🦋 hyeseong.kim

I'll post dev topics more on 🦋 bsky.app/profile/hyeseo…

here comes DeepSeek freakout 2.0: Weekly usage of Chinese AI models reached 9.223 trillion tokens, a 19.89% increase compared to the previous week; while the weekly usage of US AI models was 4.93 trillion tokens, a 16.27% increase compared to the previous week. Chinese models have surpassed the US in weekly usage for 4 consecutive weeks and remain the world's number one. DeepSeek-V4-Flash topping the charts stcn.com/article/detail…

🙏 Thank you all for the incredible love and support! Our latest Tencent Hunyuan translation models are on fire on Hugging Face: 🥰Hy-MT2-1.8B ranks #1 🥰Hy-MT2-30B-A3B ranks #4 on the open-source model trending leaderboard, with over 7K downloads already! To make it even easier for everyone, we’ve launched the Tencent Hy Translation WeChat mini-program, built on Hy-MT2. It supports voice input and offline translation, plus powerful customization of translation styles and instructions — delivering results that better match your expectations and feel far more practical. Try it out and share your feedback with us — we’d love to hear from you! Models on HF: huggingface.co/tencent/Hy-MT2… huggingface.co/tencent/Hy-MT2… GitHub: github.com/Tencent-Hunyua… #HyMT2 #TencentHunyuan #OpenSource

Three.jsでWeb流体エフェクトを即実装できるライブラリ「three-fluid-fx」が強力。 マウス連動の液体歪みやネオンの軌跡、大量のGPGPUパーティクルを記述不要で追加可能。 レガシーなWebGL（GLSL）と最先端のWebGPU（TSL）の両対応で即戦力となるオープンソース。（詳細や体験ページはリプ欄）

#TJ노래방 #PPPP #ヴァンパイア #日曜日

I just can’t get over how neat CXL type 3 is. Imagine having a 1TB bucket of memory. But! Instead of 1TB of DDR5, you have a tiered CXL accelerator. To the OS, it *looks* like regular memory, you address it in the same way. Maybe your accelerator is actually 100GB of DDR5, and ~1TB of high bandwidth flash. The first 100GB is your buffer, and a little controller slowly flushes it out. Many, many workloads are not hammering RAM enough for you to notice. Wait! You could get even more clever. With regular memory, bouncing cachelines between CPU cores is annoying. Often, you’ll program your way around this (avoiding a shared counter) by having each thread maintain a temporary local state with occasional global syncs. But, if we have a custom CXL 3 memory device, that slow global merge could be implemented in hardware instead. You’d never have to have cores fight over the same cacheline, because the shared-counter would be local to the CXL device! Aka, a remote atomic! This is essentially the concept of NDP (near-data processing), and of course there are much, much more fancy algorithms you can do with it, that’s just one example. But you can imagine, especially with database-style operations, how much bandwidth you could save not having to round-trip to the CPU and back for every operation. Imagine if your RAM could run a regex for you! We’re getting really close to that world.

@namenu_ 저도 beta때 드라이버 몇 개 터져서 안썻는데 레딧에 물어보니 요즘은 괜찮다는거 같기도.. 일단 트라이

@cometkim 저는 secure boot 꺼둬서 혹시 문제될까봐 생각도 못했는데... 후기가 궁금합니다.

let me setup Ubuntu 26.04 and finally TPM

Wow, it actually *is* DOM (See me using Chrome DevTools to inspect the DOM) HTML is <canvas> is going to open a new era of web UIs

한글이 110V 전력망 쓰는거처럼 되버렷는데 안타깝게도...

qualquer pessoa q ta numa empresa grande sabe que esta completamente insuportavel esse papo de ia nossa profissão virou pilotador de ia

AI is creating a weird HR crisis because the best engineering work in the agentic era is deletion. And deletion will never get you promoted. Here is an example of the paradox: Jerry spends six months building a sophisticated image analysis pipeline. Custom scaffolding, real-time frame extraction, a vision pre-processor that compresses and annotates before sending context back to the LLM. Thousands of lines. A system design doc. A launch review. Bob gets promoted. The work is visible, legible, and frankly impressive. Bravo Bob! Now Jeff comes in. Jeff looks at the pipeline and realizes the model does not need any of it. The last generation required that scaffolding. This one does not. Bob has 0 incentive to delete his work. Jeff goes and deletes 4,000 lines in one day. The system is faster, cheaper, more reliable. Token efficiency up 40%. Latency cut in half. Jeff will likely not get promoted for his one day scaffolding deletion. This is the core problem. Traditional engineering culture rewards addition. More features, more abstractions, more infrastructure. But elite agentic engineering is subtractive. The job is to remove every layer standing between the model and the task. Every unnecessary hop, every redundant transformation, every scaffolding that made sense six months ago and is now just dead weight. The engineer who keeps code close to the model, who deletes a lot, this person is doing the highest-leverage work available. And they look like they are barely working haha! How do we promote the Jeff?

Bro to bro, don't sleep on Cloudflare's stack.

Meu primeiro freela grande foi um e-commerce Eu tinha 16 anos, nunca tinha mexido com banco de dados Separei um dia, maratonei um curso de MySQL. No dia seguinte comecei a modelar o banco e fazer o backend em PHP Finalizei o e-commerce em ~1 mês. Rodou por mais de um ano, faturou quase 100k até o dia em que o dono descontinuou o negócio. Se você nunca fez algo assim, tá vivendo demais na zona de conforto

Tested Ubuntu 24.04: gist.github.com/lcfr-eth/2566a…

AV2 codec may be finalized on May 29 videocardz.com/newz/av2-codec…

公開まであと𝟗𝟕日 ⋆┈┈┈┈┈┈┈┈⋆ ✦特報第1.1弾 youtube.com/watch?v=vSdolb… 劇場版 魔法少女まどか☆マギカ 〈ワルプルギスの廻天〉 8月28日（金）公開 #まどかマギカ

ECMAScript excitement 😉 This week @TC39 met at @jetbrains Amsterdam & moved these proposals: 4️⃣ Atomics.pause() 4️⃣ Explicit Resource Management 4️⃣ Joint Iteration 3️⃣ Error stack accessor 3️⃣ Intl Keep Trailing Zeros 3️⃣ Iterator Chunking 3️⃣ Iterator Includes 3️⃣ Iterator Join 🧵

⚡@TimescaleDB 2.27 is out. For 10 years, we've had a consistent vision: start with Postgres, scale with Postgres. Reduce the need for complex data stacks with lagging CDC pipelines, weaker consistency, stale data, and more operational surface area. This release continues that work by making Hypercore, our Postgres-native columnstore, faster across more workloads. Specifically: • Selects. More filters now run vectorized through the standard Postgres function path, including in continuous aggregate refreshes. 30% - 200% improvements. • Updates and deletes. Bloom filters can skip decompressing compressed batches that cannot match equality predicates. Some crazy improvements up to 160x. • Upserts. Bloom filters can also prune batches during arbiter checks, avoiding unnecessary decompression on conflict-heavy write paths. Same Postgres. Less unnecessary work. Faster at scale.

We are making our discount permanent! 🎉 Enjoy building with DeepSeek-V4-Pro and bring your innovative ideas to life! 🚀

Official #Kotlin support for #VisualStudioCode is now available in alpha. #KotlinConf26