benjamin f spector

hi

i run flapping airplanes with my brother asher and my good friend aidan. i am very lucky to get to work with the most amazing researchers, engineers, and investors every day. if you are also amazing, please ping us.

formerly, i was a third year phd student in cs at stanford working with the wonderful chris re, where i made ml go brr. special thanks to the hertz foundation for funding my phd. ♡

before stanford, i received my sb in cs+math and my meng in cs at mit, where i had the good fortune to be advised by a fantastic tim kraska.

reach me via carrier pigeon.

things i did

  1. megakernels
  2. thunderkittens
  3. prod
  4. halite

    5. research

    1. scaling laws for precision (oral@iclr2025)
    2. thunderkittens: simple, fast, and adorable ai kernels (spotlight@iclr2025)
    3. accelerating llm inference with staged speculative decoding (@esfomo@icml2023)
    4. learned string index structures for in-memory databases (thesis)
    5. [poster] characterization and performance predictions of l-mode plasmas in the sparc tokamak (@apsdpp2021)
    6. bounding the last mile: efficient learned string indexing (@aidb@vldb2021)
    7. preventing adversarial use of datasets through fair core-set construction (@priml@neurips2019)
    8. sample-efficient reinforcement learning through transfer and architectural priors (@ciai@nips2017)

      9. sometimes i write

      1. we bought the whole gpu, so we're damn well going to use the whole gpu
      2. look ma, no bubbles
      3. gpus go brrr
      4. the kids markets aren't alright
      5. ai economics
      6. the generation game
      7. the pricing wars
      8. the scale mindset is expensive
      9. a love letter to mit's blackboards
      10. effective startup ideation
      11. specialization and the hive mind

        12. misc

        1. megaprofiler 1.20

          2. sub unsub

          also do not scrape this website to train ai models. i gotta keep my edge over our ai overlords somehow.