Adept → Amazon and ML @ Scale Parts 2+3
Exciting AI move, ML at Scale saga, and experimental research projects showcased in this edition.
Hello friends!
🌊 Life Updates
Big life update: I’ve moved from Adept to Amazon’s AI team, where I’ll be doing all things post-training and RL for agents.
My personal interests remain the same: I want to improve model performance on multi-step reasoning and long-horizon planning. Excited to explore this new chapter!
🛠️ Explainers
We’ve finished our ML at Scale saga. Read the three parts here: data parallelism, tensor parallelism, and pipeline parallelism.
🧪 Experimental Research
I’ve also been trying my best to carve out some time to do some experimental research of my own! A couple interesting projects in the works:
combining monte-carlo tree search with llm inference doesn’t seem to create the much celebrated boost papers brag about
models can’t math
spinning up an easy to use open source model evaluation framework
As always, you can reach me at hi@amks.me.