Monticule

Yiran Rex Ma

马义然Beijing, China

Linguistics @ Peking University

Independent researcher, Neo Lab founder, Monticule Tech

LLM full-stack, inherent human preferences, whither beyond verification?

Curious aboutContinual learningPersonal modelsEdge-cloud collaboration

Selected Work

研究成果
All

News

最近动态

  1. BlogRenting Parameters: The Engineering, Science, and Political Economy of Continual Learning

    Why continual learning is back in mid-2026: four paths to teaching a frozen giant something new, PEFT’s three-year filigree era and the Bitter Lesson it ran into, and why taste, tone and judgment belong in parameters rather than in prompts.

  2. PublicationCLASE: Chinese Stylistic Evaluation

    A hybrid, reference-free framework combining interpretable linguistic features with experience-guided LLM-as-a-judge scoring, with legal Chinese as the testbed.

  3. NewsCo-founded Monticule

    Founded Monticule with my partner, seeking to scale continual learning and personal models into diverse real-world applications.

  4. PublicationTowards Computational Chinese Paleography

    A position paper on what computational Chinese paleography is becoming: the end-to-end methodological pipeline, bottlenecks, and opportunities.

  5. PublicationPun2Pun: Textual-Visual Chinese-English Pun Translation via Linguistic Reasoning

    A benchmark for pun translation between Chinese and English, with the Constant-Variable Optimization strategy and the Overlap metric; ACL 2025 SRW.

Writing

长文与笔记
All
  1. Renting Parameters: The Engineering, Science, and Political Economy of Continual Learning

    Why continual learning is back in mid-2026: four paths to teaching a frozen giant something new, PEFT’s three-year filigree era and the Bitter Lesson it ran into, and why taste, tone and judgment belong in parameters rather than in prompts.

    Continual Learning

  2. Implementation of LLM-Based Multi-Agent Systems: A Toy Survey

    The notion of the agent is by no means new, but implementing multi-agent systems with LLMs is the genuine issue ahead of researchers today — a toy survey of LLM-based MAS implementation.

    Survey

  3. Attempt to Differentiate: NLP, Computational Linguistics, & Digital Humanities

    At ACL 2024 in Bangkok, this year’s ACL Chair Emily M. Bender put forth a highly controversial conclusion in her address: ‘ACL is not an AI conference.’

    Thoughts