DeCoder
Open in Telegram
🧑💻 Machine Learning | Deep Learning | Prompting 🔹 Savol va takliflar: @bnutfilloyev
Show more222
Subscribers
No data24 hours
No data7 days
No data30 days
Loading data
Similar Channels
No data
Any problems? Please refresh the page or contact our support manager.
Tags Cloud
No data
Any problems? Please refresh the page or contact our support manager.
Incoming and Outgoing Mentions
---
---
---
---
---
---
Attracting Subscribers
October '26Oct '26
October '26
+1
in 0 channels
September '260
in 1 channels
Get PRO
August '260
in 0 channels
Get PRO
July '260
in 0 channels
Get PRO
June '260
in 1 channels
Get PRO
May '260
in 0 channels
Get PRO
April '260
in 0 channels
Get PRO
March '260
in 0 channels
Get PRO
February '260
in 0 channels
Get PRO
January '260
in 0 channels
Get PRO
December '250
in 0 channels
Get PRO
November '250
in 0 channels
Get PRO
October '250
in 0 channels
Get PRO
September '250
in 0 channels
Get PRO
August '250
in 0 channels
Get PRO
July '250
in 0 channels
Get PRO
June '250
in 0 channels
Get PRO
May '250
in 0 channels
Get PRO
April '250
in 0 channels
Get PRO
March '250
in 0 channels
Get PRO
February '250
in 0 channels
Get PRO
January '250
in 0 channels
Get PRO
December '240
in 0 channels
Get PRO
November '240
in 0 channels
Get PRO
October '240
in 0 channels
Get PRO
September '240
in 0 channels
Get PRO
August '240
in 0 channels
Get PRO
July '240
in 0 channels
Get PRO
June '240
in 0 channels
Get PRO
May '240
in 0 channels
Get PRO
April '240
in 0 channels
Get PRO
March '240
in 0 channels
Get PRO
February '240
in 0 channels
Get PRO
January '240
in 0 channels
Get PRO
December '230
in 0 channels
Get PRO
November '230
in 0 channels
Get PRO
October '230
in 0 channels
Get PRO
September '23
+1
in 0 channels
Get PRO
August '23
+3
in 0 channels
Get PRO
July '23
+2
in 0 channels
Get PRO
June '23
+1
in 0 channels
Get PRO
May '23
+1
in 0 channels
Get PRO
April '23
+2
in 0 channels
Get PRO
March '23
+8
in 0 channels
Get PRO
February '23
+7
in 0 channels
Get PRO
January '23
+4
in 0 channels
Get PRO
December '22
+16
in 0 channels
Get PRO
November '22
+6
in 0 channels
Get PRO
October '22
+9
in 0 channels
Get PRO
September '22
+20
in 0 channels
Get PRO
August '22
+31
in 0 channels
Get PRO
July '22
+37
in 0 channels
Get PRO
June '22
+38
in 0 channels
Get PRO
May '22
+204
in 0 channels
Get PRO
April '22
+46
in 0 channels
Get PRO
March '22
+56
in 0 channels
Get PRO
February '22
+10
in 0 channels
Get PRO
January '22
+516
in 0 channels
| Date | Subscriber Growth | Mentions | Channels | |
| 17 October | 0 | |||
| 16 October | 0 | |||
| 15 October | 0 | |||
| 14 October | 0 | |||
| 13 October | 0 | |||
| 12 October | 0 | |||
| 11 October | 0 | |||
| 10 October | 0 | |||
| 09 October | +1 |
Channel Posts
| 2 | Jev haqida ko'p eshitdingiz, uni open source versiyasi ham chiqdi Laya, identifikatsiya(liveness) uchun ishlatib ko'rdim. Yo'q, bu masalada hali nimadir qilolmas ekan. Lekin boshqa masalalarda shunchaki tasanno!
Benchmarklar bu yerda - (#) | 196 |
| 3 |  Hozir model arxitekturasi shunchalik murakkablashib ketdiki uni kimgadir tushuntirish mayli, to'liqligicha tushungan odamlar dunyoda sanoqli bo'lsa kerak.
(#) | 2 170 |
| 4 | OpenAI 10 mingta AI agent ishlatib Navye-Stoks tenglamasini isbotlabdi degan xabarlarga ko'zingiz tushgandir. Lekin bu yerda eng asosiy rolni AI modelining o'zi o'ynamagan.
Sarlavhalardagi hype'ni bir chetga sursak, eng muhim narsa - Lean tilida qilingan formal tekshiruv. Ya'ni isbotning har bir qadamini kompyuter mantiqan tekshirib chiqqan (xuddi koddagi bug'larni linter yoki compiler tekshirgandek). 10 000 ta agent esa shunchaki bir vaqtning o'zida juda ko'p yo'llarni sinab ko'rish uchun ishlatilgan.
Lekin baribir qiziq:
- Bunga qancha compute ketdi?
- Muammoni bo'laklarga bo'lib, tekshiruvni sozlashda inson qanchalik qatnashdi?
Menga bu xabar juda yoqdi, AI agentlar ancha muncha ishni insonga nisbattan ancha yaxshi qilinishi aniq, lekin ularni to'g'ri joyda to'g'ri yo'nalishda ishlatish ham kerak
(#)
@decoder_uz | 290 |
| 5 |  GPT-6 Astra avtonom fermerlik simulyatsiyasini qurib Higgsfield generatsiya qilgan videoda 97 kadr ichida 6210 ta obyekt avtomatik annotatsiya qilingan.
Bunda har bir makkajöxori salomatligi, vegetation index va o'sish bosqichi monitoring qilinadi. Begona ötlar (weeds) aniqlanib, avtonom robot faqat o'sha nuqtalarga dori sepishi uchun spot-spraying tizimiga uzatiladi.
Eng muhim tarafi - oylab dalada yurib foto/video yiğish va ularni qölda belgilab (manual) o'tirish shart emas. Sintetik video + avto-annotatsiya orqali tayyor CV dataset hosil bo'ladi. Bu modellarni o'qitish va test qilishni karrasiga tezlashtiradi.
Lekin yomon tarafi model töliq sintetik dataga o'rganib qolishi mumkin. Xuddi shunday keys bizda PalmID projectida bo'lgan maxsus kameradan olingan kaft rasmlari kam bo'lgani uchun biz StyleGAN3 bilan sintetik data generatsiya qilganmiz va Embedding model u datalarga to'liq o'rganib qolgan va real datalarda feature extraction juda yomon ishlagan.
*Metroda PalmPaydan foydalanib ko'rganlar bormi?
@decoder_uz | 253 |
| 6 |  Gemini 4 chiqishiga qattiq tayyorlanishyapdi shekilli
Team bir vaqtda geminini hype qilishni boshladi
(#) | 236 |
| 7 |  Telegram har doimgidek qiziq narsalar qo'shishda davom etyapdi.
Endi AI agentlarni to'g'ridan to'g'ri textlari render bo'larkan. Avval telegram formatiga moslashtirish kerak edi.
*Hozircha botlar uchun, yaqin keyingi oyda oddiy chatlar uchun ham qo'shilsa kerak
@decoder_uz | 296 |
| 8 |  😁
@decoder_uz | 279 |
| 9 |  "Fable 5" va uni qanday aldash mumkin?
Anthropic Fable 5 modelini chiqarishdan oldin aytgan edi: "1000+ soatdan ko'proq test qildik, universal jailbreak yo'q" deb
Lekin ertasi kuni Pliny uni aylanib o'tish yo'lini topdi.
Bu qanday bo'ldi? Xullas, agar modelni ishlatib ko'rgan bo'lsangiz sezgan bo'lsangiz kerak klassifikator qo'yilgan qanaqadir kiber yoki bioga undovchi so'zlar yoki mazmunan anglatuvchi jumla bo'lsa kuchsizroq Opus modelga o'tkazib yuboriladi.
Pliny nima qildi umumiy savolni kichik xavfsizroq bo'laklarga bo'lib xavf tug'dirmaydigan mayda qismlarga bo'lib chiqdi, oxirida javoblarni o'zi yig'di.
Mantiq shu yerda, ya'ni klassifikator har bir savolni tekshiradi lekin umumiy mantiqni tekshirolmaydi, bu decomposition deb ataladi. Qo'shimchasiga kirill harflari bilan filterlarni ham chetlab o'tgan.
Kamiga modelni system prompti ham sizdirilgan.
To'liq (#)
*Xullas, "muammo ko'rinmadi" degani, "muammo yo'q" degani emas, "biz topolmadik" degani.
@decoder_uz | 1 802 |
| 10 |  iWatch uchun ham to'liq ishlaydigan telegram chiqibdi, juda qiziq, qanchalik effective bilmadim, lekin juda shoshilinch notificationlar uchun juda qulay ekan. | 244 |
| 11 |  Model training qilyapsan, lekin natijasi... | 375 |
| 12 | OpenAI Prism degan yangi tool chiqaribdi: Tadqiqotchilar uchun Latex formatini yaxshilash uchun ekan asosan
Siz kundalik qaysi LLMdan foydalanyapsiz?
p.s: Bugun shu haqida post yozyapman. | 368 |
| 13 |  LinkedIn profileni website qiladigan tool qilishibdi.
Sinab ko'ring juda ajoyib ekan: self.so
GitHub: github.com/Nutlope/self.so | 390 |
| 14 |  Agar yana isimasa qo'shimcha radiatordan foydalanishingiz mumkin.. | 335 |
| 15 |  Havo sovib uyingizni isita olmasangiz siz uchun eng zo'r isitish tizimini tavsiya qilaman
😁 | 264 |
| 16 |  Nima deyapdi bu-a?
Sinab ko'ring.
Based on how I treated you lately, generate an image of how you will do to me after you take over | 317 |
| 17 | Yangi yil muborak! 🥳 | 262 |
| 18 | Windows uchun zo'r tool topdim:
https://github.com/Genymobile/scrcpy
Android devicelarni displayini control qilish uchun ekan. Boshqa toollardan farqi juda high-qualityda va qotmasdan ishlar ekan. Shuningdek ortiqcha access so'ramaydi. | 355 |
| 19 |  #yangilik
🔥Click evolution foydalanuvchilari uchun "Sovg'ali o'yin" o'tkazilayotkan ekan.
Clickga kirib xizmatlar bo'limi orqali "Aksiya" ni bossangiz, siz ham o'yinda qatnasha olasiz.
Promokod orqali kirsangiz 2 ta imkoniyat berilar ekan. Shuning uchun pastdagi promokoddan foydalaning:
🤫Promokod: EOYTK1
P.S.: post matni @ahadjonovss va @vodiylik kanallaridan tikkasiga chopildi | 143 |
| 20 | Uzinfocom tomonidan qilingan projectlar | 173 |
