🦋 hyeseong.kim

Integration engineer / Open source hitchhiker / DX enthusiast

• Emerging tech R&D @daangnteam
• @rescriptlang compiler committer

[🦋 hyeseong.kim @cometkim]

I'll post dev topics more on 🦋

https://t.co/zYIfbNnmCO

[OpenCode @opencode]

Nemotron 3 Ultra is now free on OpenCode

text · 1M context · fully open source

NVIDIA's latest open source model

[AMD @AMD]

AM4 or AM5. DDR4 or DDR5. We’ve got you. 🤝

"AMD is bringing two X3D CPUs to market in a bid to combat rising component prices.”

Get ready for the Ryzen 7 5800X3D 10th Anniv. Edition ($349) and the Ryzen 7 7700X3D ($329).

Read the full breakdown from @tomshardware: https://t.co/6SG6Ed0XTD

[antirez @antirez]

DeepSeek v4 PRO running via SSD streaming on my 128GB MacBook m5 max. 1.6 trillion parameters.

[Jerry Liu @jerryjliu0]

We're presenting ParseBench at CVPR 2026!

ParseBench is the most comprehensive document understanding benchmark for VLMs.
✅ It contains 2k pages of real-world enterprise documents
✅ It has comprehensive evaluation metrics around tables, charts, visual grounding, semantic formatting, and content faithfulness

The core goal is measuring whether models can semantically interpret a document in the right way, without having models overfit to our precise benchmark.

Parsing 100% of PDFs to 100% accuracy is the final boss for document OCR.
In general, the latest frontier models have been tuned for coding, math, and scientific reasoning as opposed to precise visual understanding; hope more benchmarks that these will encourage overall progress towards solving this problem!

Poster is below. If you want to learn more come check out our site or 30-page ArXiv paper:

ParseBench: https://t.co/PWczfhp0OX
ArXiv: https://t.co/2dEJIaBBkr

[NVIDIA AI @NVIDIAAI]

Today we're shipping Nemotron 3 Ultra.

A 550B MoE frontier-intelligence open model built for long-running agents.

It delivers 5x faster inference and lowers the cost of complex agentic tasks by up to 30% versus other open frontier models.

[Cloudflare @Cloudflare]

VoidZero, the team behind Vite, Vitest, Rolldown, Oxc, and Vite+, is joining Cloudflare. Vite stays open source, vendor-agnostic, and built for everyone. https://t.co/DJTpX4Q9Xt

[VoidZero @voidzerodev]

VoidZero is joining Cloudflare.

Our mission stays the same: to make JavaScript developers more productive than ever before. Vite, Vitest, Rolldown, Oxc, and Vite+ remain MIT-licensed. Evan and the VoidZero team will continue leading them.

Cloudflare shares our commitment to open source. Together, we can keep investing in the tooling developers rely on every day, while bringing the Vite ecosystem and Cloudflare’s platform even closer together.

[🦋 hyeseong.kim @cometkim]

시키지도 않은 선언 이름 변경 다듬기에 호환성 명목으로 코드 부풀리기에 하라는건 안하고 라인 변경만 이쁘게 보이도록 만드는쪽으로 작업이 편향되는게 바로 보임... 전에도 이래서 치웠는데

[🦋 hyeseong.kim @cometkim]

아키텍처적인 변경을 시키는 순간 대응못하고 개떡같은 코드나 뱉고 마는건 플래닝을 아무리 해줘도 나아질 기미가 안보이네

[🦋 hyeseong.kim @cometkim]

GPT 5.5 는 진짜 답답해서 못쓰겠다. 다른걸로 잘 하다가 한도차서 GPT 5.5로 스위치 하자마자 돌던거 다 망가진다... 전체 프롬프트를 다 뒤져도 그딴거 시킨적 없는데 왜 날조하세요;

[Paul Miller @paulmillr]

> WASM, but paranoia-friendly

Announcing 3 new projects:

1. awasm: a new standard for auditable, high-security WASM packages
2. awasm-compiler: an auditable js-to-wasm compiler
3. awasm-noble: an ultra-fast WASM cryptography library (ChaCha/BLAKE3 at 6-10 GB/s)

[Rust Bytes 🦀 @rustaceans_rs]

Servo 0.2.0 is out: major Android UI refresh, huge CJK improvements, strong DevTools upgrades, and dozens of web platform wins — including first-major-engine crypto.subtle.supports(). Upgrade and test!

#rustlang #rust

[JFrog Security @JFrogSecurity]

🚨 Supply Chain ALERT New Phase of “Miasma: The Spreading Blight” 🚨

A new wave of the npm supply chain campaign, Miasma: The Spreading Blight
Targets include @‌vapi-ai/server-sdk & ai-sdk-ollama. Attackers are using a new autorun install method exploiting node-gyp shell expansion.

We have updated our previous blog:
https://t.co/gFhlSCFzOu

[🦋 hyeseong.kim @cometkim]

어우 비가 무슨 국장처럼 쏟아지네

[RyanLee @RyanLeeMiniMax]

MiniMax-M3 now is rank #8 on @ArtificialAnlys.
Due to the workload involved in open-sourcing the MSA operator in parallel, the weights will be released to everyone late next week.

[Unsloth AI @UnslothAI]

2-bit Gemma 4 12B GGUF, only 4.66 GB on disk, managed to cite 15 sites from a single prompt.

Try this locally on >6GB RAM via Unsloth Studio.

GitHub: https://t.co/aZWYAtakBP