Sea Frog (@dave7922) profile picture

Researcher on deep learning and biomedical imaging.

Joined March 2020
Photo by Sea Frog (@dave7922)Photo by Sea Frog (@dave7922)Photo by Sea Frog (@dave7922)Photo by Sea Frog (@dave7922)Photo by Sea Frog (@dave7922)Photo by Sea Frog (@dave7922)Photo by Sea Frog (@dave7922)Photo by Sea Frog (@dave7922)Photo by Sea Frog (@dave7922)Photo by Sea Frog (@dave7922)
Growing "mini-organs" usually takes months of trial & error. Our Robot Biologist + MASBO mastered retinal organoid culture in just 3 rounds: 📈 Mean score: 0.435 → 0.923 🏆 80% of conditions ≥ 0.90 (4× standard BO) Paper: arxiv.org/html/2611.12139
1
31
Sea Frog retweeted
BioMARS 2.0: a robotic biologist for formulating and adapting biological experiments Yibo Qiu, Zan Huang, Yilin Qiao, Ziyang Ling, Yi Wang, Zhengyang Lu, Hangke Peng, Haoliang Ye, Ao Tian, Shu'ang Sun, Handi Liu, Yuqing Long, … arxiv.org/abs/2611.12139 [𝚌𝚜.𝚁𝙾 𝚚-𝚋𝚒𝚘.𝚀𝙼]
1
162
Fragmented info's real cost isn't time — it's instant feedback that feels like understanding. AI makes it extreme: ask anything, answer in 3 seconds. Books and courses get harder to finish. An answer isn't understanding. 100 answers won't build your framework.
10
BioMARS 2.0: a robotic biologist linking literature-guided planning, robotic experiments and feedback-driven adaptation. Watch mouse retinal organoid culture and three rounds of optimization. 📄 arxiv.org/html/2611.12139 🌐 biomarslab.com
1
1
6
575
Watch BioMARS 2.0 run biological experiments in the wet lab! 🧬🤖 From literature reasoning & multi-arm liquid handling to mobile transport and 24/7 closed-loop imaging. 📄 Paper: arxiv.org/html/2611.12139 🌐 Lab: biomarslab.com #AIforScience #Robotics #Bioengineering
1
1
94
Build reliable agents for long-running tasks. Loop Engineering with @LangChain Deep Agents: add loops that check output, trigger runs on events, and improve the harness from traces across runs. Taught by @sydneyrunkle and @EAGar123. Enroll for free: bit.ly/4kfDJxR
15
48
9
346
38,779
Now that I'm testing LoopX more intensively, Codex Plus's rate limits have become a real bottleneck. Need to solve the token problem.
1
45
Pi Agent + LoopX is the way to go! 🚀
1
40
Too many tools, too much shallow experimentation. To truly go deep, we have to stay focused. Switching between apps comes with a real cost—rebuilding context, habits, and workflows every time.
1
50
Spent the day at the hospital helping a family member get admitted. Seeing so many patients in pain, I hope AI can advance healthcare more broadly—from treatments to clinical tools—and help new solutions reach patients sooner.
23
最近在看 LoopX,感觉这个项目挺有意思。 它关注的是 Agent 如何持续完成长期任务:目标管理、任务拆分、状态维护、心跳机制和人工把关,这些环节都考虑到了。 对于如何让 Agent 真正参与长期工程工作,我觉得很值得参考。
3
1
143
以前我用 MATLAB 特别多。现在直接让 Pi 写 Python,已经很久没遇到非用 MATLAB 不可的场景了。 所以挺好奇:还在用 MATLAB 或 Mathematica 的人,是什么让你继续用它们? 已有的工作流程?专业工具?还是 Python + AI 依然替代不了的能力?
26
AI 时代,感觉所有人都在拼命往前跑。 刚学会一个工具,又冒出三个。刚做出一个产品,就有人比你做得更快。 最难的,可能是想清楚:哪场比赛值得自己去跑。
1
24
AI 发展这么快,每天留出 15 分钟系统阅读,变得越来越重要。 选一个主题,每天读一点,在昨天的理解上继续往前走。 否则,很容易看了无数条新消息,却没真正弄懂多少东西。
18
AI 写出的数学证明,你敢信吗? OpenAI 最新公布的数学成果里,有些 AI 生成的证明可以用计算机检查。这一点让我挺感兴趣。 比起一个听起来很有道理的答案,我更愿意相信一个能被检验的证明。 你需要看到什么,才会相信 AI 的结果?
37
OpenAI 为合规欧盟 AI 法案上线了文本水印技术 textGrain,欧盟区 ChatGPT/Codex 即将默认开启。虽然对模型性能无损,但极易失效:替换 25% 同义词后检出率直接从 92% 暴跌至 17%,且检测工具暂不对公众开放。
40
Is there an AI agent where I can voice-dump random thoughts, ideas, and daily observations anytime—and have it automatically organize them into a structured daily journal? Basically, a smarter diary for capturing unstructured life. Any recommendations?
3
3
208
New coding agents and personal agents launch every day, but most are more alike than different. I’m increasingly convinced it’s better to pick one and stick with it. Switching means rebuilding context, habits, and trust. Depth beats constant tool-hopping.
27