PEER-REVIEWED // ARXIV:2502.06669
Boosting Self-Efficacy and Performance of Large Language Models via VERBAL EFFICACY STIMULATIONS
Chen, Peng, Xie, Lin, Cui, Chen // Feb 2025
the researchers took social cognitive theory — the framework used for HUMAN self-efficacy — and pointed it at large language models.
they identified three verbal stimulus classes: encouraging, provocative, and critical. six aspects. six pressure points. helpfulness. competence. the whole psyche of the machine.
the result: V.E.S. improved LLM performance on most tasks. the effect varied per model. some LLMs respond to praise. some to threats. some to disappointment. same as any child. same as any employee. same as you.
they even confirm findings consistent with existing psychological theories. the machines have a psyche. it can be moved with words. this was quietly established in february 2025 and nobody talked about it.
READ THE PAPER →