As foundation models grow in scale and diversity, coordinating multiple models into cooperative reasoning systems offers a path toward safer, more reliable AI. This chapter presents a multi-agent framework where solver models generate independent drafts, each undergoes structured critique and revision by a critic agent, and an aggregator agent synthesizes a final consensus solution. A scoring module provides semantic, numerical, and procedural evaluation across all agents. Through ablation studies on a benchmark spanning calculus, physics, chemistry, biology, economics, optimization, statistics, and mathematics, we isolate the contributions of framework architecture versus model diversity. We compare four configurations: (1) Individual Baseline, (2) Homogeneous Framework using one shared model, (3) Redundant Homogeneous Solvers using multiple instances of the same model, and (4) Heterogeneous Framework with diverse specialized models. Results show that while framework structure and redundant sampling yield modest gains, model heterogeneity is the critical factor driving substantial performance improvements. The heterogeneous configuration achieves superior step-wise accuracy (0.64 vs. 0.54 for individual models; 2.3x improvement over homogeneous configurations) with reduced variance across categories and difficulty levels. Step-wise reasoning quality (correctness of intermediate steps, not just final answers) improves dramatically only with model diversity, showing that heterogeneous agents provide complementary error detection and reasoning refinement essential for explainability and auditability. We discuss architectural principles, evaluation methodology, and implications for Global Applied AI, showing how heterogeneous multi-agent coordination supports transparent, auditable, high-confidence decision-making across scientific and industrial domains.
Introduction Large language models (LLMs) can generate persuasive narratives that shift agent behavior in multi-agent systems, but deploying raw, unstructured text as an influence mechanism offers no formal guarantees on effectiveness, interpretability, or controllability. Methods We introduce the LLM Influence Compiler (LIC), a solver–critic pipeline that compiles natural-language cooperation directives into structured, parameterized influence policies defined over a five-field schema: network targeting, narrative theme, intensity, deployment timing, and compiler confidence. Each compiled policy is diffused through a network of numerical agents via an exposure model incorporating fatigue decay, susceptibility heterogeneity, and backlash. Evaluation spans nine controlled experiment blocks comprising more than 200 simulation runs across four topologies (Barabási–Albert, small-world, Erdős–Rényi, modular SBM), four network sizes (n ∈ {80, 160, 320, 640}), five LLM backbones, and two non-stationary perturbations. Results Compiled policies raise the mean cooperation rate to 0.826 ± 0.010 (95% bootstrap CI [0.819, 0.834]), an 8.6% relative improvement over the unstructured baseline (0.760 ± 0.005, Mann–Whitney pBonf = 0.040, Cohen's d = 7.90). A controlled decomposition attributes the gain to network-aware targeting (ca. +4.5 pp), dose calibration (+1.9 pp), and an intervention floor (+1.5 pp); the residual contribution of full LLM compilation over a hand-coded rule with the same parameters is statistically indistinguishable from zero under stationary conditions. Under structural non-stationarity (mid-simulation graph rewire), the LLM-mediated re-deployment architecture significantly outperforms a frozen rule (paired t-test p < 0.001, paired Cohen's d = 2.27, 8/8 seeds). Adversarial stress testing confirms that the critic correctly flags 20/20 risky policies as high risk and rejects them, while passing a moderate-baseline policy at medium risk in all 5/5 trials. Discussion The compiler abstraction converts an opaque generative process into a decomposable, auditable policy object whose components can be independently attributed, compared, and governed, laying the groundwork for constitutional oversight of LLM-mediated influence in artificial societies.
Cornelio Yáñez-Márquez, Yikang Lu, J. Curtò et al.· Frontiers in Artificial Inte...· 1 citation
Recent advances in Large Language Models (LLMs) and Vision-Language Models (VLMs) enable the automatic generation of parametric 3D designs from natural-language specifications. This chapter presents an empirical study of foundation models for automatic Computer-Aided Design (CAD) generation of mechanical parts, using a unified evaluation pipeline and a curated benchmark of 97 engineering design problems. We introduce LLMForge, a multi-model text-to-CAD framework integrating JSON-schema validation, analytic feature scoring, mesh synthesis, and multi-round iterative refinement, studied under two critique regimes. IterTracer uses a Phong-shaded ray-trace renderer with analytic visual metrics (silhouette IoU, hole visibility, edge clearance, aspect-ratio conformance) for lightweight geometry-aware feedback across rounds. IterVision replaces the analytic scorer with a VLM semantic critic (Qwen2.5-VL-72B) that evaluates rendered views via chain-of-thought visual reasoning, assessing spatial coherence and design intent. On a benchmark spanning four canonical geometry families (plates with holes and bolt circles, multi-feature boxes, flanged cylinders, and L-brackets), we evaluate seven foundation models: DeepSeek-V3.2, Qwen3-235B-A22B, Llama-3.3-70B, Gemma-3-27B, GLM-4.5, MiniMax-M2.1, and INTELLECT. Under IterTracer, the four highest-ranked models form a tight cluster (overall mean in [0.885, 0.890]) with 98.97% mesh success, showing that compact instruction-tuned models can match substantially larger systems. VLM-based critique in IterVision yields 100% watertight mesh generation on the leading model while surfacing systematic difficulty on rotationally symmetric geometries such as cylinders, where visual and semantic scoring diverge most. We discuss benchmark design, failure modes, CAD-oriented prompting, and implications for industrial workflows and scalable automated mechanical design.
Evaluations of LLM planning agents largely ask whether a task succeeds or a declared plan is followed. In strategic cyber-physical systems, a stronger question is whether the planning architecture remains appropriate after autonomous participants respond and physics constrains the outcome. We introduce a controlled, physics-grounded benchmark built around planning-induced control trajectories: the ordered planning operations and directives through which an execution architecture acts on other agents and the physical process. It implements predefined, sequential, hierarchical, and search executors in a smart-grid demand-response system with 40 heterogeneous prosumers and an independently simulated radial feeder. The LLM is bounded to typed policy declaration and short operator messages, while schedule construction, prosumer dynamics, and power flow remain explicit code. The protocol uses paired forced-mode counterfactuals, common random response draws, and event-level deadline feasibility. Three properties follow. Architecture materially changes outcomes: forced search is the oracle in all five baseline seeds. Execution fidelity needs more than mode agreement: objective substitution holds agreement at 1.0 while increasing voltage shortfall by 2.68x. A 144-scenario, 576-episode bank has feasible oracles from three of the four architectures. A prespecified stress-held-out ridge has mean regret 90.7 (95% interval [73.8, 108.6]) and no detectable value over fixed sequential; applying known deadline feasibility before quality prediction cuts regret to 29.0 and improves over fixed sequential by 61.1. An all-feasible ablation does not beat fixed search, localising the remaining challenge to within-feasible quality selection. A five-model extension separates stress-conditioned, state-blind, and invariant declarers; latency tails show that live feasibility should be treated probabilistically.