Vin ./

ALL IN | MAXI | @Gradient_HQ

[Vin ./ @VinoisAsian]

You're working on Airdrop
I'm working on project

We are not the same

[Vin ./ @VinoisAsian]

@BennettBui1 @0xEricYang Unfortunately I’m a non phd nor master, just a rusty intern 🥲

[Commonstack @commonstack_ai]

Fraction of the bill. Same results.

Fully local, open source, works with any client.

Just > pipx install uncommon-route

https://t.co/6bJ9WhtB52

[Prasenjit @Star_Knight12]

2022 - Student
2023 - Developer
2024 - Prompt Engineer
2025 - Vibe Coder
2026 - AI Agent Babysitter
2027 - Farmer

[Vin ./ @VinoisAsian]

Trust the process, be positive, don't give up.

[Vin ./ @VinoisAsian]

Deepseek V4 available on @commonstack_ai within a day !!
The update is as fast as the throughput of the api!

[rw ./ @gradientintern]

DeepSeek V4 flash is on par with GPT 5.4 (high), the best part is that it’s much more affordable at scale:

GPT 5.4 pro vs DeepSeek V4 flash:

Input: $30/M vs $0.14/M (214x cost difference)
Output: $180/M vs $0.28/M (643x cost difference)​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​

Both at a million context, DeepSeek V4 Flash is really a bargain for intelligence.

[Hexx ./ @HexxRL]

🔬 lab works & stuff

come by to see some of the recent research works of the team.

📍 location: builder hub
🗓️ time: april 25th 1pm UTC

./ 🥼 coat on @Gradient_HQ

[Vin ./ @VinoisAsian]

@gokuoldskool How can i be like you? young rich handsome and have a great life ?

[Vin ./ @VinoisAsian]

Finally, @tryParallax getting the attention it deserves.

You can run your own too at a fraction of the cost and it's private

[rw ./ @gradientintern]

Awesome to see @tryParallax’s distributed framework for heterogeneous machines being implemented and serving up inferences!

Build and customize your own clusters for AI like never before 🤖

./ LFG @Gradient_HQ

[Vin ./ @VinoisAsian]

依照圖片我也覺得討論是蠻刺激的

[Viviennn @0xViviennn]

昨天线下活动 几乎所有人都在讨论 @NousResearch 和 @PrimeIntellect

crazy

[Vin ./ @VinoisAsian]

@0xvietnguyen job.

[Hexx ./ @HexxRL]

🤖 Gradient Live Knowledge Trivia!

Come by to join us on a 20 question trivia! See where you stack up for knowledge among Grads!

📍 Location: Builder Hub
🗓️ Time: April 11th 1PM UTC

./ @Gradient_HQ memory engine on 🧠

[Gradient @Gradient_HQ]

Our cofounder @0xEricYang sat down with @yacinelearning to walk through Echo-2’s distributed RL architecture.

Dive in to learn about async RL with distributed infra, and how we are scaling this for businesses to win in the agentic era.

[Yacine Mahdid @yacinelearning]

for those interested in distributed reinforcement learning I just finished a ~1h tutorial on the echo2 framework by @Gradient_HQ

we check:
- how to do async RL
- infra split between rollout workers and centralized learner
- interview with gradient cofounder eric yang himself!

[Vin ./ @VinoisAsian]

@gradientintern @Gradient_HQ All W !!!

[adam @adamy_ai]

china's AI giants just launched a price war over AI coding plans.

zhipu, minimax, kimi, alibaba, bytedance, tencent all competing for the same developer wallet.

entry price: ¥7.9/mo (~$1.10) for the first month.

[Vin ./ @VinoisAsian]

@hahnbeelee Do you hire a random dude from Asia? I happen to know a friend that is useless and wanted a job

[Vin ./ @VinoisAsian]

I tested a local OCR setup using:

mlx-community/GLM-OCR-bf16
mlx-vlm
glm-ocr
Tested Inputs

test.jpeg: Traditional Chinese overtime form photo
eng-test.png: clean printed English document photo
hello-123.png: synthetic image with simple text
How It Was Tested
I tested both:
through the glm-ocr pipeline

directly with mlx_vlm.generate

Result
The model did not return usable OCR text.
Instead, outputs were mostly:
placeholder tokens like <|begin_of_image|>
empty markdown blocks

or a wrong single token such as 境

This happened even on:
a clean printed English image
a simple synthetic text image (HELLO 123)
Conclusion
The current local OCR setup is not working correctly for actual text extraction.

[Vin ./ @VinoisAsian]

We're expanding pre-train and model size, it's time to explore post training.

like how we, human learn, experience, we learn as we do , on the go, everyday!

[Gradient @Gradient_HQ]

Benchmarks that test what models have memorized are saturating fast. ARC-AGI-3 is asking a harder question: can AI actually learn something new on the fly?

One direction we've been exploring: multi-agent orchestration. In our study, coordinating four frontier LLMs across multiple turns consistently matched or outperformed the strongest single model, even on tasks none of them could solve alone.

The gap between "best single model" and "best coordination of models" is where a lot of the real progress is hiding.

More on our multi-turn, multi-agent orchestration study: https://t.co/LIwRoeggBB

[Vin ./ @VinoisAsian]

@HuggingModels okay wtf? my 0.6 can't even understand the task i gave

[Vin ./ @VinoisAsian]

@0xBagsAI_Sol @NguyenPhan171 sếp hỗ trợ e với :)))