BeNN
Kanalga Telegram’da o‘tish
From simple ML algorithms to Neural Networks and Transformers — and from Number Theory to Topology, Cosmology to QED — dive into the world where code meets the cosmos.👨💻🌌 For Any Questions @benasphy
Ko'proq ko'rsatishMamlakat belgilanmaganTexnologiyalar & Aralashmalar38 685
1 042
Obunachilar
-324 soatlar
-107 kun
-930 kun
Ma'lumot yuklanmoqda...
O'xshash kanallar
Taglar buluti
Kirish va chiqish esdaliklari
---
---
---
---
---
---
Obunachilarni jalb qilish
Sentabr '26
Sentabr '26
+1
0 kanalda
Avgust '26
+22
1 kanalda
Get PRO
Iyul '26
+7
0 kanalda
Get PRO
Iyun '26
+8
0 kanalda
Get PRO
May '26
+16
1 kanalda
Get PRO
Aprel '26
+87
5 kanalda
Get PRO
Mart '26
+6
0 kanalda
Get PRO
Fevral '26
+7
2 kanalda
Get PRO
Yanvar '26
+11
0 kanalda
Get PRO
Dekabr '25
+14
3 kanalda
Get PRO
Noyabr '25
+30
3 kanalda
Get PRO
Oktabr '25
+35
9 kanalda
Get PRO
Sentabr '25
+15
0 kanalda
Get PRO
Avgust '25
+13
1 kanalda
Get PRO
Iyul '25
+109
6 kanalda
Get PRO
Iyun '25
+6
1 kanalda
Get PRO
May '250
1 kanalda
Get PRO
Aprel '250
0 kanalda
Get PRO
Mart '250
0 kanalda
Get PRO
Fevral '250
0 kanalda
Get PRO
Yanvar '250
0 kanalda
Get PRO
Dekabr '240
0 kanalda
Get PRO
Noyabr '24
+2
0 kanalda
Get PRO
Oktabr '24
+5
0 kanalda
Get PRO
Sentabr '24
+3
0 kanalda
Get PRO
Avgust '24
+7
0 kanalda
Get PRO
Iyul '240
0 kanalda
Get PRO
Iyun '240
0 kanalda
Get PRO
May '240
0 kanalda
Get PRO
Aprel '240
0 kanalda
Get PRO
Mart '240
0 kanalda
Get PRO
Fevral '240
0 kanalda
Get PRO
Yanvar '24
+1
0 kanalda
Get PRO
Dekabr '23
+6
0 kanalda
Get PRO
Noyabr '23
+10
0 kanalda
Get PRO
Oktabr '23
+7
0 kanalda
Get PRO
Sentabr '23
+20
0 kanalda
Get PRO
Avgust '23
+18
0 kanalda
Get PRO
Iyul '23
+19
0 kanalda
Get PRO
Iyun '23
+10
0 kanalda
Get PRO
May '23
+24
0 kanalda
Get PRO
Aprel '23
+8
0 kanalda
Get PRO
Mart '23
+9
0 kanalda
Get PRO
Fevral '23
+10
0 kanalda
Get PRO
Yanvar '23
+18
0 kanalda
Get PRO
Dekabr '22
+23
0 kanalda
Get PRO
Noyabr '22
+51
0 kanalda
Get PRO
Oktabr '22
+74
0 kanalda
Get PRO
Sentabr '22
+97
0 kanalda
Get PRO
Avgust '22
+475
0 kanalda
Get PRO
Iyul '22
+45
0 kanalda
Get PRO
Iyun '22
+41
0 kanalda
Get PRO
May '22
+59
0 kanalda
Get PRO
Aprel '22
+45
0 kanalda
Get PRO
Mart '22
+57
0 kanalda
Get PRO
Fevral '22
+49
0 kanalda
Get PRO
Yanvar '22
+138
0 kanalda
Get PRO
Dekabr '21
+89
0 kanalda
Get PRO
Noyabr '21
+99
0 kanalda
Get PRO
Oktabr '21
+229
0 kanalda
Get PRO
Sentabr '21
+148
0 kanalda
Get PRO
Avgust '21
+143
0 kanalda
Get PRO
Iyul '21
+106
0 kanalda
Get PRO
Iyun '21
+109
0 kanalda
Get PRO
May '21
+171
0 kanalda
Get PRO
Aprel '21
+73
0 kanalda
Get PRO
Mart '21
+34
0 kanalda
Get PRO
Fevral '21
+72
0 kanalda
Get PRO
Yanvar '21
+281
0 kanalda
Get PRO
Dekabr '20
+1 297
0 kanalda
| Sana | Obunachilarni jalb qilish | Esdaliklar | Kanallar | |
| 05 Sentabr | 0 | |||
| 04 Sentabr | 0 | |||
| 03 Sentabr | +1 | |||
| 02 Sentabr | 0 | |||
| 01 Sentabr | 0 |
Kanal postlari
This is the model that OpenAI President Greg Brockman stated they're close to AGI. another reason was according to various tweets I just saw It has scored 99.9% on ARC-AGI-3. However, some studies from ARC Prize says they're just optimizing the model to achieve that and the real percentage is 63%.
I still think we are kinda little far from AGI, but it is just the time issue. I still insist on 2030 like I said before and let's see👀
| 2 | I hope it meets expectations. | 76 |
| 3 |  🤖 GPT-6 Rolls Out to Codex
Pro users got first access in Codex and ChatGPT.
An official prompting guide for the new model is also out.
📊@tech | 72 |
| 4 |  Lol I thought 2019 was Yesterday 😭 | 110 |
| 5 | https://youtu.be/g7FmEc5GLWs?si=pR0HjjQ9qQ2Mkibe | 92 |
| 6 | because tg can't render LaTeX try to understand it from this pic | 143 |
| 7 |  Matn yo'q... | 140 |
| 8 | Cont'd
3. Positional Encoding Degradation - Transformers are naturally "order-blind", they use Positional Encodings (like RoPE) to track word order.
The Problem: Models are trained on fixed context lengths (e.g., 8k tokens).
Impact: Feeding input past that trained length distorts the positional geometry, causing the model to lose logic or hallucinate wildly.
Modern AI fixes this problem using stuff like FlashAttention, Rotary Embeddings (RoPE) Scaling, State Space Models (Mamba/Hybrids) | 127 |
| 9 | Have you wondered why LLMs u use will say input token is limited(technical word for it is, LLM Context Windows Are Limited)? The reason you can copy past the whole book on it?
Well u may think it is only hardware problem, but it is mainly baked directly into the Transformer architecture.
1. The first problem is the O(n^2) Self-Attention Bottleneck - The core motor of a Transformer is the Self-Attention mechanism. Every input token calculates its similarity score relative to every other token in the sequence to compute attention weights using the below formula.
For sequence length N, computing the dot product QK^T requires O(n^2) operations. This means Doubling your prompt length quadruples the attention compute and memory overhead.
2. The KV Cache Scaling Problem - During generation (inference), models save the calculated Key (K) and Value (V) vectors of all previous tokens in memory so they don't have to recompute them at each step.
While attention computation is optimized, the KV Cache footprint scales linearly O(N) per request, so For massive prompts (e.g., 128k+ tokens), the cache alone can consume dozens of GBs of VRAM per user, causing GPUs to run out of memory (OOM) before compute limits are even hit. | 136 |
| 10 |  +1Twitter is mad place 😭😭now they're arguing between Young and old Messi | 185 |
| 11 | Whenever Friday comes this is the song that keeps playing on ma mind 😭🤣 | 206 |
| 12 | https://youtu.be/U6n2NcJ7rLc?si=MNl7VI3psyK2haUO | 201 |
| 13 | On the topic of the necessity of AGI.... It's plausible to think that if we can build Narrow intelligence that's specialized let's say on Oncology to solve Cancer and the same on physics to solve Theory of Everything.
However, the thing is To solve Cancer the model need to know not only oncology, but Chemistry, Genetics, patient data and the same with ToE, it needs to know about Quantum Gravity then whenever it tackles one issue it may need to formulate some very new mathematical formula that hasn't been there.
In such cases AGI could solve it as it can understand vast areas. It doesn't mean it needs to know about Cartoon Network. Narrow AI that specializes in Oncology doesn't know that much about chemistry so won't make much progress that's why We need a General Scientist like human Being. If u again think why we don't add this feature of chemistry and then others.... Surprise 🙌🙌The moment you keep bolting on more and more domains until the system can actually invent, transfer knowledge, and reason across them…
you’re no longer building Narrow AI, but AGI! | 184 |
| 14 |  Matn yo'q... | 165 |
| 15 | JUST IN: OpenAI says it will achieve AGI by the end of this year.
@WatcherGuru | 155 |
| 16 | If you are in high school, whenever u get chance please study Astronomy, Physics, General Science, History, Philosophy, Religion and other bunch of stuff! You may think that you will read such stuff later on in life once you settle, but trust me that's not gonna work like that😁
Again For everybody here what I always say is to always stay curious and learn anything that you wanna understand. If you think u can read such stuff later on once you have good job or u settled then idk honestly, I personally read those stuff not just to read, but to apply to my real life.
Good example would be before I read and understood some of Astrophysics concepts, I thought I was some Amazing creature and if I will have good job at end(like joining big companies Google and stuff) then it will be amazing. However, After Reading it I understood that I am nothing in this whole vast Universe! If I leave this world tomorrow nothing is gonna happen it will continue from Universe perspective. You are just in a very little Pale blue dot and You are not even noticeable in Grand scheme of Universe! I got Humbled because of this and most of all my inner sense has got peace and still crave for more, but now at least I happily crave for it.
"DON'T DIE BEFORE YOU'RE DEAD" | 322 |
| 17 | So far from the LLM that I tried
• Frontend - Gemini 3.5,6,7
• Basic to Intermediate coding tasks - Kimi K3
• High level Research - Claude Fable 5
• High level Coding - Claude Opus 5, GPT 5.6 SOL PRO
lol this are some of it and I want to try out also Deepseek, but I am not sure where to use. Another thing is be careful with Fable 5, Claude Opus 5 and GPT 5.6 SOL PRO... They are really expensive.
Most of all always check every code and changes that has been done to your code don't accept what they do.
@BeNN_Pi | 260 |
| 18 |  Enawra | Ethiopia's 1st Peer-to-Peer Language Exchange Platform
You teach, and you learn. Casually.
Local languages and International languages.
All for the price of less than 1$
Here: @enawrax_bot
Disclaimer: If you face any issues contact the username in the bot's description | 251 |
| 19 |  Here's the Result | 257 |
| 20 | I was Listening to Brian Greene Podcast and remembered one cool thing when I was into Astrophysics, which is traveling to the future lol!
The idea is like since Time is not Absolute(f you think time for you and someone traveling at high speed passes at the same rate, then you better start researching Special Relativity, you’ll enjoy it.)
I wanted to know At what speed would I need to travel so that, for every 1 year I experience, 1,000,000 years pass on Earth? It is not like u travelling there and coming back, it is more of you will only age 1 years while earth ages 1,000,000 years.
If you have watched Interstellar you may get the idea(Even though the interstellar uses the idea of General Relativity) so I decided to do the math and I’ve attached my paper calculation below with the result. | 260 |
