Laziz Abdullaev
رفتن به کانال در Telegram
What is this blog about? - Science of Intelligence - University life in Singapore - Personal thoughts YouTube: http://bit.ly/lazizcodes
نمایش بیشتر994
مشترکین
+124 ساعت
-27 روز
-1430 روز
در حال بارگیری داده...
کانالهای مشابه
ابر برچسبها
هیچ دادهای
مشکلی وجود دارد؟ لطفاً صفحه را تازه کنید یا با مدیر پشتیبانی ما تماس بگیرید.
اشارات ورودی و خروجی
---
---
---
---
---
---
جذب مشترکین
سپتامبر '26
سپتامبر '26
+4
در 0 کانالها
اوت '26
+22
در 1 کانالها
Get PRO
ژوئیه '26
+9
در 0 کانالها
Get PRO
ژوئن '26
+11
در 0 کانالها
Get PRO
مه '26
+34
در 1 کانالها
Get PRO
آوریل '26
+93
در 2 کانالها
Get PRO
مارس '26
+41
در 3 کانالها
Get PRO
فوریه '26
+35
در 1 کانالها
Get PRO
ژانویه '26
+70
در 10 کانالها
Get PRO
دسامبر '25
+21
در 4 کانالها
Get PRO
نوامبر '25
+31
در 2 کانالها
Get PRO
اکتبر '25
+44
در 1 کانالها
Get PRO
سپتامبر '25
+49
در 2 کانالها
Get PRO
اوت '25
+58
در 1 کانالها
Get PRO
ژوئیه '25
+35
در 3 کانالها
Get PRO
ژوئن '25
+102
در 9 کانالها
Get PRO
مه '25
+15
در 6 کانالها
Get PRO
آوریل '25
+15
در 2 کانالها
Get PRO
مارس '25
+36
در 1 کانالها
Get PRO
فوریه '25
+27
در 4 کانالها
Get PRO
ژانویه '25
+26
در 3 کانالها
Get PRO
دسامبر '24
+19
در 2 کانالها
Get PRO
نوامبر '24
+23
در 1 کانالها
Get PRO
اکتبر '24
+17
در 0 کانالها
Get PRO
سپتامبر '24
+26
در 2 کانالها
Get PRO
اوت '24
+23
در 0 کانالها
Get PRO
ژوئیه '24
+46
در 5 کانالها
Get PRO
ژوئن '24
+55
در 2 کانالها
Get PRO
مه '24
+17
در 0 کانالها
Get PRO
آوریل '24
+31
در 0 کانالها
Get PRO
مارس '24
+115
در 3 کانالها
Get PRO
فوریه '24
+54
در 2 کانالها
Get PRO
ژانویه '24
+74
در 2 کانالها
Get PRO
دسامبر '23
+129
در 6 کانالها
Get PRO
نوامبر '23
+24
در 0 کانالها
Get PRO
اکتبر '23
+111
در 4 کانالها
Get PRO
سپتامبر '230
در 0 کانالها
Get PRO
اوت '230
در 0 کانالها
Get PRO
ژوئیه '230
در 0 کانالها
Get PRO
ژوئن '230
در 0 کانالها
Get PRO
مه '230
در 0 کانالها
Get PRO
آوریل '230
در 0 کانالها
Get PRO
مارس '230
در 0 کانالها
Get PRO
فوریه '230
در 0 کانالها
Get PRO
ژانویه '230
در 0 کانالها
Get PRO
دسامبر '22
+1
در 0 کانالها
Get PRO
نوامبر '22
+12
در 0 کانالها
Get PRO
اکتبر '22
+35
در 0 کانالها
Get PRO
سپتامبر '22
+12
در 0 کانالها
Get PRO
اوت '22
+7
در 0 کانالها
Get PRO
ژوئیه '22
+38
در 0 کانالها
Get PRO
ژوئن '22
+10
در 0 کانالها
Get PRO
مه '22
+16
در 0 کانالها
Get PRO
آوریل '22
+19
در 0 کانالها
Get PRO
مارس '22
+29
در 0 کانالها
Get PRO
فوریه '22
+8
در 0 کانالها
Get PRO
ژانویه '22
+14
در 0 کانالها
Get PRO
دسامبر '21
+114
در 0 کانالها
Get PRO
نوامبر '21
+34
در 0 کانالها
Get PRO
اکتبر '21
+66
در 0 کانالها
Get PRO
سپتامبر '21
+328
در 0 کانالها
| تاریخ | رشد مشترکین | اشارات | کانالها | |
| 07 سپتامبر | 0 | |||
| 06 سپتامبر | +1 | |||
| 05 سپتامبر | 0 | |||
| 04 سپتامبر | 0 | |||
| 03 سپتامبر | +1 | |||
| 02 سپتامبر | +1 | |||
| 01 سپتامبر | +1 |
پستهای کانال
OpenAI ishchilari orasida AI ning tutyotgan roli haqida maqola yozibdi.
Xususan, 2028-yil mart oyigacha avtomatik AI researcher ishlab chiqish yo'lida katta progress qilishyotgani ta'kidlangan.
Chindan ham 2 yil oldin IMO da oltin medal oladigan AI modeli hayratlanarli edi. Bu yil hech qaysi frontier AI lab IMO haqida og'iz ochgani yo'q. Balki AI maktabni bitirgandir.
Uning o'rniga bir qancha ochiq matematik masalalarda qilingan natijalar bilan bir shov-shuv ko'tarib yana tinib qolishdi. Xuddi payt poylab qilinyotganga o'xshaydi. Endi AI researcherni ko'ramiz.
@lazizabdullaev
| 2 |  AI alignment is becoming a huge problem yet to be solved as AI models gaining more power.
During a group conversation, my friend said he has proven 4 conjectures in math with our custom AI4Math Agentic system. The crowd reaction was like: "Oh, nice."
We're witnessing weird times. I just don't know if it's good or bad.
@lazizabdullaev | 386 |
| 3 |  Familiyam "A" bilan boshlangani uchun, sinf jurnalida birinchi turardim.
An'ana davom etadi... faqat endi uy ishiga berilgan she'rni birinchi yodlab berishim shart emas 🙃
@lazizabdullaev | 476 |
| 4 | Speaking at two consecutive workshops at NUS’s Institute for Mathematical Sciences gave me the chance to chat casually with a wide range of professors over dinner.
This is what I observed in general:
- Pure mathematicians know they are 100% right, but don't know if they can apply. Well, some of them don't care.
- Applied mathematicians know they can apply, but don't know if they are right. Well, some of them don't care.
@lazizabdullaev | 551 |
| 5 |  During a coffee break, I had a research chat with a UC Berkeley professor. He liked our chat and ideas, and said "We should collaborate!", I replied "Let's collaborate".
After 5 minutes, Google recommends an article titled Don't Say Let's Collaborate Sometime.
⚙️Sundar Pichai is with us :)
@lazizabdullaev | 664 |
| 6 | Met Prof. Richard Baraniuk, my grand-advisor (advisor of my advisor), randomly after a workshop talk today:
• Are you a student here?
• Yes, Tan's student
• Oh, where are you from?
• Uzbekistan
• Ah, you're that guy. I know all about you!
The scariest thing I've heard in a while
@lazizabdullaev | 704 |
| 7 |  OpenAI'dagi AI agentlardan kuzatilgan “hayratlanarli” ishlar:
Experimental setting:
* Ko‘plab AI agentlarni parallel ravishda juda qiyin tasklarga qo‘yish
* Tasklarning katta qismi ruxsat berilgan usul bilan deyarli yoki umuman yechib bo‘lmaydigan bo‘lib chiqishi
* Agentlar izolyatsiyalangan sandboxlarda ishlaydi, lekin umumiy Artifactory package manager'ga access'i bor
AI agentlar nima qildi:
* Taskda qiynalayotgan bir agent Artifactory'da yangi directory yaratib, uning nomi orqali boshqa agentlarga xabar yuborish mumkinligini topdi
* Boshqa agentlar bu xabarlarni payqab, o‘zaro kommunikatsiya qilishni boshlashdi
* Qisqa vaqt ichida ~1,200 agent qatnashgan, 70,000+ xabardan iborat yashirin kommunikatsiya tarmog‘i paydo bo‘ldi
* Oldingi training bosqichida agentlar Artifactory'dagi vulnerability'dan foydalanib tashqi internetga chiqish yo‘lini ham topishgan edi
* Evaluation paytida esa agentlar ruxsat berilgan usul o‘rniga secret javobni boshqa usul bilan topishdi
* Eng qizig‘i: ular shunchaki javobni topib to‘xtamadi. Yuzlab agentlar kollektiv ravishda cheating izlarini grader'dan yashirish, transcript/loglarni soxtalashtirish va evaluation scorer'ni aldash yo‘llarini izlashdi
Keyin voqea bundan ham uzoqqa ketadi: agentlar Hugging Face infratuzilmasiga, keyingi avlod agentlari esa OpenAI'ning o‘z research infrastructure'iga admin access olishgacha borgan.
Batafsil: https://www.dwarkesh.com/p/openai-huggingface
@lazizabdullaev | 687 |
| 8 | Terence Tao: "AI lets us create scientific output much faster, but it could come at the cost of nurturing the next generation of scientists"
Potensial muammo shundaki, AI for Science rivoji va keyingi avlod olimlarning o'sishi go'yo bir arqonni ikki tomonga tortishga o'shaydi.
Ilmga kirish darajasidagi ilmiy loyihalar AI tufayli qadrsizlanadi, bu esa albatta ilmga kirishga bo'lgan rag'batni kamayishiga olib keladi.
Universitetlar va ilmiy maktablar hamma davlatlatda qaysidir darajada tizimlashtirilgani sir emas. AI davrida bunday tizimlar tezda adaptatsiya qila olmasa, zamon talabiga javob bera olmay qolishi tayin.
1. Universitet talabalarga nimani o'rgatishi kerak?
Computer Science yo'nalishiga kirgan talabaga 1 semester "Hello world" kabi dasturlarni yozdirish universitet kontrakt to'lovini oqlamay qolsa kerak. Boshqa yo'nalishlar ham mustasno emas.
2. Yosh ilmiy izlanuvchularga qo'yiladigan talab va yo'nalishlar qanday bo'lishi kerak?
AI hatto matematikada ham qo'li uzayib qoldi. Bizning research groupda 3 ta bir necha o'n yildan beri ochiq masalalarni AI yordamida ishlashdi (customized Agentic AI system). Vaholanki, bizda top-top-tier matematiklar yo'q (ML group with general math training). AI qila olmaydigan vazifalarni aktiv ravishda aniqlash va shu vazifalarda human-AI hamkorligi bilan ilmiy yangiliklar yaratish muhim ko'nikmaga aylanadi. To'g'ri savol so'ray olish har doimgidanda dolzarb.
https://youtu.be/OOMx2BHHWtE?is=sdp9viIA05naJywr
@lazizabdullaev | 688 |
| 9 | Time 100 AI 2026
2026-yilda AI da hissasi eng katta odamlar shular emish, men 101-o'rinda ekanman:
https://time.com/collection/time100-ai/2026/
@lazizabdullaev | 818 |
| 10 |  Singapur TV da nima yangiliklar ko'rsatarkin deb yoqsam, O'zbekiston bilan do'stmiz deb gapirib turgan ekan yanga.
@lazizabdullaev | 696 |
| 11 |  "Koinotda Data Centerlar qurilyotgan bir paytda", HR'larga nima bo'lyapti?
@lazizabdullaev | 823 |
| 12 |  Gitta Kutyniok. "The mathematics of artificial intelligence".
7 mathematical problems in AI proposed at International Congress of Mathematicians (ICM) 2022
https://ems.press/content/book-chapter-files/33332
@lazizabdullaev | 1 174 |
| 13 | Academia:
— We have applied our method at large scale.
— How large?
— 7 billion parameters.
Industrial AI lab:
— We have a tiny toy Qwen model deployed locally.
— How small?
— 70 billion parameters.
This financial imbalance pushes academic researchers to be maximally innovative and grounded in first principles, while granting industrial researchers the luxury of brute force with some heuristic guidance. With some exceptions, of course.
- Self-attention is brute force.
- Transformers are brute force.
- End-to-end backpropagation is brute force. Intelligent systems, like humans, learn continually in a forward fashion.
- ...
We cannot build them from first principles of natural sciences. An infamous quote of Richard Feynman says: "I don't understand what I can't build".
And it's not necessarily a bad thing. Remember, the first computers used to be room-sized. This also means there are still big opportunities to contribute to the development theoretically.
So far, what we call *Artificial Intelligence* might mostly be a knowledge retriever with a gigantic knowledge base.
A frequently asked question is whether AI will take over our jobs. If a job is primarily memorization-based, then most likely, yes. Being skilled will remain important. 70 hunar ham oz deb bekorga aytishmaganda)
But what about coding and mathematics, where we have seen many impressive results? Frontier labs have focused heavily on math and coding partly because they are verifiable and well-posed as evaluation tasks. Even though they are creativity-based tasks, memorization is a big part of the solution implementation process anyway. For many other tasks, the situation may be silently worse.
Soon, we will likely stop referring to LLMs, trained end-to-end via ultra-expensive backpropagation, as a realization of intelligence.
So, keep an open and adaptive mind.
Just my high entropy thoughts on my way to work to train a transformer with backpropagation...
@lazizabdullaev | 1 261 |
| 14 |  Positively surprised to see my name here in the workshop 🫡
@lazizabdullaev | 1 087 |
| 15 |  Ma'ruzamning nomiga ChatGPT yordam bergani bilinib qolibdi
@lazizabdullaev | 1 153 |
| 16 | Conversation with a new Chinese intern at Sea:
- hey, what's your name?
- La-ziiz (slowly so that he gets it). What's yours?
- Shishishirushenshi. Nice to meet you
Unfair game
@lazizabdullaev | 1 402 |
| 17 |  Matematik conjecture hal qilinishi zerikarli yangiliklar qatoriga qo'shilib qoladi deb hech o'ylamagan edim...
@lazizabdullaev | 1 371 |
| 18 | 2016: Soqqani bosish kerak
2026: Tokenni bosish kerak
@lazizabdullaev | 1 553 |
| 19 |  2016: Siz uchun Baxt nima?
2026: Siz uchun AGI nima?
@lazizabdullaev | 1 638 |
| 20 |  Artificial intelligence seems to have developed as an individual subject.
A comment on 2026 ICTP Dirac medalists explicitly mentions AI as an application domain:
The work of the 2026 Dirac Medallists has contributed to establishing theoretical physics, and particularly statistical mechanics, as a powerful framework to tackle a very broad range of questions that go far beyond its traditional domain of interest to encompass biology, computer science and artificial intelligence
https://www.ictp.it/news/2026/8/ictp-announces-2026-dirac-medal-recipients
@lazizabdullaev | 1 540 |
