Alex

AI Agents, Coding, AI Tools & Tech in general

16 y/o

[Yuchen Jin @Yuchenj_UW]

AI won’t erase the gap between people.

YouTube, Coursera, and Internet made world-class knowledge free. Most people still chose TikTok.

AI is no different. Everyone gets Claude Code, Codex, and ChatGPT. The best will use them better and compound faster.

AI isn’t an equalizer. It’s a multiplier.

[Alex @krab75292]

@_avichawla Which method is the best for fine-tuning a model for OCR on a custom dataset?

[Alex @krab75292]

Could Ox Alpha be a new open source model from OpenAI?

[Alex @krab75292]

@ravikiran_dev7 just posting obvious stuff

"Could bring stronger performance on complex multi-step tasks and autonomous workflows"

like no way I thought the performance of Opus 5.1 will be worse than of Opus 5

[Alex @krab75292]

Ox Alpha has 100T tokens/day capacity according to OpenCode.

Even at $0.50 per 1M tokens + 80% margin that’s $10M/day if fully used ($70M/week).

Either it’s a tiny cheap model or a US company. No Chinese lab could afford this — or has the compute to serve a big model for free

[Alex @krab75292]

just 5 months ago sota was under 1% on Arc ARC-AGI-3

[NVIDIA AI @NVIDIAAI]

Our general-purpose coding agent just scored 100% on the ARC-AGI-3 interactive reasoning benchmark.

NVIDIA AVO completed all 183 levels across all 25 public environments, figuring out what to do with no instructions, explicit rules, or stated goals.

[Devin Desktop @devindesktop]

GPT-5.6 Sol is now 70% off in Devin Desktop and Devin CLI, through October 3.

On FrontierCode 1.1, Sol achieves top-tier results. With this promo, it's one of the most cost-effective frontier models you can run in Devin.

[Alex @krab75292]

@AnonymerNutze12 @TheGeorgePu https://t.co/BMRaP19bzj
In this post you can see that Qwen3.8 27B fails the test on xhigh, while it passes it on no-think or low-think.

Of course there probarly are also areas where xhigh performs better.

[Lotto @LottoLabs]

This is the follow up article to then CRUDbench task that nonthinking aced, but stumped xhigh.

Low and medium passed but poorly and with more time and tokens.

[Alex @krab75292]

@kimmonismus It would be better if they dropped an open source model.

[Alex @krab75292]

@weswinder It got rebranded to Devin Desktop.

[Alex @krab75292]

What tools and skills do you use for ai coding?

[Alex @krab75292]

How many of you vibe code but don’t make any money?

[Alex @krab75292]

@LottoLabs Qwen thinks too much for its own good

[Alex @krab75292]

@YottaMindset @LottoLabs True

[Alex @krab75292]

@alexeixbt only if you use it

[Alex @krab75292]

@danellisona Places where human interaction matters, like therapy.

It can’t replace the human part.

[Alex @krab75292]

Every time AI crosses a line we said meant “real intelligence”, we just redraw the line.
Text generation? “Just autocomplete”.

If AGI is ever achieved, it will change the AI space so fast you will know it’s AGI.

[Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 @elder_plinius]

honestly wild that no one has formally declared “AGI” yet…

[Qwen @Alibaba_Qwen]

We promised open weights for Qwen3.8. Now, time to meet them! 🎉

⚡ Qwen3.8-27B:
- A native multimodal dense model. With just 27B parameters, it outperforms Qwen3.7-Plus overall and shines in real-world coding & office workflows.
- 262K native context, easily extendable to 1M tokens via YaRN.
- Built for builders. Highly efficient, high-quality, and licensed under Apache 2.0.

🚀 The open weights for Qwen3.8-2.4T-A95B (Max-level) have also been released recently.

Whether you're shipping lightweight applications with Qwen3.8-27B locally or building agents with Qwen3.8-2.4T-A95B, they're yours now!

Download, deploy, and build something we haven't imagined yet. 👀👇
- Hugging Face：
https://t.co/4kaAcqYEVj
- ModelScope：
https://t.co/eRIMZCGkhC