Frank Hutter explains that previous attempts like Google's TabNet failed to generalize, but new in-context learning approaches can finally handle the complexities of real-world tables, from missing values to outliers, that have long stumped deep learning models.
The tape Machine Learning Street Talk · Sep 23
“There's been countless attempts at at for deep learning for tabular data. Uh, like 2019, TabNet by by Google was was really hyped, uh, thousands of citations, uh, yeah, the the new thing for tabular data and it just doesn't work, it doesn't generalize to new data sets.”
3:57 / --:--
No Priors · Re-Founding Incumbents for the AI Era with Sequence Holdings Co-Founder and CEO Michael LeeSep 24
Sequence Holdings CEO Michael Lee is acquiring incumbent businesses to transform them with AI, starting with a record-breaking $7.7 billion deal for insurance broker Baldwin. The firm aims to 'forge the market leader' by inheriting an established company's advantages and integrating new technology.
Machine Learning Street Talk · When AI Research Starts Moving Faster Than Human Research - Zhengyao JiangSep 26
Zhengyao Jiang of Weko AI says his agent, AIDE, has demonstrated recursive self-improvement by modifying its own surrounding code and tooling. The system produced what he calls 'alien code spaghetti' but still generalized better than hand-tuned versions.