This study examines representation bias in open-source Qwen models for investment decisions.
problem Representation bias in financial applications of large language models.
method Balanced round-robin prompting over 150 U.S. equities, constrained decoding, token-logit aggregation.
result Firm size and valuation increase model confidence, while risk factors decrease it.
CoE modularizes LLMs for scalable, cost-effective AI systems.
problem Challenges in scalability, cost, and customization of monolithic LLMs.
method Composition of Experts (CoE) leverages multiple expert LLMs with a router for dynamic selection.
result CoE achieves superior performance with reduced computational overhead and cost.
Paper introduces a trading agent using LLMs for risk assessment and trading recommendations.
problem Developing a trading agent that can handle financial risks effectively.
method Extending CPPO algorithm with LLM-generated risk assessment and trading signals from financial news.
result Backtesting shows improved performance of the trading agent compared to benchmarks.
Deleting refusal directions from models leads to systematically more optimistic decisions.
problem The impact of removing refusal directions from models on decision-making outcomes.
method Ablation study using a frozen pipeline of 21,600 weekly equity decisions.
result Ablation of refusal directions makes models more optimistic and justifies themselves more, but also reduces confidence.
System detects financial misinformation and generates clear explanations.
problem Identifying and explaining fraudulent financial content.
method Combined large language models, pre-processing, and sequential learning.
result Achieved F1-score of 0.8283 for classification and ROUGE-1 of 0.7253 for explanations.
Enhances LLM quantization with MDBF, improving perplexity and accuracy.
problem Limited performance of Double Binary Factorization in extreme quantization.
method Introduces Multi-envelope DBF, retaining sign matrices and replacing single envelope with rank-l envelope. result Improves perplexity and zero-shot accuracy over previous binary formats.
The study examines how verifier imperfections impact test-time scaling techniques.
problem Understanding how verifier imperfections affect test-time scaling methods.
method Proves the instance-level accuracy of Best-of-N and Rejection Sampling methods using the geometry of the verifier's ROC curve.
result RS outperforms BoN for fixed compute, but both converge to the same accuracy in the infinite-compute limit.
IVON improves LoRA finetuning with minimal overhead and significant accuracy gains.
problem Improving LoRA finetuning with Bayesian methods.
method IVON, a variational algorithm, with posterior pruning.
result Significant accuracy improvements over AdamW and other Bayesian methods.
Study shows training duration impacts model merging quality, suggesting joint selection of duration and method.
problem Impact of expert training duration on model merging quality for large language models (LLMs).
method Systematically fine-tuned experts on five domains across three model sizes, evaluating five merging methods at each duration.
result Training duration affects merging quality, with simple averaging degrading sharply and sparsification-based methods performing well past the validation optimum.
Study shows training duration affects model merging quality, suggesting joint selection of duration and method.
problem Impact of expert training duration on model merging quality for large language models (LLMs).
method Systematically fine-tuned experts on five domains across three model sizes, evaluated five merging methods at each duration.
result Training duration and merging method should be chosen jointly, not independently.
LLMs can fail to maximize aligned values even after training, due to irrational reasoning.
problem Value misalignment in LLMs' reasoning despite training alignment.
method Formalized rational value risk and decomposed estimation error.
result Rational value risk is widespread and cannot be fully eliminated.
SLED improves factuality in LLMs without external knowledge.
problem Unreliable or factually incorrect outputs from large language models.
method Contrasts final layer logits with early layers' logits, uses approximate gradient to refine outputs.
result Consistently improves factual accuracy over existing methods.
TENP prunes experts and neurons in Mixture-of-Experts models for efficient deployment.
problem Efficient deployment of large language models constrained by static parameter footprint.
method Structured Trapezoidal ExpertNeuron Pruning (TENP) identifies and retains important experts and neurons.
result DeepSeek model achieves 10% better performance on code generation tasks with 40% expert sparsity.
New methods improve LLM preference optimization by intelligently weighting multiple reference models.
problem Improving LLM preference optimization with multiple reference models.
method Introducing four new weighting strategies for multiple-reference preference optimization.
result All four new weighting strategies outperform current methods on preference accuracy.
Stable and consistent model alignment for language models without assuming human preference models.
problem Lack of statistical consistency in existing alignment methods.
method Relative density ratio optimization between preferred and mixture of preferred and non-preferred data distributions.
result Our approach achieves statistical consistency and stability, providing tighter convergence guarantees.
A new protocol corrects confounding effects to measure alignment-induced activation shifts accurately.
problem Confounding effects in measuring alignment-induced activation shifts using naive methods.
method Introduces a four-variant decomposition to separate alignment shift from template effects.
result Correctly measures alignment-induced activation shifts, recovering behaviorally active subspace.
New analysis shows interpretability doesn't guarantee steering utility in LLMs.
problem Does higher interpretability lead to better steering utility in large language models?
method Trained 90 SAEs across three LLMs, evaluated interpretability and steering utility, used Kendall's rank coefficients for analysis.
result Interpretability is only weakly associated with steering utility, and features selected by Delta Token Confidence improve steering performance.
CoDistill-GRPO improves small models in GRPO by distilling knowledge from a larger model.
problem Small models in GRPO struggle with sparse rewards on difficult tasks.
method Simultaneously trains a large and small model using co-distillation and GRPO objectives.
result Significant improvement in small model performance over standard GRPO on mathematical benchmarks.
Signed-permutation coordinate transport improves model alignment across checkpoints.
problem Improper alignment of coordinate-indexed objects across model checkpoints.
method Introduces sign-marginalized Hungarian matching and coordinate-preserving transport.
result Recovering signed-permutation gauge improves coordinate alignment and model performance.
Train a lightweight carry-on model on existing LLMs for faster customization.
problem Customizing large language models for specific tasks is computationally expensive.
method Train an additional branch of transformer blocks on the final-layer embedding of pretrained LLMs, then merge them with a carry-on module.
result Training a 100M carry-on layer requires less than 1GB GPU memory, making it scalable and affordable.
Mamba outperforms Reformer in minute-level stock prediction using LLM sentiment scores.
problem Improving minute-level stock market prediction accuracy in volatile markets.
method Combining sentiment scores from top LLMs with stock price data, training Mamba and Reformer models.
result Mamba achieved lower error rates across all tested LLMs, especially with LLaMA 3.3--70B.
IPO Finance Agent evaluates LLMs on SpaceX IPO due diligence, surpassing Finance Agent v2.
problem Evaluating language models on financial tasks like IPO due diligence.
method Introducing IPO Finance Agent and an evaluator-optimizer pipeline.
result The best-performing model reaches 79.4% accuracy at 0.30 USD per query.