Answer.AI - A few tips for working on high-surface-area problems

A few tips for working on things that involve a lot of moving pieces, that have a high “surface area”.

A few tips for working on high-surface-area problems – Answer.AI Note A note from Johno I’ve been writing brief notes on my own blog for years. I’ll be posting on the Answer.AI blog too now that I work here – here’s what is hopefully the first of many posts. This one is just a few thoughts inspired by some recent work, which I hope you’ll find interesting. I’d love to hear what you think; you can find me here on Twitter/X . Some of the many questions/directions we faced when trying to get QLoRA working with FSDP - an example of a high-surface-area problem! Some problems are fairly well-defined

saved by

• Benedict Neo
• Amos You

related reading

• How to Land a Frontier Lab Jobvladfeinberg.com
• Tips for Empirical Alignment Research — AI Alignment Forumalignmentforum.org
• Tips and Code for Empirical Research Workflows — AI Alignment Forumalignmentforum.org
• What's new | Updates on my research and expository papers, discussion of open problems, and other maths-related topics. By Terence Taoterrytao.wordpress.com
• Tips and Code for Empirical Research Workflows — LessWronglesswrong.com
• Tips and Code for Empirical Research Workflows — LessWronglesswrong.com
• IsoCompute Playbook: Optimally Scaling Sampling Compute for RL Training of LLMscompute-optimal-rl-llm-scaling.github.io
• How We Use Claude Code Skills to Run 1,000+ ML Experiments a Dayhuggingface.co
• GitHub - google-research/tuning_playbook: A playbook for systematically maximizing the performance of deep learning models. · GitHubgithub.com
• Sara Fish • LLM Evals Research Advicesara-fish.github.io
• Sara Fish • LLM Evals Research Advicesarafish.com
• Harvard CS197: AI Research Experiencescs197.seas.harvard.edu