Seoul just had a Grok Bot meetup. Would love to run one in Daejeon too.
KAIST city, so good spot for a casual build + share session (I’ve hosted several meetups before).
@benln interested?
The hardest part of designing a wrist rest? Making it stay in one piece when I picked it up.
Several tolerance tweaks later, these two finally get along.
My honeycomb wrist rest for the Rule 60 V3. No screws, no magnets, just printed parts.
Free files:
makerworld.com/en/models/349…
Picked up a @BambulabGlobal X2D AMS Combo for 16% off during Korea’s Black Friday sale.
Now, what should I print first?
Woonggi Min retweeted
gpu-unlock has been trending lately. Happened to have two RTX 3090s on hand, and since there weren't many real-world LLM benchmarks out there, I ran a 100K context test myself.
See comment for more details.
notion is dead
acard.dev/
Stop using Notion and Obsidian.
Let AI do the organizing.
Just talk to your AI. The conversation becomes your notes, your docs, your record.
Acard - local, open source
I'll make, $5 gets you the web app and everything else.
Woonggi Min retweeted
acard.dev/
Stop using Notion and Obsidian.
Let AI do the organizing.
Just talk to your AI. The conversation becomes your notes, your docs, your record.
Acard - local, open source
I'll make, $5 gets you the web app and everything else.
Woonggi Min retweeted
Dynamic sparse attention reads fewer tokens per query. So why doesn’t the KV cache necessarily shrink to match?
Sliding-window attention only looks at recent tokens. Once a token falls outside the window, that layer no longer needs its KV state to continue the sequence, so the runtime can release it.
Dynamic sparse attention works differently. Each query selects which tokens to attend to from across the history. A token skipped by this query may be exactly what the next one needs.
So the runtime may still need to keep the full KV history, even though each step reads only a subset. And selecting positions and gathering noncontiguous KV entries adds its own cost.
Fewer attention reads ≠ a smaller cache. For long-context agents, compute and memory are two different budgets.
Our deep dive covers what this, and hybrid models like Kimi K3, mean for prefix reuse, caching, and speculative decoding: friendli.ai/blog/llm-archite…
prime time to switch to @AsideAI
We've decided to sunset Arc on December 22, 2026.
That's not an easy sentence to write. We spent years building Arc and celebrating each and every release along the way: we have so much pride and love for what we made alongside the Arc community.
Our priority is to deliver best-in-class security for our members. Keeping a browser secure takes constant, dedicated work that is only increasing in the age of AI. Dia was built from the ground up to meet today’s security standards, so that's where our engineering and security teams will focus moving forward.
We've worked to make switching to @diabrowser as easy as possible for Arc members. Here's how:
· Your passwords, bookmarks, extensions and history come over to Dia seamlessly.
· Spaces, Folders, and Tab Groups are in Dia and import over with your data intact.
· Dia now surpasses Arc benchmarks in stability, speed, and security.
· The browser features you use in Arc are free in Dia, and the AI is optional.
Thank you for using Arc, for pushing us to make it better, and for building alongside us. The team that made Arc is the same team behind Dia, and some of our most ambitious work is just ahead. We'd love for you to be there with us when it lands.
Woonggi Min retweeted
Another interesting chart showing FriendliAI's TTFT performance for GLM-5.3 on Vercel AI Gateway over the past week.
The chart shows each provider's TTFT relative to the slowest provider, on a log scale.
At its peak, FriendliAI was 2,891× faster than the slowest provider in TTFT.
Woonggi Min retweeted
Interesting trend from @vercel AI Gateway:
Open-weight models account for 71.5% of token usage, but only 23% of total spending.
That's a lot of inference volume at a relatively low cost.
While this reflects Vercel's workloads rather than the entire AI market, it's a compelling signal of how cost-efficient open-weight models are becoming for real-world applications.
vercel.com/blog/ai-gateway-p…
Picked up a @BambulabGlobal X2D AMS Combo for 16% off during Korea’s Black Friday sale.
Now, what should I print first?
Woonggi Min retweeted
I used Reverse Engineer Anything to re-create raycast on @omarchy. And I was blown away! In just a couple hours I have a fully-functioning raycast-like launcher! Right now it:
* Learns your habits
* Apps + every Omarchy action in one search
* Aliases & favorites
* Keyboard-first
* Follows your Omarchy theme live, opens in ~40ms
repo: github.com/typecraft-dev/oma…
Woonggi Min retweeted
A matching prefix doesn't always mean reusable state.
As AI agents tackle longer, multi-step tasks, modern LLM architectures are changing how inference works.
Explore how MLA, hybrid architectures, and smarter caching are reshaping inference for agentic workloads. 🚀
🔗 friendli.ai/blog/llm-archite…
Woonggi Min retweeted
디시인사이드도 20일 개선 챌린지 시작합니다.
평일 20일, 하루에 하나씩 디시를 바꾸고 매일 공개합니다.
못 지키면 다음 날 사과문 쓰고 쏩니다. (일단 KFC가 같이 쏴주기로 함, 다른 파트너도 환영입니다)
Reverse-engineered Lenovo's Windows driver (synaWudfBioUsb136.dll) to bring the ThinkPad fingerprint LED to Linux (Omarchy): green on ready, amber blink on mismatch, matching Windows behavior 1:1.
github.com/omacom/omarchy-pk…
