Searcharxiv⌕ Search

arXiv subjects

Chao Peter Yang

Publications and source records attributed to Chao Peter Yang.

5 recordsLinked to original sources

Don't Blame The Preferences: Examining The Emergence of Sycophancy in LLM Training

AI sycophancy, the tendency of large language models (LLMs) to flatter users and affirm their beliefs, can undermine accuracy and contribute to misinformation, dependence, and overconfidence. Although prior work has benchmarked sycophancy in released models and proposed mitigations, its origins during training remain poorly understood. Preference alignment methods such as RLHF and DPO are often blamed, but their effects are rarely compared with earlier training stages. We investigate sycophancy across the training pipelines of three model families and three benchmarks. We find that sycophancy is largely established by the end of supervised fine-tuning (SFT), rather than introduced by preference alignment. Across all nine model benchmark combinations, SFT checkpoints attain at least 92% of preference aligned models' sycophancy scores and at least 85% of final models' scores; on average, they are about as sycophantic as final models. Preference alignment increases sycophancy in only four of nine cases, by an average of 4.2 percentage points when it does. To probe the underlying causes, we isolate model representations associated with sycophancy using mechanistic interpretability and measure their activation on training datasets from each stage. SFT datasets show roughly the same average activation as other training data but contribute substantially more examples, and the highest-scoring individual datasets all come from SFT. Our findings suggest that addressing sycophancy requires interventions throughout the training pipeline, particularly in data selection and augmentation for pre-training and SFT.

cs.AI↗

StepSafe: Learning Timely Intervention with Context-Aware Safety Reasoning

Safety in agentic systems is inherently sequential: whether an action is safe depends on prior actions, tool outputs, information disclosures, and interactions among agents. Recent work has shifted agent safety guardrails from post-hoc trajectory-level assessment toward step-level action diagnosis and proactive risk prediction, yet existing training objectives provide limited supervision for how step evidence accumulates over time and when it becomes sufficient to justify intervention. We propose \methodlong~(ASPO): a training framework for step-level timely intervention. Pre-fix level stage alignment allows crediting each decision at its own step: it compares each response with the other responses sampled at the same prefix and rewards the action that fits the stage, continuing before the risk is visible, asking while clarification can still avert harm, and stopping before harm is irreversible. Adaptive feedback self-distillation further credits the reasoning behind each decision, with the model itself as a teacher that sees the step annotations on intervention steps and rules mined from its own false alarms on safe steps. Its training data carry a violation taxonomy and an intervention window at every step and are extended through self-evolving multi-agent red-teaming. Finally we present \textsc{StepSafe}, a step-level reasoning guardrail with timely intervention, that achieves the state-of-the-art peformance among open-source guardrails across five agent-safety benchmarks on safety-utility pareto curve with Youden's $J$ of 53.0 (4B) against 46.2.

cs.AI↗

Getting Motif-ated: Controllable AI Compositions from Injected Motif Prompts

Deep learning has transformed symbolic music generation by borrowing the training paradigms of large language models, with systems such as NotaGen now producing complete, stylistically convincing classical scores from a short prompt. These systems could become powerful creative partners, helping musicians generate endless possibilities. However, current systems expose almost no control handles on the music itself. In principle, control handles could be built into a foundation model trained from scratch, but this is rarely practical without massive amounts of quality annotated data and compute resources. We therefore present MotiGen, a recipe for retrofitting pretrained symbolic music models to use new instruction prompts. MotiGen injects a musical motif as a structured prompt line, reinforces it with a scalar attention bias toward the motif tokens, and learns the association with a two-phase curriculum. First, it learns from focused excerpts cropped around motif occurrences in the training data, then full scores including the motifs. Our experiments show that our model composes with the prompted motif in over 92.3\% of generated pieces. Generated pieces using a variety of motifs are included in our sample site: https://motigen-site.github.io/.

cs.LG↗

Auditing Engagement Incentives in the Kidfluencer Ecosystem: A Multimodal Weak Supervision Approach

The rise of `kidfluencers' on YouTube has raised ethical concerns about child digital labor and exploitation. While emerging legislation attempts to regulate this ecosystem, empirical evidence linking exploitation to engagement remains scarce, given the difficulty of operationalizing exploitation at scale. This study presents a multimodal AI audit of 5,051 videos across 79 kidfluencer channels, using weak supervision to detect exploitation signals without large-scale manual labels. We aggregate noisy labeling functions -- including LLM-based classification of titles and GPT-4 Vision analysis of thumbnails and descriptions across six literature-grounded dimensions -- to assign a probabilistic exploitation score to each video. A multi-annotator validation study (N=107) shows strong agreement with human judgment (macro-average F1 $= 0.911$) and high sensitivity for overall exploitation risk (recall $= 0.960$, F1 $= 0.793$). Our findings reveal a significant engagement premium for performative labor, emotional bait, and privacy violations. Exploitation scores correlate with view counts (Spearman $ρ= 0.229$, $p < 10^{-50}$), and mixed-effects regression controlling for channel-level variation shows that a one-unit increase in exploitation score yields a $4.4\times$ increase in views ($p < 0.001$). Within-channel analyses indicate median view boosts of $+65.6\%$ for emotional bait and $+56.0\%$ for performative content (FDR-corrected $p<0.001$), with effects holding in same-year robustness checks ($p=0.030$). Explicit commercial content (product placement), by contrast, shows no premium ($-3.8\%$, n.s.), suggesting the platform rewards commodification of the child's identity and labor over traditional advertising. These findings challenge policy frameworks focused solely on financial trusts, showing that engagement is systematically tied to the intensive, performative labor of children.

cs.CY↗

FinCom: A Financial Multi-Agent Demo with Disagree-or-Commit Deliberation

Multi-agent systems powered by large language models (LLMs) are increasingly used for financial analysis and decision support. However, existing coordination schemes, especially those emphasizing consensus or debate, are vulnerable to sycophancy: agents conform to peer reasoning instead of evidence, leading to premature agreement and degraded outcomes. We introduce FinCom (Financial Committee), a governed multi-agent framework and interactive system that operationalizes the Disagree-or-Commit (DoC) protocol to embed structured dissent into financial AI committees. A central Supervisor orchestrates three ReAct-enabled specialist agents: Research, Quantitative, and Risk. Each agent is equipped with role-specific tools for retrieval, computation, and stress testing. During deliberation, agents must either explicitly critique or commit to their peers' reasoning before converging on a unified recommendation. This demonstration showcases how FinCom supports committee-style financial analysis through coordinated multi-agent interaction, including structured report generation and interactive decision support. Evaluated across the most recent financial agent benchmark, in addition to 90 internal handcrafted financial tasks using an LLM-as-a-Judge protocol, DoC improves reasoning accuracy and risk awareness significantly over a consensus-seeking baseline on both an in-house and external evaluation set. By reframing disagreement as a governance primitive rather than noise, FinCom offers a lightweight, prompt-only recipe for improving accountability, transparency, and epistemic robustness in agentic financial systems.

cs.MA↗