RESEARCH

→ View original post on X — @tunguz

Training XGBoost Models for Machine Learning

By

@tunguz

–

24 March 2026 23h15

Anyways, back to training some super duper XGBoost models. @trainxgb

→ View original post on X — @nandodf

Good wishes for AI resilience research work

By

@nandodf

–

24 March 2026 23h14

Best of luck @woj_zaremba – I’m glad that smart people like you are working on AI resilience.

Large Scale Training Infrastructure Kernels and Distributed RL Scaling

By

–

24 March 2026 23h09

We go into detail about the infrastructure behind large scale training including the kernels we developed and open-sourced for the project. We also discuss distributed training and environment scaling for RL.

Cursor Introduces CursorBench for Realistic Coding Problem Evaluation

By

–

24 March 2026 23h09

We describe our internal benchmark CursorBench which represents a more realistic sampling of coding problems. We discuss why we think it is important to include the complex problems software engineers see everyday.

Reinforcement Learning Phase Critical for Model Performance

By

–

24 March 2026 23h09

The reinforcement learning phase is critical for final performance. We discuss the algorithms we apply for this stage. We find that simple approaches often work best, and improve performance broadly.

Composer 2 Uses Continued Pretraining and RL for Coding

By

–

24 March 2026 23h09

Composer 2 had three main efforts: continued pretraining, reinforcement learning, and benchmark development. The goal of each was to closely emulate the Cursor environment to produce a highly intelligent coding model.