Small LLMs outperform large ones on simple tasks without extra labelling costs.
problem Performance of large commercial models in simple classification tasks.
method Logistic Regression on small LLM embeddings.
result Small LLMs equal or outperform large LLMs in 'tens-of-shot' classification tasks.
This paper simplifies fine-tuning for small LLMs, reducing barriers for developers.
problem Limited resources for fine-tuning large language models (LLMs) by individual developers and small organizations.
method Instruction-tuning datasets, small-sized LLMs (3B to 7B parameters), various training configurations and strategies.
result Improved model performance on benchmarks with specific training configurations, and insights into early termination and hyperparameter simplifications.
The CLT fails for LLM evaluations with small data, leading to underestimation of uncertainty.
problem Inaccurate uncertainty estimates in LLM evaluations with small datasets.
method Alternative frequentist and Bayesian methods for uncertainty quantification.
result CLT-based methods underestimate uncertainty in small data settings.
This work shows fine-tuning improves LLM uncertainty estimates efficiently.
problem Ensuring high-stakes predictions from large language models.
method Fine-tuning on a small dataset of correct and incorrect answers.
result Fine-tuned LLMs produce good generalization and tractable uncertainty estimates.
Diffusion LLMs can efficiently generate harmful prompts for adversarial testing.
problem Generating harmful prompts for adversarial testing is resource-intensive and costly.
method Transformed adversarial prompt optimization into an efficient inference task using pretrained Diffusion LLMs.
result Only a few conditional samples are required to generate harmful prompts with high reward.
LLM evaluation suffers from systematic biases and lacks reliable positive judgments.
problem LLM evaluation suffers from systematic biases and lacks reliable positive judgments.
method Formulate LLM evaluation as a positive-unlabelled learning problem and propose a geometric auditing framework based on Partial Optimal Transport.
result Improved alignment with human preferences, increased robustness to presentation biases, and interpretable confidence estimates.
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.
The paper tests if LLMs' capabilities are executed by small subnetworks (circuits).
problem Understanding how LLMs execute their capabilities.
method Formalized criteria for circuits, developed hypothesis tests, applied to six circuits.
result Synthetic circuits align with idealized properties, while Transformer circuits vary in their alignment.
SmoothLLM defends LLMs from jailbreaking attacks by randomly perturbing inputs.
problem Adversaries can fool large language models into generating objectionable content.
method SmoothLLM randomly perturbs multiple copies of a prompt and aggregates predictions to detect adversarial inputs.
result SmoothLLM sets the state-of-the-art for robustness against various jailbreak attacks.
Dropping a tiny fraction of preferences can significantly alter the rankings of top LLMs.
problem Robustness of LLM ranking systems to small changes in preference data.
method A computational method based on the Bradley-Terry model to evaluate robustness.
result Top LLM rankings can be highly sensitive to the removal of a small fraction of preferences.
CJE calibrates cheap LLM judges against an oracle, achieving high accuracy at a fraction of the cost.
problem Inexpensive LLM judges can produce biased rankings, leading to unreliable outcomes.
method CJE uses a small oracle to calibrate cheap scores, then evaluates at scale with valid uncertainty.
result CJE achieves 99% pairwise ranking accuracy at 14x lower cost compared to a 16x oracle/judge cost ratio.
ChatGPT predicts stock market reactions from news headlines without financial training.
problem Predicting stock price movements using non-financial data.
method Used post-knowledge-cutoff headlines to train ChatGPT-4, which forecasts stock market reactions.
result ChatGPT-4 can predict stock market reactions with high accuracy, especially for small stocks and negative news.
LLMs struggle to optimize hyperparameters efficiently, but hybrid methods can improve performance.
problem Optimizing hyperparameters of small language models using LLMs.
method Comparison of classical HPO algorithms and LLM-based methods, introducing Centaur hybrid approach.
result Hybrid Centaur approach achieves best results, outperforming classical and pure LLM methods.
This paper fine-tunes LLMs for stock return prediction using financial news.
problem Improving stock return forecasting accuracy using LLMs.
method Fine-tuning LLMs with text and forecasting modules, comparing encoder-only and decoder-only models, and integrating token-level representations.
result LLMs' aggregated token-level embeddings enhance return predictions for long-only and long-short portfolios.
Look-Ahead-Bench evaluates financial LLMs for lookahead bias, revealing significant differences in model performance.
problem Measuring and mitigating lookahead bias in financial LLMs.
method Standardized benchmark evaluating model behavior in practical financial scenarios, analyzing performance decay across market regimes.
result Standard LLMs exhibit significant lookahead bias, while Pitinf models show improved generalization and reasoning abilities.
Bayesian approach quantifies uncertainty in LLM-based systems.
problem Uncertainty quantification in LLM-based systems, especially for high-stakes applications.
method Interpreting prompts as parameters in a Bayesian model, using MHLP for inference.
result Improvements in predictive accuracy and uncertainty quantification on various benchmarks.
MoA framework improves financial LLM responses at low cost.
problem Complexity and cost in LLM research for financial domains.
method Mixture of Agents (MoA) framework using customized small models.
result MoA produces higher quality responses at lower costs.
CROQ optimizes LLM decision-making by narrowing down choices and improving accuracy.
problem Uncertainty in LLM outputs poses risks in high-stakes domains.
method Conformal prediction (CP) and optimization (CP-OPT) to minimize prediction set sizes.
result CROQ improves LLM accuracy, especially with CP-OPT.
This paper evaluates LLMs on large graph property estimation tasks.
problem Limited context length of LLMs limits their evaluation on large graphs.
method Developed EstGraph dataset and introduced four tasks for LLMs to estimate large graph properties.
result LLMs perform better on graph property estimation tasks when provided with context-rich prompts based on random walks.
New method corrects quantization errors in LLMs using low-rank matrices.
problem Correcting quantization errors in large language models.
method Introducing low-rank weight matrices to correct quantized activations in LLMs.
result Reduces accuracy gap with original model by more than 50% using low-rank matrices.
Modern LLMs fail at authorship attribution without fine-tuning, but topic embeddings outperform them.
problem Authorship attribution of the Federalist Papers using modern LLMs.
method Examined popular LLMs, compared word/phrase embeddings, and used Bayesian analysis with topic embeddings.
result Topic embeddings trained on 'function words' outperform default LLM embeddings in authorship attribution.
VERA uses variational inference to jailbreak LLMs without manual optimization.
problem Lack of principled objective for gradient-based optimization in jailbreaking LLMs.
method VERA casts black-box jailbreak prompting as a variational inference problem, training a small attacker LLM to approximate the target LLM's posterior over adversarial prompts.
result VERA achieves strong performance across various target LLMs, demonstrating the value of probabilistic inference for adversarial prompt generation.
LLMs outperform strong tabular baselines on industrial car retrofit prediction.
problem Industrial retrofit planning on structured operational data
method Embedding features, direct prompted classification, and ML+LLM stacking
result LLMs outperform strong tabular baselines on industrial car retrofit prediction
Improved financial sentiment analysis using simple instruction tuning of LLMs.
problem Lack of accurate financial sentiment analysis by large language models.
method Instruction tuning of general-purpose LLMs with a small portion of financial sentiment data.
result Significant improvement in financial sentiment analysis, especially in complex scenarios.
Fine-tuned open-source LLMs match or exceed closed-source models in social science research.
problem Limited scalability and high costs of large LLMs in social science research.
method Fine-tuning open-source models for specific tasks, exploring training set size effects, proposing hybrid workflow.
result Small, fine-tuned open-source LLMs achieve equal or superior performance to commercial alternatives.
Efficiently allocate budgets for LLM-assisted virtual screening to reduce costs.
problem Reducing the cost of evaluating alternatives in large-scale screening tasks.
method Propose a top-m greedy evaluation mechanism and the EFG-m algorithm for efficient budget allocation. result Prove that EFG-m is both sample-optimal and consistent in large-scale virtual screening. A simple model explains phase transition in large language models.
problem Understanding the emergence of abilities in large language models.
method Modeling LLM as a sequence-to-sequence random function and using a list decoder.
result A critical threshold exists where the expected number of erroneous sequences grows exponentially.
Proposes a new algorithm for efficient online model selection of LLMs considering the increasing-then-converging trend.
problem Balancing cost and performance in choosing the best LLM among a diverse set of models.
method Introduces a time-increasing bandit algorithm (TI-UCB) that predicts model performance increases and balances exploration and exploitation.
result Achieves a logarithmic regret upper bound, indicating a fast convergence rate in model selection.
New method uses model comparison signals to improve LLM evaluation accuracy.
problem Limited benchmark sizes and model stochasticity in evaluating LLMs' mathematical reasoning.
method Combines standard labeled outcomes with model comparison signals to design a statistically efficient evaluation framework.
result Semiparametric estimator achieves the semiparametric efficiency bound and substantially improves ranking accuracy.
SAGE improves memory efficiency by selectively adding, merging, or ignoring new facts.
problem Efficiently managing new facts in agentic LLMs to avoid costly write-time reasoning.
method SAGE uses a von Mises-Fisher-based density estimator to score and route candidate facts.
result SAGE achieves the best average token-F1 on LoCoMo and reduces add-phase API cost by 3.4x on GPT-4o-mini.
LemonadeBench evaluates LLMs' economic intuition through a simulated lemonade stand.
problem Evaluating LLMs' economic understanding and decision-making in simple markets.
method Simulated lemonade stand business to test LLMs' long-term planning and profit maximization.
result Models achieve profitability but exhibit local rather than global optimization.
SLIP secures LLMs on edge devices by splitting computation and protecting sensitive parts.
problem Protecting LLMs on edge devices from theft and unauthorized use.
method SLIP uses matrix decomposition to split model computation between secure and vulnerable resources, ensuring zero accuracy degradation and minimal latency.
result SLIP is the first practical, secure hybrid protocol for protecting LLMs on edge devices.
AutoScale improves LLM pre-training by adjusting data mixtures at different scales.
problem Data mixtures that work well at small scales may not perform as well at larger scales.
method AutoScale uses a two-stage approach: fitting a model to predict loss under different compositions and extrapolating optimal compositions to larger scales.
result AutoScale accelerates convergence and improves downstream performance.
Kernel smoothing improves LLM reasoning efficiency.
problem Efficiently estimate value functions with limited samples for reinforcement learning.
method Kernelized advantage estimation using classical nonparametric statistics.
result Improved policy optimization with accurate value and gradient estimation.
A new sparse benchmark metabench identifies key abilities from large benchmarks.
problem Redundancy and compression in existing benchmarks.
method Data from 5000+ LLMs to identify most informative items, distilling a sparse benchmark.
result Sparse benchmark metabench captures underlying abilities with high accuracy.
Study shows LLMs can remove half of layers without significant performance drop.
problem Understanding knowledge storage in LLMs' weights.
method Layer pruning and finetuning to identify and remove unnecessary parameters.
result Minimal degradation of performance after removing up to half of layers.
Study identifies key parameters and input dimensions making LLMs and VLMs brittle.
problem Vulnerability of large language and vision-language models to perturbations.
method Proposed FI measure based on information geometry to quantify sensitivity.
result Small subset of high FI parameters significantly contribute to brittleness.
LLMs add value in commodity portfolio construction when information set and implementation rules are held fixed.
problem Commodity portfolio construction
method Multi-Agent LLM Framework
result LLM strategies outperform Rule Agent in Sharpe terms
InvestLM is a financial domain LLM tuned on LLaMA-65B for investment advice.
problem Improving financial text understanding and advice generation for investment.
method Curated financial instruction dataset, LLaMA-65B, less-is-more-for-alignment approach.
result InvestLM provides comparable responses to state-of-the-art commercial models.
A trading system uses LLMs to adapt to volatile crypto markets.
problem Volatility and market sentiment in cryptocurrencies make traditional models ineffective.
method Specialized LLM agents for technical analysis, sentiment evaluation, and decision-making; verbal feedback for continuous improvement.
result Agents outperform buy-and-hold strategy with consistent gains across market phases.
Transformer architecture struggles with complex tasks due to limitations in function composition.
problem Transformer architecture's limitations in handling complex tasks.
method Used Communication Complexity to prove limitations in composing functions.
result Transformer layer is incapable of handling large domain functions, even when domains are small.
Dynamic Vocabulary Pruning stabilizes LLM training by removing low-probability tokens.
problem Training Large Language Models (LLMs) with Reinforcement Learning (RL) causes numerical divergence between inference and training.
method Dynamic Vocabulary Pruning (DVP) constrains the RL objective to a safe vocabulary that excludes low-probability tokens.
result DVP stabilizes training by reducing systematic bias introduced by the extreme tail of the token distribution.
Traditional methods outperform LLMs in forecasting corporate credit ratings.
problem Forecasting corporate credit ratings using LLMs.
method Comparison of traditional methods (XGBoost) and LLMs (LLaMA) on credit rating forecasting.
result XGBoost outperforms LLMs in forecasting corporate credit ratings.
Training-free source selection for LLM families with shared vocabularies
problem Source selection for LLM families with shared vocabularies
method Fisher alignment at vocabulary scale
result Fisher alignment is a cosine between kernel mean embeddings in the joint activation-error space
A new, low-cost method speeds up membership inference attacks on large language models.
problem Membership inference attacks on large language models.
method An ensemble of small quantile regression models to determine model training set membership.
result Comparable or improved accuracy with significantly reduced computational cost.
Unified framework to bridge human and LLM judgments.
problem Systematic discrepancies between human and LLM evaluations.
method Latent human preference score and linear transformations of covariates.
result Higher agreement with human ratings and exposure of systematic gaps.
LLM-Lasso uses LLMs to improve feature selection in Lasso regression.
problem Improving feature selection in Lasso regression with domain-specific knowledge.
method Combines LLMs with Lasso regularization to generate feature weights.
result Outperforms standard Lasso and feature selection baselines in biomedical studies.
EORM boosts LLM accuracy with a lightweight, energy-based verifier.
problem Efficiently verifying mathematical reasoning in large language models.
method Energy-based framework to rank Chain-of-Thought solutions using simple outcome labels.
result EORM boosts LLM accuracy to 90.7% on GSM8k and 63.7% on MATH with only 55M parameters.