We work with top frontier AI labs building RL environments for computer use and world class coding data -- g2i.ai
Organizers of @ReactMiamiConf@AIEMiami
If your verifier passes bad work, your agent learns to repeat it.
In RL, the verifier is part of the training data. Humans have to decide what counts as done and what evidence proves it.
At G2i, that's the human data work we care most about.
This agent understood every update and still submitted the wrong sheet.
At G2i we build human data for frontier labs, and this is the gap we care most about: how experts notice the task changed, recheck, and verify before they submit.
Clean demos never teach it.
I'm looking for 10 engineers. You'll work directly with a top AI frontier lab. Privacy engineering. $200 - $300 /hr for 3 months (ability to extend to 6).
Ideal fit: Staff level eng in the USA. Python/distributed systems experience.
Apply here: jobs.ashbyhq.com/g2i/b2f1846c-aโฆ