Can AI self-improve on its own and reach superhuman performance? 🧠...
Bonnie Li@bonniesjli
27 views
Dec 11, 2025
~1 min read
Advertisement
1
Can AI self-improve on its own and reach superhuman performance? 🧠
In our Sima 2 paper, we dropped a Gemini agent into an unseen 3D world. The model acted as the task proposer, the agent, and the reward model - autonomously learning from self-generated experience. It surpassed human performance through self-improvement iterations!
In our Sima 2 paper, we dropped a Gemini agent into an unseen 3D world. The model acted as the task proposer, the agent, and the reward model - autonomously learning from self-generated experience. It surpassed human performance through self-improvement iterations!
2
The model act as both the 1) task setter - instructing the agent to perform tasks in the environment 2) agent, executing trajectories 3) and reward model - scoring its own trajectories. This builds a self-improvement flywheel purely through self-generated data.
3
We deploy the self-improvement process in ASKA, enabling the agent to autonomously acquire and improve upon new skills. The agent is capable of progressing significantly further than the SIMA 1 and SIMA 2 agents, despite only ever training on self-generated experience in ASKA. More details in our paper arxiv.org/pdf/2512.04797…
4
Paper w/ @_timharley, Frederic Besse, @VladMnih, @ShaneLegg, @demishassabis & the fantastic Sima team @GoogleDeepMind
arxiv.org/abs/2512.04797
arxiv.org/abs/2512.04797
