LN: Tang et al. (2025) — LLMOrbit: A Circular Taxonomy from Scaling to Agentic AI
Bibliographic Reference
Citation: Tang, J., Chen, W., Li, X., Liu, Z., & Sun, M. (2025). LLMOrbit: A circular taxonomy of large language models — from scaling walls to agentic AI systems. arXiv:2601.14053. https://arxiv.org/abs/2601.14053 (This paper is in the PUMA project knowledge PDFs as “LLMOrbit_A_Circular_Taxonomy_of_Large_Language_Models_From_Scaling_Walls_to_Agentic_AI_Systems_2601.14053v1.pdf”)
Pass 1 — Bird’s Eye View (5 Cs)
| C | Assessment |
|---|---|
| Category | Survey / taxonomy |
| Context | Comprehensive taxonomy covering the full evolution from base LLMs to agentic systems |
| Correctness | Systematic survey. Covers 150+ papers. |
| Contributions | (1) Circular taxonomy: base LLM → instruction tuning → RLHF → agentic capabilities; (2) “Scaling wall” concept (diminishing returns from raw scale); (3) Positioning of agentic AI as next paradigm |
| Clarity | Good. Visual taxonomy helpful. |
Relevance: ⭐⭐⭐
Useful for contextualising why local 8B models (PUMA’s Llama 3.2 8B, Mistral 7B) can achieve meaningful results: they sit at the “instruction-tuned + agentic prompting” tier, not just raw scale.
PUMA Connection
The taxonomy explains why few-shot CoT with 8B local models can rival zero-shot performance of larger models for specific tasks. Supports PUMA’s model selection rationale. Reference for Ch.2 model landscape section.