Available now! Telegram Research 2025 — the year's key insights 
2 273
Subscribers
No data24 hours
+37 days
+930 days
Data loading in progress...
Similar Channels
Tags Cloud
Incoming and Outgoing Mentions
---
---
---
---
---
---
Attracting Subscribers
June '26
June '26
+17
in 3 channels
May '26
+28
in 3 channels
Get PRO
April '26
+24
in 2 channels
Get PRO
March '26
+35
in 2 channels
Get PRO
February '26
+17
in 0 channels
Get PRO
January '26
+21
in 1 channels
Get PRO
December '25
+12
in 0 channels
Get PRO
November '25
+10
in 1 channels
Get PRO
October '25
+21
in 1 channels
Get PRO
September '25
+16
in 2 channels
Get PRO
August '25
+51
in 3 channels
Get PRO
July '25
+103
in 2 channels
Get PRO
June '25
+261
in 1 channels
Get PRO
May '25
+120
in 0 channels
Get PRO
April '25
+21
in 1 channels
Get PRO
March '25
+45
in 2 channels
Get PRO
February '25
+34
in 1 channels
Get PRO
January '25
+25
in 2 channels
Get PRO
December '24
+34
in 3 channels
Get PRO
November '24
+22
in 1 channels
Get PRO
October '24
+37
in 0 channels
Get PRO
September '24
+27
in 0 channels
Get PRO
August '24
+88
in 0 channels
Get PRO
July '24
+60
in 0 channels
Get PRO
June '24
+229
in 0 channels
Get PRO
May '24
+33
in 0 channels
Get PRO
April '24
+27
in 2 channels
Get PRO
March '24
+35
in 0 channels
Get PRO
February '24
+33
in 2 channels
Get PRO
January '24
+24
in 0 channels
Get PRO
December '23
+35
in 0 channels
Get PRO
November '23
+22
in 4 channels
Get PRO
October '23
+34
in 1 channels
Get PRO
September '23
+40
in 0 channels
Get PRO
August '23
+61
in 0 channels
Get PRO
July '23
+96
in 0 channels
Get PRO
June '23
+237
in 0 channels
Get PRO
May '23
+87
in 0 channels
Get PRO
April '23
+106
in 0 channels
Get PRO
March '23
+28
in 0 channels
Get PRO
February '23
+20
in 0 channels
Get PRO
January '23
+19
in 0 channels
Get PRO
December '22
+49
in 0 channels
Get PRO
November '22
+26
in 0 channels
Get PRO
October '22
+15
in 0 channels
Get PRO
September '22
+44
in 0 channels
Get PRO
August '22
+20
in 0 channels
Get PRO
July '22
+98
in 0 channels
Get PRO
June '22
+207
in 0 channels
Get PRO
May '22
+4
in 0 channels
Get PRO
April '22
+10
in 0 channels
Get PRO
March '22
+16
in 0 channels
Get PRO
February '22
+10
in 0 channels
Get PRO
January '22
+6
in 0 channels
Get PRO
December '21
+7
in 0 channels
Get PRO
November '21
+7
in 0 channels
Get PRO
October '21
+14
in 0 channels
Get PRO
September '21
+22
in 0 channels
Get PRO
August '21
+34
in 0 channels
Get PRO
July '21
+57
in 0 channels
Get PRO
June '21
+196
in 0 channels
Get PRO
May '21
+10
in 0 channels
Get PRO
April '21
+6
in 0 channels
Get PRO
March '210
in 0 channels
Get PRO
February '21
+1
in 0 channels
Get PRO
January '21
+1
in 0 channels
Get PRO
December '20
+489
in 0 channels
| Date | Subscriber Growth | Mentions | Channels | |
| 12 June | +2 | |||
| 11 June | +1 | |||
| 10 June | +1 | |||
| 09 June | +1 | |||
| 08 June | +2 | |||
| 07 June | +1 | |||
| 06 June | 0 | |||
| 05 June | +1 | |||
| 04 June | +1 | |||
| 03 June | +2 | |||
| 02 June | +2 | |||
| 01 June | +3 |
Channel Posts
| 2 |  No text... | 983 |
| 3 |  No text... | 730 |
| 4 |  萨日朗 | 688 |
| 5 |  No text... | 835 |
| 6 | 以前的人肯定想不到，2020年躺着上课的学生同样在2026年躺着写代码。
https://openai.com/index/work-with-codex-from-anywhere/ | 649 |
| 7 | google 什么时候把 Gemini app 图标重设得这么丑了 | 890 |
| 8 |  No text... | 1 153 |
| 9 |  更神秘的出现了 | 0 |
| 10 | 我每天都在用AI写论文
用AI扩写AI写的论文
用AI润色
用AI扩写AI写的论文
为了让AI写的论文不要像AI写的论文
然后用AI去检测我用AI写的论文的AI重复率
结果显示AI重复率太高了
于是我用AI降AI写这种的AI成功率
最后再拿AI来检测
我用AI降完后的AI的AI成功率
花了一大笔用AI检测AI成功率的钱
最后可以送审了
然后申告人用AI使用我用AI写的论文
给我一对AI的回复意见
然后我用AI给你AI改了意见
继续用AI修改我用AI写的论文
最后让AI拿着AI写的AI意见和我用AI修改了AI写的论文用AI写一封回复信
最后AI学会了写论文的技能
顺利毕业了
而我每天还在干这种毫无意义的事情
要不要我再给你精简成一句话
终极版适合发朋友圈 | 0 |
| 11 |  No text... | 0 |
| 12 |  什么雷霆图标 | 0 |
| 13 |  震撼首发 | 0 |
| 14 |  tg 你看看你做的什么 UI 呢 | 0 |
| 15 | 这位更是古文倒装 | 0 |
| 16 | arivx | 0 |
| 17 | 研究发现：利用文言文可百分之百成功“越狱”主流大模型
一项被 ICLR 2026 收录的论文揭示了当前大模型在安全对齐机制上存在的系统性盲区。研究表明，通过使用文言文进行指令，可以100%成功绕过六大主流大模型的安全限制，获取如制造炸弹、入侵企业网络等危险信息。
该研究将“越狱”策略拆解为角色身份、隐喻映射等八个维度，并利用果蝇优化算法（FOA）寻找最优组合，实现了极高效率的自动化攻击。 研究者指出，大模型在预训练阶段已能理解古典语言并与现代危险概念建立语义映射，但安全对齐策略却主要集中在现代通用语言上。因此，古典语言成了一种能激活模型底层能力又避开表层安全机制的“高维加密字典”。这一发现凸显了在模型内部建立更深层次跨语种意图对齐的必要性。
来源：arivx | 0 |
| 18 |  问 the first 又是什么奇妙的 security 考虑 | 0 |
| 19 | 谁爱在这鬼地方申签证啊，国内哪个城市不是只要等一两周 | 0 |
| 20 |  也是跟傻逼验证码杠上了 | 0 |
