When the meetings are recorded, can it be in a database so that is queryable or comes up with solutions to prior problems or new areas of exploration?
Can Handoff heavy be managed well if about 300 claude agents manage the design?
Knowing exactly what to remember and what to forget and when to be "Fifty first dates"
https://readable-intelligence.com/
https://felixrieseberg.com/how-to-be-more-ambitious-a-tutorial-for-humans-with-models/
https://www.chiplog.io/p/why-stacked-3d-dram-is-the-right
https://www.thonking.ai/p/what-shapes-do-matrix-multiplications
https://www.thonking.ai/p/strangely-matrix-multiplications - flops/watt
https://siboehm.com/articles/22/CUDA-MMM
https://x.com/bradenjhancock/status/2104209695200485432 - shape of language model
Compiler and Hardware Co-Design for Accelerator Architectures
https://udlbook.github.io/udlbook/ - understanding deep learning
compilers for machine learning
https://nanjiang.cs.illinois.edu/2026/09/30/proof.html - Breaking down proofs
https://x.com/shao__meng/status/2099281275370754227/photo/1 notes
reinventing ed as next token pred
https://www.interconnects.ai/p/open-source-ai-reading-list
https://deep-learning-alchemy.github.io/ from 9/28
https://x.com/gordic_aleksa/status/2099884333305860516 - "double descent can be understood as a phase transition between memorizing individual examples and learning reusable structure from the data"
https://web.stanford.edu/class/msande319/ -
