I'm starting to get really proficient in the "spot the difference" game, detecting small but significant deviants between what I prompted and what the image model produced. I think we'll soon all be experts at this.

Called it

It's been an interesting 10 years!

I’ve been teaching my heart out for the past decade+. I’m honored to share that today, these efforts were recognized with NYU’s highest university-wide honor for teaching excellence: the Distinguished Teaching Award. nyu.edu/academics/awar…

Thanks to my colleagues and students at the Center for Data Science for the kind and generous support. I feel fortunate being able to share my passion for finding insights in data so deeply.

Do you also get an afterimage when looking at this? Courtesy Akiyoshi Kitaoka

GPT is what happens when you do a sextillion dot products right

It is debatable whether a Gorilla would prevail in a fight with 100 men. But I am very confident that the men would *see* the Gorilla - however vicious the beast might be, it would not be able to ambush them. pnas.org/doi/10.1073/pn…

Garden of forking paths strikes again. Still. Endemically. Very sad.

When I was a child, I explored the horrors ensuing from repeatedly making photocopies of photocopies, seeing the information in the original slowing draining away. This issue has taken on a new urgency as LLMs are increasingly training on LLM created outputs, leading to

CDS Clinical Professor Pascal Wallisch was selected as one of only four NYU Trailblazers for 2025! His presentation explored what we learned from disagreeing about #theDress for over a decade. His research at Fox Lab examines how people construct their subjective reality.

No personality test is good at predicting life outcomes, but astrology predicts quite literally nothing.

2 weeks without smartphone internet significantly improved sustained attention. The effects were similar to being a decade younger.

More than 50% of high-profile scientific findings cannot be independently reproduced by other scientists. Lab research is the worst, with irreproducibility for top publications reaching nearly 90%. This is a waste of more than $28,000,000,000 from taxpayers in America alone.

And it was on a Wednesday, no less

All current ML approaches need ridiculously large amounts of training data.

Boring sun, big moon, cold Jupiter, lucky dirt. This place is truly special. Unique.

Hopefully, this is of use to someone

Brutal