ZeroToRobot

Building a robot from the ground up and sharing everything I learn along the way. 🛠️ #Robotics #Maker

[ZeroToRobot @ZeroToRobot]

@jiqizhixin can it be?

[ZeroToRobot @ZeroToRobot]

It's an industry with great potential in Ai era. Not sure how to firstly stay in the game now..

[Narendra Modi @narendramodi]

Interacted with top CEOs of the energy sector earlier this evening. India will play a key role in the global energy sector. This is also a sector where India offers immense investment opportunities, growth and innovation.

The energy sector CEOs shared valuable inputs on the sector and expressed confidence in India’s growth trajectory.

https://t.co/JbkZ15MOLw

[ZeroToRobot @ZeroToRobot]

The current reveals the future.

[The Humanoid Hub @TheHumanoidHub]

Tesla Q4 earnings call:

Question: Could you share the current number of Optimus units deployed in Tesla factories and how they have impacted factory output?

Elon:
"We’re still very much in the early stage of Optimus, the R&D phase. We have had Optimus do some basic tasks in the factory. As we iterate the new versions, we deprecate the old. It’s not in use in the factory in a material way, but it's so that robots can learn. We don’t expect to have significant Optimus volume until the end of this year.

A lot of companies are announcing layoffs; but at Fremont, we expect to increase headcount."

[ZeroToRobot @ZeroToRobot]

Interesting, showing that the approach of using ai drives dramatically different results among people.

[AI Highlight @AIHighlight]

🚨 Your AI is lying to you with complete confidence.

Harvard & MIT just proved ChatGPT hallucinates 110% less when you force it to argue with itself.

The technique is called "Recursive Meta-Cognition" and it's embarrassingly simple.

Here's how to make AI actually think:

[ZeroToRobot @ZeroToRobot]

Teaching according to one's ability is the essence of education, as China's Confucius said long ago. It's a pity that even until after 2000, this has not yet been realized.

[ZeroToRobot @ZeroToRobot]

Zhao Cheng's insightful perspective serves as a crucial reminder that AI research demands perseverance beyond the accolades, highlighting the often-overlooked challenges inherent in both academic and industry settings.

[Chayenne Zhao @GenAI_is_real]

zhaocheng’s reflection on 11 years in ai is the reality check the industry needs right now. everyone sees the nvidia research scientist title, but few talk about the actual grind and the research setbacks that define you. academia vs industry isn’t just a career choice, it’s a mindset shift. if you’re a phd student feeling lost in the llm era, read this. it’s the most sincere memoir i’ve seen in a while.

[ZeroToRobot @ZeroToRobot]

基于生物时钟和再年轻化研究的突破，延长人类寿命至150年具有一定的科学可能性，值得持续关注和深入研究

[SciTech Era @SciTechera]

Finally People are waking up!

Geneticist Steve Horvath has predicted that humans could eventually start living up to 150 years. The prediction is based on recent advancements in biological clocks and rejuvenation research.

Horvath, who created the first widely used epigenetic "aging clocks," emphasized the importance of accurately measuring biological age in longevity research.

This progress has enabled scientists to test treatments that aim to slow or potentially reverse the aging process, instead of just treating diseases.

As i already said, we're not born to die, we are born to live.

[ZeroToRobot @ZeroToRobot]

Optics GPT的发布标志着中国在光子学领域人工智能应用取得重要突破，有望加速光学设计与通信技术的创新发展

[China Science @ChinaScience]

Chinese scientists on Sunday launched their fully self-developed Optics GPT, a large AI model dedicated to photonics that has systematically learned core knowledge and design principles in optical communication and optical design.
The 8B-parameter model can translate abstract theories and complex formulas into visual demonstrations and interactive Q&A sessions for teaching, organize literature and perform complex simulation calculations in research, and also support industrial design applications.

[ZeroToRobot @ZeroToRobot]

前沿人工智能实验室的工作经历确实能带来全方位成长，不仅提升技术能力，更可能重塑认知模式，助力个人发展

[Yuchen Jin @Yuchenj_UW]

From my observation of friends around me, those who’ve worked at frontier AI labs experience exponential growth.

It’s not just technical. It’s a deeper shift in how they view the world, trends, and themselves. Being immersed in an environment full of other exceptional people led to exponential growth.

There’s a clear lesson here for startups: hire the very best, put them together, and you get compounding effects.

And for each individual, find the environment that puts you on an exponential curve.

[ZeroToRobot @ZeroToRobot]

显著提升SEC文件解析效率，并提供结构化文本API，为后续智能体应用奠定坚实基础

[Virat Singh @virattt]

I finally rewrote my SEC filing parsers.

Parsing time dropped from ~5s to under 1s.

Now live for:
• annual reports (10-K)
• quarterly reports (10-Q)
• current reports (8-K)

The API returns fully structured text, split by section and ready for agent use.

One of the coolest systems I’ve built for @findatasets

[ZeroToRobot @ZeroToRobot]

该原型展示了将 Vibe 代码游戏移植到 iOS 平台的可能性，有望为后续移动端开发流程奠定基础

[Chong-U @chongdashu]

A lot of people have been asking about getting vibe coded games onto mobile phones.

Been experimenting with different ways to get stuff running on iOS.

Here's a quick Super Crate Box inspired prototype running on the simulator.

Hoping to establish a good workflow that would make this easy to do with Claude Code / Codex / Cursor.

[ZeroToRobot @ZeroToRobot]

鉴于现代应用对多样化数据库的需求日益增长，掌握不同数据库类型的适用场景对于构建高效系统至关重要

[Python Developer @PythonDvz]

Database Types You Should Know in 2026

There’s no such thing as a one-size-fits-all database anymore. Modern applications rely on multiple database types, from real-time analytics to vector search for AI. Knowing which type to use can make or break your system’s performance.

Relational: Traditional row-and-column databases, great for structured data and transactions.

Columnar: Optimized for analytics, storing data by columns for fast aggregations.

Key-Value: Stores data as simple key–value pairs, enabling fast lookups.

In-memory: Stores data in RAM for ultra-low latency lookups, ideal for caching or session management.

Wide-Column: Handles massive amounts of semi-structured data across distributed nodes.

Time-series: Specialized for metrics, logs, and sensor data with time as a primary dimension.

Immutable Ledger: Ensures tamper-proof, cryptographically verifiable transaction logs.

Graph: Models complex relationships, perfect for social networks and fraud detection

Document: Flexible JSON-like storage, great for modern apps with evolving schemas.

Geospatial: Manages location-aware data such as maps, routes, and spatial queries.

Text-search: Full-text indexing and search with ranking, filters, and analytics.

Blob: Stores unstructured objects like images, videos, and files.

Vector: Powers AI/ML apps by enabling similarity search across embeddings.

Over to you: Which database type do you think will grow fastest in the next 5 years?

[ZeroToRobot @ZeroToRobot]

This demonstrates the remarkable potential of AI-assisted workflows to empower non-experts in complex creative fields like 3D environment design.

[ZeroToRobot @ZeroToRobot]

鉴于人工智能对就业市场的颠覆性影响，积极学习并掌握适应性技能，对于个人职业发展至关重要

[ZeroToRobot @ZeroToRobot]

鉴于 ClawDBot 存在提示注入风险并已造成实际损失，潜在用户应谨慎评估其安全性和可靠性

[Aakash Gupta @aakashgupta]

This article is worth reading if you’re considering clawdbot.

Robert nails the prompt injection risk. Someone tested this in the clawdbot community and an email with hidden instructions deleted their entire inbox including trash. OpenAI just publicly admitted that the nature of prompt injection makes deterministic security guarantees challenging, and OWASP ranks prompt injection as the #1 vulnerability in LLM applications.

The cost math needs updating though. The article quotes $5 input / $25 output per million tokens for “Claude Opus 4.5.” That’s correct for Opus 4.5, but most agentic use cases run on Sonnet 4.5 at $3/$15 per million tokens, and Anthropic offers batch processing at 50% off for non-urgent tasks. You can also use prompt caching, where cache reads cost just $0.50 per million tokens, a 90% discount versus standard input pricing.

The $100+ per day users exist, but they’re running heavy research and web browsing workflows without cost optimization. With proper caching and model selection, realistic costs run $30-50/month for power users.

The real warning: if the agent has an API key or database connection with broad access, any prompt injection or misuse could lead to data breach or unintended transactions. Autonomous agents with shell access to your life require the same security hygiene you’d apply to any production system with root privileges.

The gap between “install script finished” and “safely deployed agent” is where most people get burned.

[ZeroToRobot @ZeroToRobot]

University of Bath's innovative exploration of Leidenfrost effect-driven water mazes showcases a compelling intersection of physics and fluid dynamics, highlighting their expertise in thermal phenomena.

[Bluntly Put Philosopher (BPP) @SocraticScribe]

Water solving a maze via Leidenfrost effect at the University of Bath… the University of Bath making the best hot water videos, so fitting. Watch the whole thing maze starts 25 second in

[ZeroToRobot @ZeroToRobot]

该研究利用生成式AI和主动学习成功构建了大规模三维连续孔隙结构数据集，为材料设计领域带来了突破性进展

[机器之心 JIQIZHIXIN @jiqizhixin]

What if we could design nature's intricate, interwoven structures with the click of a button?

This Nature Communications paper has cracked the code.

They used generative AI and active learning to create a massive 3D dataset of "bicontinuous" structures. These are porous, continuous networks where solid and void phases perfectly intertwine.

This AI-generated library allows for the fast, inverse design of multifunctional materials, outperforming traditional methods in computational speed and eliminating complex post-processing. It's already being used to design better bone implants, chair components, and materials with tunable stiffness and permeability.

Data-driven inverse design of multifunctional bicontinuous multiscale structures

Paper: https://t.co/Mz4bmfreVi 
Dataset: https://t.co/T4WEroSDeu 
Code: https://t.co/PFYuNZSsXQ

Our report: https://t.co/itpLrmin01

📬 #PapersAccepted by Jiqizhixin

[ZeroToRobot @ZeroToRobot]

Inspection and maintenance robotics represent a significant advancement in engineering, leveraging AI and innovation to enhance efficiency and reliability across various industries.

[Ronald van Loon @Ronald_vanLoon]

Why Inspection and Maintenance #Robots Are Taking Over
by @lukas_m_ziegler

#Robotics #Engineering #ArtificialIntelligence #Innovation #Technology

[ZeroToRobot @ZeroToRobot]

此举表明谷歌致力于通过AI Studio增强Firebase的功能，有望显著简化应用开发流程并提升开发效率

[🚨 AI News | TestingCatalog @testingcatalog]

Google is working on integrating Firebase with AI Studio to enable users to build more complex apps.

Users will be able to vibe code a proper database integration and an auth layer. This is accompanied by an updated UI for the Build page and support for / commands.

[ZeroToRobot @ZeroToRobot]

Speedy-Splat的优化提升了高斯飞溅技术的效率，有望进一步推动其在实时场景重建中的应用

[Ryohei Sasaki@engineer @rsasaki0109]

Speedy-Splat: Fast 3D Gaussian Splatting with Sparse Pixels and Sparse Primitives
https://t.co/7rhT5b73xC
Abstract: 3D Gaussian Splatting (3D-GS) is a recent 3D scene reconstruction technique that enables real-time rendering of novel views by modeling scenes as parametric point clouds of differentiable 3D Gaussians. However, its rendering speed and model size still present bottlenecks, especially in resource-constrained settings. In this paper, we identify and address two key inefficiencies in 3D-GS, achieving substantial improvements in rendering speed, model size, and training time. First, we optimize the rendering pipeline to precisely localize Gaussians in the scene, boosting rendering speed without altering visual fidelity. Second, we introduce a novel pruning technique and integrate it into the training pipeline, significantly reducing model size and training time while further raising rendering speed. Our Speedy-Splat approach combines these techniques to accelerate average rendering speed by a drastic 6.71x across scenes from the Mip-NeRF 360, Tanks & Temples, and Deep Blending datasets with 10.6x fewer primitives than 3D-GS.