarjima qilmang.
- 5. …if I get access. If dan keyin kelasi zamon uchun Present Simple.
- 6. Could you explain to me how… yoki Could you explain how…
- 7. We evaluated three models, and the second one performed better. Qisqa fe’l tabiiyroq.
- 8. Please read the guidelines before labelling. Aniq, ma’lum hujjat — the.
Stand-up va intervyuni ovoz chiqarib mashq qilish
Yozish va o‘qishni o‘zingiz rivojlantirasiz, gapirishga esa suhbatdosh kerak. BandMate ilovasida Bandi bilan ish mavzusidagi suhbatlarni ovoz chiqarib mashq qilishingiz, loyihangizni tushuntirib ko‘rishingiz va talaffuz bo‘yicha fikr olishingiz mumkin. Bu AI fikri: foydali mashq, lekin haqiqiy ish suhbati yoki til sertifikati o‘rnini bosmaydi.
Bugun bitta vazifa: yuqoridagi stand-up qolipiga o‘z kuningizni qo‘yib, uni uch marta ovoz chiqarib ayting va eng yaxshi variantini yozib oling.
BandMate’ni Google’da afzal manba qiling
