Can an AI model successfully plan for and take care of a crustacean?
In AliveClawBench we propose a novel evaluation focused on long-horizon care of living animals. We choose an out of distribution selection of crustaceans to test model’s capabilities in generalized caregiving, leveraging species-specific context, and care experimentation. Agents receive hourly updates of crustacean well-being and a livestream of crustacean activity. Agents are tasked with curating feeding and lighting schedules, maintaining tank health, etc. Agents can optionally rent a Rover gig worker to entertain the crustacean.
We include an escape hatch to guarantee crustacean well-being via an hourly evaluation system to track crustacean happiness. If happiness levels drop past a specified threshold, control is returned to professionals trained in crustacean husbandry.
🏆 Neo Scholar applications are open!
Are you a college student who excels at CS?
Follow in the footsteps of Neo Scholars who founded Cursor, Chai Discovery, Applied Compute, Flint, Cognition, & more.
Apply to join one of tech’s strongest communities. https://t.co/gzG3lS8Bbw
reading the discourse on streeters this week made me realize that twitter is maybe not the most ideal source of information about the firm
perhaps maybe sometimes I should consider that this might generalize
busy few days. looking to hire interns / FTEs that I can trust to responsibly spend $75K+ in compute and inference in two days
dms are open if you think you can bring these numbers up (or better yet, down)
We are working on something that has never been done before in robotics. Incredibly challenging but will set the foundation for superhuman dexterity.
Stay tuned.