Author: Aswin Sarang

Aswin Sarang is a technology professional and entrepreneur working across robotics, artificial intelligence, and automation. He focuses on building practical systems that bridge engineering, strategy, and real-world deployment, with an emphasis on clarity, scalability, and long-term value. His work spans product development, system integration, and technology consulting, helping organizations navigate complex technical decisions and translate emerging technologies into usable solutions. Known for a first-principles approach, Aswin prioritizes fundamentals over hype and execution over speculation. Beyond technology, he maintains a strong interest in human performance, learning, and personal development, bringing a multidisciplinary perspective to both his professional and creative pursuits.

How Many Human Raters Do Your AI Benchmarks Really Need?

Google’s new “Forest vs Tree” study shows the usual 3–5 raters per item make safety and toxicity benchmarks brittle—and offers a roadmap for better N/K trade-offs.

Perseverance’s Latest Drive Was Scripted by Generative AI

NASA let a vision-language model pick every waypoint for two Perseverance traverses—showing how generative AI will plan kilometer-scale drives on future missions.

ChatGPT Flunked a 719-Hypothesis Reality Check

A Washington State University study shows ChatGPT flips answers, misses false statements, and only barely beats chance when judging scientific hypotheses.

DNA Robots Are the Next Supply Chain for the Human Body

Researchers are turning DNA into hinges, springs, and controllers for nanoscale robots that could deliver drugs, trap viruses, and assemble molecular devices.

AI’s Energy Appetite Looks Big—Until You Zoom Out

A new “Watts and Bots” analysis finds U.S. AI workloads consume Iceland-sized energy, but the real story is local grid impact and how operators source power.

Roadrunner’s Skating Biped Shows What Multimodal Locomotion Can Do

A 15-kilogram prototype from the Robotics and AI Institute swaps between wheeled and stepping gaits using a single control policy, hinting at more fluid legged robots.

Tomato Harvest Robots Learn to Judge Pickability Before Making a Move

Osaka Metropolitan University’s harvest-ease model teaches tomato-picking robots to score difficulty, replan angles, and share rows with human crews.

Axplorer Puts AI Math Discovery on a Laptop — Why That Matters for Robotics

$Illustration of mathematicians collaborating with an AI interface, via MIT Technology Review$

Axiom Math open-sourced Axplorer, a PatternBoost successor that runs on a single workstation and could speed up the math that underpins robotics and AI.

Deepfake X-Rays Are Fooling Radiologists. Here’s How Healthcare Should Respond

A new Radiology study shows synthetic X-rays are fooling both clinicians and AI models, raising the stakes for provenance across healthcare and robotics.

5 Predictions About the Future of AI Forensics That’ll Shock You

AI Forensics: Tracing the Origins of Digital Content Introduction In the digital era marked by the exponential growth of artificial intelligence, the significance of AI forensics is paramount. As AI systems, particularly text-to-image models, become prevalent in generating digital content, the authenticity and ownership of these materials come into question. This is where AI forensics […]