DeepSeek, nima u? AI olamida katta ta’sirga ega bo’lgan voqelik...

DeepSeek – Xitoyning 2023-yil tashkil topgan startup kompaniyasi tomonidan yaratilgan yangi AI model va bu model hozirda mavjud OpenAI modellaridan yaxshiroq va arzonroq. Agar siz kuchli AI modellarini yaratmoqchi bo‘lsangiz, sizga kuchli AI chiplar kerak bo‘ladi. Aksriyat 'advanced' AI (GPU) chiplarni tayyorlaydigan kompaniya bu Amerikaning Nvidia kompaniyasi. Hozirda bu chiplar juda muhim bo‘lgani uchun, AQSh hukumati Nvidia’ga ushbu chiplarni xitoylik kompaniyalarga sotishni taqiqlagan (chunki AI sohasidagi hukmronlik o’ta muhim).

Savol tug’iladi - ular qanday qilib OpenAI’dan ham kuchliroq model yaratishga muvaffaq bo‘lishdi?

Qiziq joyi shu. Ularda juda cheklangan miqdorda Nvidianing H800 chiplari bo’lgan (ba’zi expertlar bunga qo’shilmayapti) va bu miqdor kuchli model yaratish uchun bu yetarli emas. Lekin ular ajoyib katta texnik innovatsiyalar qilishdi - AI modellarini o’ta samarali o‘rgatish ‘training’ qilish yo’lini topishdi. Natijada, OpenAI, Microsoft yoki Google kabi kompaniyalar ishlatgan resurslarning , masalan, atigi yuzdan bir qismini ishlatish orqali yaxshiroq AI model yaratishga muvaffaq bo‘lishdi va o’ta arzonga. Ular $6 mln dollardan kamroq sarflashgan bo’lsa, bu qiymat boshqalarda milliardlab dollarga teng!

Ammo bilasizmi, yanada qiziq joyi nima?
Bu AQShning ko‘plab kompaniyalari uchun katta zarba bo‘ldi. Chunki, nafaqat ular o‘zlarining ajoyib AI modellarini yaratishdi, balki uni ochiq manba (open-source) qilib qo‘ydilar. Ya’ni, ular butun dunyoga: "Biz buni qanday qilganimizni nafaqat aytamiz, balki AI modellarining kodini ham beramiz, xohlagan narsangizni qilishingiz mumkin," deb e’lon qilishdi. Bu esa OpenAI’ning ko‘plab modellarining qiymatini sezilarli darajada tushirib yubordi, chunki endi undan yaxshiroq va bepul alternativ bor. Bu hozirda Xitoyning innovatsion kuchi!

Doim to’g’ri emas, lekin ba’zida 'eng yaxshi innovatsiyalar qattiq zaruriyat paytida kelib chiqadi'

@saidovhakimjon

Bilaman, bu yerda ko'pchiliklaringiz Master yoki PhD programmalariga topshirish niyatlaringiz bor. Ko'pgina top grad school programmalarda yaxshi GRE bo'lishi talabi bor. Abdurashid tomonidan tashkil qilinayotgan workshopga qatnashishni tavsiya qilaman. Juda foydali bo'ladi sizlarga deb o'ylayman.

@saidovhakimjon

GRE 101: Your Roadmap to Grad School Success

Let's face it - staring at GRE practice questions can feel like reading hieroglyphics, especally the Verbal section.

"Wait, what's a 'cacophony'?"; "Why does this geometry question look nothing like what I learned?"; "How am I supposed to write a 400-word essay in English in just 30 minutes?"... Sound familiar?

I've been there. And next Tuesday, I'm doing something special - hosting a FREE workshop where I'll:

-Address your questions
- Break down what actually matters on the GRE (hint: not everything in those thick prep books!)
- Show you the patterns in Quant questions that most students miss
- Reveal my vocabulary learning system that actually sticks
- Share time-tested strategies that have helped my students boost their scores by 10+ points

Best part? You won't have to sit through a boring lecture. We'll solve real GRE questions together, dissect them, and you'll leave knowing exactly how to tackle them on test day.

📅Date: January 21st
🕘Time: 9:00pm Tashkent Time
📌Channel: @GradSchoolAdmissions
Share this post with your friends!

Hech nima davlatni ilmchalik rivojlantirmaydi va dunyoga tanitmaydi.

Qochirimni yaxshi koradigan rivojlangan davlatdagi bir odamdan sorashdi, sizlar bu rivojlanishga qanday erishdingiz? U javob berdi:
-biz bu holatga kelish uchun sportga katta mablagajratdik, shunga parallel ravishda ilmga sportchalik e'tibor qaratmadik, jamiyatga sport ortidan boyib ketilishini propoganda qildik, hatto utyukni tokka tiqqanda ham sportchilar chiqib aqllilik qilib ilm bilan sportni solishtirib ilm qilish sportdan oson ekanligini aytaverdi. Shunda xalq ichida kuchli sportchilar kopaydi. Bu ishlarni va sportni kuchini qarangki bu sportchilar bir joyga yigilib har xil ixtirolar qilib boshlashdi, ular atom bombasi yaratdi, robortlar yaratishdi, uchuvchisiz uchuvchi uchoqlar yaratishdi va hokazo. Shular ortidan davlatimiz juda boyib ketdi, mana ozim ham millionerman. Gapirsam gap kop lekin hozir universitetda matematikadan darsim bor shunga bormasam bolmaydi, sizlar esa men aytgan yoldan boravinglar hammasi zor bo`lib ketadi.

Bu gaplarim bilan hech nima demoqchi emasman!

Mayli aytmoqchiman: sportga e'tibor beraverish kerak lekin hamma narsa ilmga bolgan katta e'tibordan keyin bolishi kerak.

@karimraximov

Yuqoridagiga qo'shimcha, bu professorni shaxsan taniyman - juda yaxshi mentor PhD uchun, yo'nalishi ham juda qiziq va muhim. PhD ga open positioni bor ekan, qarab ko'ringlar:

Info: "Can you imagine actually “seeing” the complex interface between a nanostructure and a molecule?

Application deadline: Aug 31, 2024.

You will break new ground at the absolute forefront of what is possible. You will advance (cryo) electron microscopy imaging & spectroscopy techniques representing a critical avenue for refining our understanding of quantum dot (QD) functionalization, particularly in the area of surface ligand engineering tailored to imaging complex molecular systems, but also in (self)assembly of coupled QD arrays for intriguing electronic and photonic properties.

The project is part of an interdisciplinary research collaboration at DTU Nanolab, in close collaboration with Prof. Heiner Friedrich at Eindhoven University of Technology (NL). The group at TU/e utilizes functionalized quantum dots as probes to ‘mark’ macromolecules for comprehensive correlative dynamic studies of interactive polymeric materials, aiming at (bio)engineering and (bio)medical applications. You will focus on understanding the complex interaction between quantum dots and organic ligands to functionalize them for various applications, here as biomarkers to label specific structures in complex molecular systems. Link"

@saidovhakimjon

DTU Candidate Site

PhD Scholarship in Structural Characterization of Functionalized Quantum Dots – DTU Nanolab

Can you imagine looking at the interface between a quantum dot and a molecule? In collaboration with TU Eindhoven (NL), you will push the boundaries of electron microscopy. You will advance the design and understanding of quantum dots for imaging of complex…