Yu Yang

PhD, PE, PMP. Data scientist. @MIZZOU Tiger. Interested in LLM, AI, Webapp. Side Hustle = Real Estate. Opinions = mine.

[Yu Yang @Yang_ML_Estate]

Summary of my 2025.

This is my third year as a senior data scientist. I will keep a super short summary.

1. Solo-built and launched Estate AI (https://t.co/1snMWJlsWS) this year—an AI-agent real-estate analysis app—owning the end-to-end stack: Next.js (front end deployed on Vercel), FastAPI (back end deployed on Railway), Supabase for auth and database, and Stripe for billing. One of my friends subscribed to it, showing her support. Other than that, there are no active users yet. ~~But I am still proud of it, because it will be a great tool for my own investment.

2. Wrote 8 blogs on Medium, covering topics such as GRPO, reinforcement learning, sparse attention, and MCP. These posts accumulated a total of 16,838 presentations and 2,236 views.

3. My 8-year-old elder boy still does not speak any words. However, he shows more spatial intelligence than the most advanced robot in the world. He remembers all the trails we have walked and knows exactly where to turn left or right. Some of the tricks I learned from reinforcement learning could somehow be applied to him as well ~~

4. Changed jobs, taking on more responsibility.

5. Invested in a mobile home park project as an LP. I have known the GP for many years. Before investing, I also spoke with his previous investors. So far, so good.

6. No personal real-estate deals this year. I’ve been getting super busy and would like to focus more on learning reinforcement learning and post-training.

7. Listened to about 200 hours of AI-related podcasts. Happy to better understand AI, robotics, and the long-term trends around nuclear fusion.

Hopefully all of my friends, both online and offline, also had a fruitful year. I am becoming more optimistic about my elder son’s future, with advances in AI for decoding brain signals (such as Meta’s work this year) and progress in brain–computer interfaces.

Happy New Year to all of you!

[Yu Yang @Yang_ML_Estate]

Summary of my 2024

This year marked my second year working as a senior data scientist, and I turned 42. I remain strongly motivated to work in data science and to learn about the frontiers of large language models (LLMs). Before the end of 2024, I reflected on what I had achieved for myself and my family. My work evaluation was completed before Christmas and is not included here.

• March 2024: Published a book titled "Poems from Tang Dynasty Meet AI" on Amazon. The book showcases 100 poems from the Tang Dynasty (618–907 A.D.), presented in both Chinese and English. Each poem is accompanied by a beautiful painting, initially generated by Midjourney and refined with Stable Diffusion and PhotoPea. The English translations, first produced by ChatGPT, were reviewed and edited by me. This collection enables readers to connect with the loneliness, sorrows, and joys experienced over 1,000 years ago, making it an invaluable resource for language learners. It stands as one of the first bilingual books in the era of AI-generated content (AIGC).

• June 2024: My elder son finally learned to tolerate an electronic toothbrush after over 13 months of training and demonstration. I was proud of myself for successfully training the most challenging model - a seven-year-old child with severe autism who is non-verbal.

• April - August 2024: Completed "The Complete 2024 Web Development Bootcamp" taught by Dr. Angela Yu. The course teaches fundamental skills in HTML, CSS, JavaScript, Express, React, and DeFi. It also provides many useful tips for learning and practicing programming and web development. Dr. Angela Yu is an exceptional teacher whose course has been taken by over 1.3 million people in the past few years, earning a rating of 4.7 out of 5 from 395,461 students.

• September - December 2024: Worked on a real estate web app. Being a side project, I had to pause it for a few weeks when I needed to focus on my day job. Making slow but steady progress.

• Tech Blog Writing: Published 10 articles on Medium this year, covering visualization plugin development, transformers, AWS EC2/ECS with GPU, and prompt engineering for LLMs.

• Real Estate Investment: No progress this year. Maintained our current properties. Too busy with other projects.

Finally, I would like to sincerely thank everyone who followed, commented on, and liked my tweets (or Xs). Personally, I have been looking forward to more progress in real AI applications in brain/neuron sciences, which could help identify the root causes of severe autism and other brain conditions, advance new bio-compatible materials for brain chips, and eventually lead to combined solutions (gene editing or brain chips) for adults with autism who lack basic living skills. Remember, if you can read this tweet, you are already blessed compared to people with severe autism who lack basic living skills. Let's make the best use of our intelligence and make progress, little by little.

[Yu Yang @Yang_ML_Estate]

@hxiao @sgl_project @radixark Interesting project 👍👍

[Yu Yang @Yang_ML_Estate]

What’s your experience in using X ads?

I just paid $25 to boost my post last week, hoping the free children speak app could help more people. It told me that “boost is paused” without any explanation 🤣🤣

No refund either.

The link to adspolicy shows nothing 🤣🤣

[Yu Yang @Yang_ML_Estate]

I have an idea to craft a game of adventure, which combines moves of Onmyoji and Daoist.

[Gérier | Only Indies @OnlyGerier]

Tengo sentimientos encontrados con Cook The Dungeon, un roguelike deckbuilder en primera persona que utiliza un sistema de caza y cocina de monstruos para mejorar tus stats.

El juego está ENTERO hecho con IA. Pero no parece un AI slop, de hecho luce muy bien.

¿Qué pensáis?

[Yu Yang @Yang_ML_Estate]

@ZeyuanAllenZhu @syhw 世人闻此皆掉头，有如东风射马耳

[Yu Yang @Yang_ML_Estate]

@paulg “ AI is great, but please do not build data centers in my neighborhood “ 😜😜

[Yu Yang @Yang_ML_Estate]

@jietang Long-horizon reasoning is like playing chess. If you can think 20 moves ahead, you’ll be a much stronger player than an opponent who can only think five moves ahead.

[Yu Yang @Yang_ML_Estate]

Just released ChildrenSpeak v1.0.2 on the iOS App Store! It’s free without any ads!

ChildrenSpeak is an AAC (Augmentative and Alternative Communication) app I built for my elder son, who is on the autism spectrum. It helps children express their needs, feelings, and interests through visual word cards with audio feedback. The whole philosophy of design is to keep it as simple as possible with only the words that matter most.

If you're a parent, therapist, or educator working with non-verbal or minimally verbal children, l'd love to hear from you. Are there other languages you'd like to see supported -- Chinese, Spanish, or others? Let me know in the comments.

[Yu Yang @Yang_ML_Estate]

Distillation on human beings is still going on

[Yu Yang @Yang_ML_Estate]

@jiahuiyu @finkd @alexandr_wang 👍👍Looking forward to your next exciting journey!

[Yu Yang @Yang_ML_Estate]

ByteDance is clearly capable of building products as impressive as Seedance, which says a lot about the strength of its organization and the density of its talent.

So it probably doesn’t need to rely heavily on distillation from proprietary models. And when distillation does make sense, it can do it intelligently using open-source models, which also helps avoid potential compliance issues.

[Yu Yang @Yang_ML_Estate]

I don’t completely buy this argument.

Three reasons:

1. The more I study RL, the more I appreciate the importance of clearly defined value functions and rewards for robots. Today’s robots still seem far from animal-level intelligence, particularly in spatial reasoning.

2. I’ve listened to podcasts with researchers from Physical Intelligence (π) and Google DeepMind who have worked on robotics for years. There is still a long way to go from impressive demos to reliable production systems.

Autonomous driving, for example, is still not completely solved after 20+ years of effort—and the vehicle motion-planning problem is relatively constrained, often modeled with just 3 DOF: x, y, and yaw. A humanoid robot with dexterous hands can easily have 20+ DOF. The complexity explodes.

3. Elon has a track record of overselling timelines and capabilities.

[Yu Yang @Yang_ML_Estate]

@antirez @liuliu @drawthingsapp thanks. Will give it a try on my Mac m2

[Yu Yang @Yang_ML_Estate]

Yes. In the movie from South Korea. The lawyer was proud of his job. The cast was one of my favorites.

The Attorney (2013) is a South Korean legal drama inspired by the real-life "Burim case," about a materialistic tax lawyer who transforms into a human rights advocate after defending a student falsely accused of being a communist by the authoritarian government in the 1980s.

[Yu Yang @Yang_ML_Estate]

@RyanEls4 Next time when my dev friends talk to me, I will ask them to start with “ you honor…”. Because I have the title of Data Scientist 😁😁

[Yu Yang @Yang_ML_Estate]

@AlexChhk Zhongguancun core tech area at Beijing? Moonshot, Zhipu, Minimax …

[Yu Yang @Yang_ML_Estate]

Great to see another neo AI lab working on AI4science. Somehow it reminds me the story of Yonghui Wu, who is currently the leader behind Seedance.

Yonghui Wu spent 17 years at Google DeepMind, rising to Research Vice President and Google Fellow, and worked on core Gemini R&D, before ByteDance recruited him in February 2025 to lead the overall Seed org, reporting directly to CEO Liang Rubo.

[Yu Yang @Yang_ML_Estate]

@YiTayML @sur4js Haha, that’s also my argument. Google still has a great pool of top talent. If a small team like DeepSeek could build a tier-1 LLM, Google DeepMind has the more-than-enough talents to build frontier multimodal LLMs and AI4Science.

[Yu Yang @Yang_ML_Estate]

@JeffDean @Sanjay_Ghemawat Followed . Looking forward to hearing interesting stories from him.

[Yu Yang @Yang_ML_Estate]

多模态能力对于生存很重要，盲人可以生存且有高智力依赖于agent collaboration. 假设某个地区所有人都是盲人，种植，交通，书籍和其他智力资源的生产和提供，饮用水处理和废水处理，没有视觉理解能力，完成这些任务基本不现实。

AGI首先没有一个统一的概念。如果宣称实现了AGI, 但是缺乏视觉理解和物理世界交互能力，这可能算是虚拟世界的弱AGI 。

[Yu Yang @Yang_ML_Estate]

@lasagna_silent @jasondeanlee @nabeelqu @littmath @Jacob_Tsimerman For most of common people like me, we do not understand math at the Fields Medal level🤣🤣