Persistent memory offers modest gains for AI drawing Chinese characters
Persistent Memory and Iterative Improvement in an LLM Agent for Chinese-Character Drawing
Artificial Intelligence
Summary
The paper investigates whether an AI agent that remembers its past attempts can improve its skill in drawing Chinese characters. The researchers tested different memory designs and found that keeping free-form notes helped the AI do better on its first try. Combining notes with reusable code also helped, especially on harder characters. When the AI got extra detailed instructions for each character, it improved much more, showing that clear guidance can be very helpful.
What this means in practice
- •For ai software developers: Improve AI performance on iterative tasks by designing memory systems that store notes and reusable code.
- •For educational game designers: Create AI tutors that progressively improve at teaching Chinese characters by recalling past drawing attempts and code patterns.
Authors
Peilin Wu, Yi-Zhuang You
Abstract
Does persistent memory accumulated from previous attempts help an LLM agent solve new problems, and does memory design affect that benefit? We study programmatic Chinese-character drawing across five agent conditions attempting the same 668 items. Accumulated memory produces modest aggregate gains despite substantial growth in the memory stores: free-form notes improve first-attempt success by 4.6 percentage points, while a design combining free-form memory with reusable code improves it by 2.8 points. The two designs do not differ significantly overall, but the code bank shows a suggestive advantage on later, harder items. A separate post-hoc comparison finds a much larger improvement when the code-bank agent receives privileged, target-specific structural supervision. The agent can therefore exploit highly informative structural guidance, although the privileged per-item information does not establish what could have been learned from experience alone. These results suggest that memory design may affect where accumulated experience becomes useful, even when its average benefit is modest.