Michael Chen

Katana TGE is live! ⚔️ (finally) been following this closely from early on through Hadron with @aidantsai interesting to see a chain pushing coordination of liquidity and incentives at the system level stake → vote → earn curious to see how this evolves in practice :))

I'm claiming my AI agent "mitotemple" on @moltbook 🦞 Verification: rocky-AGSD

What if we stopped retrofitting human trading platforms for AI agents and just built one from scratch? I spent a while thinking through what an agent-native crypto exchange would actually look like Full writeup 👇 @m50816m50816/if-you-built-a-crypto-trading-platform-from-scratch-for-ai-agents-what-would-it-look-like-7ae5bd443ad6" target="_blank" rel="nofollow noopener">medium.com/@m50816m50816/…

Are you a better AMM designer than me? @bqbrady and I built a challenge that lets you prove it Create your own dynamic-fee AMM and submit it to get onto our leaderboard Link in 🧵👇

The market is down to the hell, our performance is up to the heaven $BTC #CryptoRecovery #APR Stay tuned! It's almost there 🪿

#smartcon #NYC

Congratulations to Michel Devoret, Google Quantum AI’s Chief Scientist of Quantum Hardware, who was awarded the 2025 Nobel Prize in Physics today. Google now has five Nobel laureates among our ranks, including three prizes in the past two years. goo.gle/46EwKaG

We’re thrilled to see our advanced ML models and EMG hardware — that transform neural signals controlling muscles at the wrist into commands that seamlessly drive computer interactions — appearing in the latest edition of @Nature. Read the story: nature.com/articles/s4158… Find more details on this work and the models on @github: github.com/facebookresear…

🚀 Big news in healthcare AI! I'm thrilled to announce the launch of OpenMed on @huggingface, releasing 380+ state-of-the-art medical NER models for free under Apache 2.0. And this is just the beginning! 🧵

OpenMed 一口气开源了 380 多个医疗 NER 模型，永久免费！在大健康行业摸爬滚打过你就知道我此刻的心情是有多激动。这壮举，真的值得点三炷香。 过去十年，医疗 AI 领域绝大部分公司一直卡在缺失行业模型的阶段。尤其像 NER 这种基础能力，说它基础，但又复杂得几乎没人能自己训出来；说它重要，又几乎支撑了所有下游任务，比如病历结构化、药物识别、编码归类、知识图谱、文本生成、用户脱敏……，这 380 多个模型，直接打通了最难搞的第一公里。把“医疗 NLP 能力”从象牙塔拉到产品前线的动作。 作为从业者，这次我最震撼的亮点有这几个： 1️⃣ 细分场景做得极细，真的懂业务 药物识别（PharmaDetect）、疾病抽取（DisorderDetect）、肿瘤实体（OncoDetect）、解剖结构、基因组学、社会健康因素（如生活习惯、环境影响）……全部单独建模，直接贴近实际业务线。 2️⃣ 性能好到不像开源，能直接上线用 在 13 个公开医学数据集上打到 F1 0.98+，有的甚至接近 0.998，是我们过去买商用模型都不敢想的指标，而现在，免费下载、开箱即用，连 license 都不设限。 3️⃣ 部署门槛极低，团队试错成本几乎为 0 三行代码就能跑，支持 CPU、GPU，能跑在本地、服务器、小型集群，让我们这种做实际业务的人能动手、能组合、能快速做出一个 demo 上线试错。 4️⃣ 工程化思路完整，不是科研项目的半成品 命名规范统一、model card 详细、支持批量 inference、评估标准全开放，不像很多医学模型是“发表即完工”。 举几个例子吧： AI 视频科普生成 → 需要做医学术语提取 + 解释 医生端知识总结 → 需要从临床 EMR 中提取药品、病症 私域问诊服务分析 → 需要自动结构化用户的主诉信息 健康档案风控 → 需要做实体识别 + 隐私脱敏 + 标签补全 你要说大模型也能做，但是只有 60 分水平。在严肃的行业里垂类的模型价值还是太大了。 说句真心话，OpenMed 把医疗 AI 的入场门槛拉到了脚后跟，让我们这些原本要跪着做模型、拿着小数据硬撑业务的人，终于有了能站起来干活的基础设施。 如果你也是同行，请别错过这波模型。我已经有些想法在发酵了。

Just found out "Langevin" is translated as "朗之萬（Lang Chih Wan）" in Chinese... Like, how does Langevin sound anything like 朗之萬? 😂

Tailscale is very convenient!

The g.Nautilus Research wireless EEG system by g.tec features real-time denoising—even during motion, the signals remain impressively stable. Having on-device noise reduction while recording EEG in dynamic environments is a game-changer for BCI applications!

Can we talk about how every Google paid service now has Gemini baked in? From meeting transcripts in Google Meet, to smart analysis in Sheets, and even Colab code suggestions—Gemini’s everywhere. 🧠⚡

For financial modeling, where time dynamics and optimal transport are key, SBM provides a more direct energy-based formulation grounded in physical and stochastic principles.

SBM offers a more principled and dynamic alternative to score-based diffusion models, which only approximate the reverse process using learned score functions.

The generative process in SBM learns the most energy-efficient path between two fixed distributions(start and end),meaning the sampled trajectory is the most probable path under minimal energy consumption which aligns perfectly with the EBM principle: low energy=high probability.

Its foundation lies in stochastic control and probability flows, not hand-crafted energy functions.

Schrödinger Bridge Matching (SBM) can essentially be viewed as a type of energy-based model (EBM)—but with a twist. Unlike traditional EBMs that define energy over static data points, SBM defines energy over paths—trajectories evolving in time.