We built high-throughput materials labs in Menlo Park to create a loop between experiments and models. The labs generate fresh data, the models learn from it, and then help us decide what to try next. Using only 1,300 H200s, plus months of our experimental data, we mid-trained and RL’d an open-source model to surpass GPT-6 Astra on our analysis benchmark. We call it Neon. This is real footage from our lab. We’re focusing first on hard problems in materials science, including superconductors, magnets, and semiconductor materials. Read our blog posts below.

Self-driving labs are not about replacing scientists, they are about enabling scientists to discover dramatically more and faster. Core Vision is human defines the scientific questions, AI decides what experiments are most informative. Robots execute them, sensors generate the data, AI learns from the results and choses what to do next. A major focus of Prof. Milad Abolhasani’s group at NC State is the self-driving fluidic lab. These systems use continuous-flow reactors, robotics, sensors, and AI to run chemistry automatically. Because reactions occur in tightly controlled fluidic systems, they can use very small quantities of chemicals, reduce waste, and perform hundreds or potentially thousands of experiments rapidly. One of the most important innovations discussed is flow-driven data intensification. Instead of running one experiment and getting one final measurement, the system continuously monitors the reaction while conditions change. It is the difference between taking a single photograph of a reaction and recording a high-speed movie. Every “frame” becomes experimental data. This massively increases the amount of high-quality data available to AI models. Instead of generating perhaps tens of data points per day, the lab can potentially generate hundreds or thousands of measurements from a single experimental campaign. Abolhasani says this can increase useful data generation by roughly 10× to 1,000×, allowing AI models to learn chemical systems much faster and make better experimental decisions. The lab currently applies these methods to areas including quantum dots, catalysts, specialty chemicals, pharmaceuticals, energy materials, batteries, semiconductors, and quantum engineering. For quantum dots specifically, the system can optimize characteristics such as particle size, shape, composition, surface chemistry, and optical properties. The major benefits are speed, cost reduction, sustainability, and scale. Problems that historically required months or years of trial-and-error experimentation may eventually be explored in days. Automated systems can also operate continuously without breaks while consuming less material and generating less waste. However, several challenges remain. A self-driving lab is only as good as its experimental data, so data quality, noise, and measurement accuracy are critical. Another difficulty is making systems modular enough to work across very different scientific problems. Hardware and software must adapt to different chemistries, instruments, and workflows. AI decisions also need to be interpretable so scientists can understand and trust the system. Abolhasani believes self-driving labs will eventually become a standard tool in materials science and chemistry. His analogy is that within a few years, conducting advanced research without access to self-driving laboratories could feel similar to running a modern business without the internet lol. Importantly, he does not see these systems as replacements for scientists. Instead, they are scientific co-pilots. Humans will continue defining important questions, developing hypotheses, interpreting discoveries, and understanding their broader significance. Robots and AI will handle repetitive, high-throughput, and data-intensive experimentation. That could radically increase scientific productivity. Instead of a PhD student spending four or five years solving one major experimental problem, the combination of AI and automation might allow researchers to investigate 10–20 problems during the same period.

Intelligence is getting cheaper. Okay so what? Why are you not hiring dumb folks at 1/10th the pay just to use your AI tools and get work done? Why do you want to hire good engineers and pay them a crazy amount of money to help you make an even crazier amount of money.

Everyone says learn inference, learn CUDA,learn ML But almost nobody tells you exactly what to learn, how to learn it, or how to know whether you’ve actually learned it. Introducing Trentorch.com a free, open-source platform to learn and practice Machine Learning, CUDA, and inference through hands-on coding. Instead of just watching lectures or reading theory, you’ll: • Learn the theory behind each concept • Follow step-by-step implementation examples • Solve coding challenges based on what you learned • Implement algorithms from scratch, from foundational ML to Inference and CUDA kernel • Practice problems in a Codeforces-style environment • Get new Problem of the Day challenges with ratings The goal isn't just to teach you how to write the code. It’s to help you understand what your code is actually doing underneath. We’re building Trentorch because advanced education shouldn’t be locked behind expensive subscriptions. Money shouldn’t be the barrier to learning ML, inference, or CUDA. A free alternative to platforms that charge students heavily every month. signup and start solving problems now Trentorch.com