@pratyushmaini on Backlist

2 appearances on the backlist front page in the last 30 days.

38.

The community still hasn’t fully absorbed that you can train a 30B MoE on 10T tokens, from scratch, for just 500K

The community still hasn’t fully absorbed that you can train a 30B MoE on 10T tokens, from scratch, for just 500K The big caveat is you need to get data, infra etc right from day 1. Once this ecosystem of enablers is working in unison, own

by @pratyushmaini (Pratyush Maini) · backlist 2026-06-21 · on X ↗

79.

Having worked on unlearning for multiple years, it was clear that post-training "fixes" alone were a dead-end. Mo…

Having worked on unlearning for multiple years, it was clear that post-training "fixes" alone were a dead-end. Model learning is way too entangled. With 𝗡𝗨𝗟𝗟𝘀 we decided to architect unlearnability into the model, and scaled it to 1B+

by @pratyushmaini (Pratyush Maini) · backlist 2026-06-17 · on X ↗