The frontier lab building Large Tabular Models for enterprise prediction
San Francisco
Joined January 2025
- Tweets97
- Following3
- Followers344
- Likes28
Faster power restoration starts before the storm hits.
Better predictions are changing how utilities prepare for the worst, and showing what the right information is worth when it matters most.
Large tabular models are opening up a new frontier in AI research.
For PhD students choosing where to focus, the opportunity to help shape an emerging field is hard to ignore.
The result? More than 20% less prediction error than the existing approach.
And performance consistent enough across all three targets to make removing the physical hardware viable.
The idea was simple: use the data the car already collects to predict what a physical sensor would measure.
With NEXUS, we tested that approach across three slip-sensor targets.
The tradeoff: sensors provide critical data, but they also add weight, wiring, and complexity.
In a sport obsessed with marginal gains, that's a problem worth solving.
In Formula 1, every gram matters.
But what if you could remove a physical sensor without losing the signal it provides?
We put that question to the test 🧵
Many claim state-of-the-art results, but how do you know what actually works?
Former Dataiku VP of Research @dsleo explains why rigorous benchmarking is essential to meaningful progress in machine learning.
Formula 1 is a precision sport. A few percentage points in predictive accuracy can mean a very different result on track.
“All the words, all the phrasing, all the sentences in my blog posts are handwritten,” says Google DeepMind’s @sedielem. While he uses AI to test phrasing and dissect complex research papers, he ensures every final word remains entirely his own.