General learned delegation by clones
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917274331381760 |
|---|---|
| author | Li, Darren Chen, Meiqi Shao, Chenze Meng, Fandong Zhou, Jie |
| author_facet | Li, Darren Chen, Meiqi Shao, Chenze Meng, Fandong Zhou, Jie |
| contents | Frontier language models improve with additional test-time computation, but serial reasoning or uncoordinated parallel sampling can be compute-inefficient under fixed inference budgets. We propose SELFCEST, which equips a base model with the ability to spawn same-weight clones in separate parallel contexts by agentic reinforcement learning. Training is end-to-end under a global task reward with shared-parameter rollouts, yielding a learned controller that allocates both generation and context budget across branches. Across challenging math reasoning benchmarks and long-context multi-hop QA, SELFCEST improves the accuracy-cost Pareto frontier relative to monolithic baselines at matched inference budget, and exhibits out-of-distribution generalization in both domains. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_13262 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | General learned delegation by clones Li, Darren Chen, Meiqi Shao, Chenze Meng, Fandong Zhou, Jie Artificial Intelligence Computation and Language Frontier language models improve with additional test-time computation, but serial reasoning or uncoordinated parallel sampling can be compute-inefficient under fixed inference budgets. We propose SELFCEST, which equips a base model with the ability to spawn same-weight clones in separate parallel contexts by agentic reinforcement learning. Training is end-to-end under a global task reward with shared-parameter rollouts, yielding a learned controller that allocates both generation and context budget across branches. Across challenging math reasoning benchmarks and long-context multi-hop QA, SELFCEST improves the accuracy-cost Pareto frontier relative to monolithic baselines at matched inference budget, and exhibits out-of-distribution generalization in both domains. |
| title | General learned delegation by clones |
| topic | Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2602.13262 |