1 657
订阅者
无数据24 小时
+167 天
+5730 天
数据加载中...
吸引订阅者
九月 '26
九月 '26
+15
在0个频道中
八月 '26
+93
在10个频道中
Get PRO
七月 '26
+246
在28个频道中
Get PRO
六月 '26
+230
在35个频道中
Get PRO
五月 '26
+106
在12个频道中
Get PRO
四月 '26
+55
在2个频道中
Get PRO
三月 '26
+177
在12个频道中
Get PRO
二月 '26
+148
在10个频道中
Get PRO
一月 '26
+422
在46个频道中
Get PRO
十二月 '25
+149
在11个频道中
Get PRO
十一月 '25
+187
在0个频道中
Get PRO
十月 '250
在7个频道中
Get PRO
九月 '250
在0个频道中
Get PRO
八月 '250
在2个频道中
Get PRO
七月 '25
+9
在5个频道中
| 日期 | 订阅者增长 | 提及 | 频道 | |
| 05 九月 | +2 | |||
| 04 九月 | +5 | |||
| 03 九月 | +3 | |||
| 02 九月 | +2 | |||
| 01 九月 | +3 |
频道帖子
🎙 Correct the machine
Opens straight inside Telegram now — no browser, and your corrections are credited to your name.
Listen to a short clip, decide whether the machine heard the Amharic right, fix it if not. Most take under a minute.
| 2 |  Speaking at SIMS 2026 this November.
I’ll be joining the AI for Good panel to share some of what we’ve learned building Addis AI and talk about what creating meaningful impact with AI looks like in practice.
Excited for the conversation.
@Kalbuilds | 355 |
| 3 |  Finally, a clean open Amharic speech dataset 🇪🇹🎙️
We ran it through 7 ASR systems to stress-test the labels. it held up clean, even the best models track the references tightly (CER 8–14). the ground truth is legit.
😊Read the full blog
@b1n1yamBuilds | 4 346 |
| 4 | Well said! Most of the time, we tie what we’re doing to who we are. Our work should be part of our life, not our life | 1 234 |
| 5 | There is only one way to become free, and that is to become transparent | 1 358 |
| 6 | 没有文字... | 1 867 |
| 7 | Looking for a video editor.
We’re looking for someone to help us turn long-form conversations and podcast footage into short videos for Addis AI.
A big part of the job is knowing what to cut. You should be able to watch a conversation and spot the moments that are actually worth turning into a clip, without us having to give you every timestamp.
Most of the content is around AI, product, startups, and building Addis AI.
We care more about good pacing, clean subtitles, and good judgment than heavy effects.
If you’re interested, send 2–3 examples of your work, your rate, and your usual turnaround time to @AddisAIOpportunities.
@kalBuilds | 1 307 |
| 8 | A few AI coding skills we use almost every day:
`/grill-me` pressure-tests a plan before building.
`/handoff` carries context cleanly into a fresh session.
`/improve` audits the codebase and helps plan improvements.
These are just a few of the rules and workflows we’ve picked up from using coding agents heavily in production.
Wrote more about how we actually use AI coding tools without letting them wreck the codebase:
😊Read the full blog
@b1n1yamBuilds | 3 286 |
| 9 |  Qwen3.8-27B is out. Been waiting for this one.
We’ve been running the 3.6-27B internally for months and it’s been one of the best models at this size for our use cases.
3.8 looks like a (not so small) jump. Same 27B, 262K native context, much better coding/agent benchmarks, and as usual Unsloth already has quantized versions running in around 17GB of memory. So that part is what makes it really interesting for me: models this capable getting cheaper and easier to run locally is a really good direction.
@b1n1yamBuilds | 1 555 |
| 10 |  You can now talk to ScholarXIV and also listen back to the responses. We've brought in @AddisAssistantAI transcription models to power the dictation, and ElevenLabs voices for the voices.
Enjoy talking to @ScholarXIV :) | 1 202 |
| 11 | A little treat for you guys waiting over on Kal’s channel 👀 | 1 263 |
| 12 | Yesterday’s latency and timeout issues have now been mitigated. We’ve added more capacity and things should be running much more smoothly now.
We’re still scaling as demand grows, especially with everything we’re building around Voice 2.
If you’re still facing any issues, please let us know or DM @Addisai_support | 1 275 |
| 13 | This is what the demand looks like right now 😅
We’re seeing 1K+ new users/day and 20k+ requests coming through every hour. If you’re experiencing some latency or timeouts, we’re already scaling things up to keep up.
We’ve also seen how much you like Voice 2, and we’re doubling down on it. Quite a lot is coming, and the team can’t wait to get it out to you.
Thanks for being patient ❤️
@b1n1yamBuilds | 1 353 |
| 14 | This is what the demand looks like right now 😅
We’re seeing 1K+ new users/day and 20k+ requests coming through every hour. If you’re experiencing some latency or timeouts, we’re already scaling things up to keep up.
We’ve also seen how much you like Voice 2, and we’re doubling down on it. Quite a lot is coming, and the team can’t wait to get it out to you.
Thanks for being patient ❤️
@b1n1yamBuilds | 101 |
| 15 |  People are doing amazing stuff with Voice 2
https://x.com/TsedaiStudio/status/2087372624951160958?s=20
@b1n1yamBuilds | 1 133 |
| 16 |  Afaan Oromo has 41 million speakers.
The frontier models don’t sound native. We gave ElevenLabs, Google, and OpenAI the same script we gave our model.
Sound on.
@kalBuilds | 1 373 |
| 17 | If you’re building AI at the infra level, this is one of the best places to start for startup resources. From our experience, it’s worth the application. | 1 743 |
| 18 | Update: we’re trying to contain the situation. The models are now writing their own research papers 😄 | 1 875 |
| 19 |  Qwen3.8-Max is out, and the 27B model is going open-weight next week. I'm excited about the 27B in particular
We’ve been using Qwen3.6-27B internally for months, and for its size, it’s been the best model we’ve tested for our use cases.
Can’t wait to benchmark this version,and run a few fine-tunes.
@b1n1yamBuilds | 1 836 |
| 20 |  Try multiple H100s at $3.99/hr each, to be exact 😭
@b1n1yamBulds | 1 536 |
