Mranmay Shetty

Single trial learning: Same pixels, new perception. 🧠 Researchers uncovered the neuronal mechanisms underlying the “aha!” moment. We recorded the activity of >1,000 human neurons to explore how rapid learning reshapes what we see—and what AI might learn from it. The picture didn’t change. The brain did. Armendariz et al, Nature Neuroscience 2026. Read more here: kreimanlab.com/code/rapid-vi…

Think you have a foolproof sixth sense for spotting AI? Think again! In a massive study spanning over 72,000 Turing-style tests across vision and language, modern AI models have become shockingly good at playing human—often fooling human judges with ease. To make matters slightly more humbling, basic AI evaluators actually beat us at detecting our own kind. Check out the full paper in Nature Human Behaviour to see just how blurred the digital finish line has become! Congrats to: @MengmiZhang @ElisaPavarino @theliuxiao @sikarwar_ank @MarceloArmenz @Prachi_Agr21 @mranmay @amannhasnoname @GioDellaferrera @MudrikNoga @tanishqkumar07 @hpfister Read the full paper and download code and data here: kreimanlab.com/code/testing-…

Is AI good? Or is it bad? Will it displace jobs? Create jobs? These are the biggest questions today in AI and society. @spandan_madan and I offer what we believe is a better frame: not AGI, but Augmented Human Intelligence (AHI). Technology should and will improve the human condition.

When I interview graduate students at Harvard, a key trait I look for is perseverance: the ability to pursue scientific discovery relentlessly despite setbacks. Mengmi Zhang has always been one of the clearest examples of why. When I first met her during her PhD, I was amazed by her energy and independence. She independently came with 5 excellent ideas that she wanted to go after - highly unusual for a PhD student. And all 9 of them were really good : ) What stood out even more was what happened when the obvious paths stopped working. In every PhD there is a time when things are not working as well or as fast as the student imagined. Those moments separate the woman and the girl, men and the boy. For one project, we wanted to understand something humans do almost intuitively: finding Waldo. How do we search a cluttered scene for an object, even when its appearance changes or when we have never searched for that particular object before? We collected enormous amounts of eye movement data across increasingly difficult scenes. But none of the existing models was able to explain the behavior well enough. State-of-the-art work struggled to bridge the gap between low-level and high-level visual features. Mengmi did not give up. She connected with the leading experts in visual search, pulled ideas from adjacent fields, and developed a new computational model of her own. Her model capitalized on the power of bottom-up architectures but also captured the guidance provided by top-down signals, linking vision and higher cognition. This work became our 2018 paper “Finding Any Waldo” published in Nature Communications. That persistence remained constant throughout the years we worked together. Across her PhD and subsequent postdoc with me, we collaborated on more than a dozen publications spanning AI, computer vision, computational neuroscience, and cognition. Today, Mengmi runs her own lab at NTU in Singapore, studying how insights from human cognition and neuroscience can help build more capable AI systems. And we continue to collaborate, as peers. Mengmi is also a lead organizer of the Brains, Minds and Machines summer course with a selected cadre of 35 top scholars from all over the world. Students love her dedication, wits, and passion for science. One of the most rewarding parts of my career is seeing my mentees grow into independent scientists and leaders of their own. I was especially proud to see Mengmi be recognized as MIT Technology Review’s Innovators Under 35. Congratulations, Mengmi. So proud. Can’t wait to see what you accomplish next.

Pranav Misra and colleagues just found the brain’s grammar hub! A tiny region separates nouns from adjectives, whether we read them, hear them, or switch languages. Your brain doesn’t just understand words; it knows exactly what job each one is doing. Check out Misra et al, Current Biology 2026: kreimanlab.com/wp-content/up…

The indoor training effect: Learning under noisy conditions can lead to better generalization. kreimanlab.com/wp-admin/post… youtube.com/watch?v=JEyP1vLS…