Large language models (LLMs) have demonstrated significant progress in multilingual language understanding and generation. However, due to the imbalance in training data, their capabilities in non-English languages are limited. Recent studies revealed the English-pivot multilingual mechanism of LLMs, where LLMs implicitly convert non-English queries into English ones at the bottom layers and adopt English for thinking at the middle layers. However, due to the absence of explicit supervision for cross-lingual alignment in the intermediate layers of LLMs, the internal representations during these stages may become inaccurate. In this work, we introduce a deep supervision fine-tuning method (DFT) that incorporates additional supervision in the internal layers of the model to guide its workflow. Specifically, we introduce two training objectives on different layers of LLMs: one at the bottom layers to constrain the conversion of the target language into English, and another at the middle layers to constrain reasoning in English. To effectively achieve the guiding purpose, we designed two types of supervision signals: logits and feature, which represent a stricter constraint and a relatively more relaxed guidance. Our method guides the model to not only consider the final generated result when processing non-English inputs but also ensure the accuracy of internal representations. We conducted extensive experiments on typical English-centric large models, LLaMA-2 and Gemma-2, and the results on multiple multilingual datasets show that our method significantly outperforms traditional fine-tuning methods.

Personalizing large language models (LLMs) is essential for delivering tailored interactions that improve user experience. Many existing personalization methods require fine-tuning LLMs for each user, rendering them prohibitively expensive for widespread adoption. Although retrieval-based approaches offer a more compute-efficient alternative, they still depend on large, high-quality datasets that are not consistently available for all users. To address this challenge, we propose CHAMELEON, a scalable and efficient personalization approach that uses (1) self-generated personal preference data and (2) representation editing to enable quick and cost-effective personalization. Our experiments on various tasks, including those from the LaMP personalization benchmark, show that CHAMELEON efficiently adapts models to personal preferences, improving instruction-tuned models and outperforms two personalization baselines by an average of 40% across two model architectures.

We introduce LADDER (Learning through Autonomous Difficulty-Driven Example Recursion), a framework which enables Large Language Models to autonomously improve their problem-solving capabilities through self-guided learning by recursively generating and solving progressively simpler variants of complex problems. Unlike prior approaches that require curated datasets or human feedback, LADDER leverages a model's own capabilities to generate easier question variants. We demonstrate LADDER's effectiveness in the subject of mathematical integration, improving Llama 3.2 3B's accuracy from 1% to 82% on undergraduate-level problems and enabling Qwen2.5 7B Deepseek-R1 Distilled to achieve 73% on the MIT Integration Bee qualifying examination. We also introduce TTRL (Test-Time Reinforcement Learning), where we perform reinforcement learning on variants of test problems at inference time. TTRL enables Qwen2.5 7B Deepseek-R1 Distilled to achieve a state-of-the-art score of 90% on the MIT Integration Bee qualifying examination, surpassing OpenAI o1's performance. These results show how self-directed strategic learning can achieve significant capability improvements without relying on architectural scaling or human supervision.

Large Language Models (LLMs) can engage in human-looking conversational exchanges. Although conversations can elicit trust between users and LLMs, scarce empirical research has examined trust formation in human-LLM contexts, beyond LLMs' trustworthiness or human trust in AI in general. Here, we introduce the Trust-In-LLMs Index (TILLMI) as a new framework to measure individuals' trust in LLMs, extending McAllister's cognitive and affective trust dimensions to LLM-human interactions. We developed TILLMI as a psychometric scale, prototyped with a novel protocol we called LLM-simulated validity. The LLM-based scale was then validated in a sample of 1,000 US respondents. Exploratory Factor Analysis identified a two-factor structure. Two items were then removed due to redundancy, yielding a final 6-item scale with a 2-factor structure. Confirmatory Factor Analysis on a separate subsample showed strong model fit ($CFI = .995$, $TLI = .991$, $RMSEA = .046$, $p_{X^2} > .05$). Convergent validity analysis revealed that trust in LLMs correlated positively with openness to experience, extraversion, and cognitive flexibility, but negatively with neuroticism. Based on these findings, we interpreted TILLMI's factors as "closeness with LLMs" (affective dimension) and "reliance on LLMs" (cognitive dimension). Younger males exhibited higher closeness with- and reliance on LLMs compared to older women. Individuals with no direct experience with LLMs exhibited lower levels of trust compared to LLMs' users. These findings offer a novel empirical foundation for measuring trust in AI-driven verbal communication, informing responsible design, and fostering balanced human-AI collaboration.

To answer one-to-many factual queries (e.g., listing cities of a country), a language model (LM) must simultaneously recall knowledge and avoid repeating previous answers. How are these two subtasks implemented and integrated internally? Across multiple datasets and models, we identify a promote-then-suppress mechanism: the model first recalls all answers, and then suppresses previously generated ones. Specifically, LMs use both the subject and previous answer tokens to perform knowledge recall, with attention propagating subject information and MLPs promoting the answers. Then, attention attends to and suppresses previous Figure 1: To answer one-to-many factual queries, we answer tokens, while MLPs amplify the found that LMs first use attention to propagate subject suppression signal. Our mechanism is corroborated information to the last token, which is used by MLPs by extensive experimental evidence: in to promote all possible answers. Attention then attends addition to using early decoding and causal to and suppresses the subject and previous answer tokens, tracing, we analyze how components use different while MLPs amplify the suppression and further tokens by introducing both Token Lens, promote new answers.

Recent advances in Large Language Models (LLMs) have raised interest in their formal reasoning capabilities, particularly in mathematics. While closed LLMs like GPT-4 perform well on mathematical benchmarks, e.g., GSM8K, it remains unclear whether small to medium-sized open LLMs can achieve similar performance, questioning their reliability. To close this gap, we propose a post-training approach leveraging a mixture of opinions (MoO) from weaker ancillary LLMs to enhance a (relatively) stronger LLM's reasoning. For that, each post-training sample is augmented with Chain-of-Thought (CoT) reasoning steps and answers from ancillary LLMs, enabling the main LLM to learn from diverse perspectives. We compare MoO with standard supervised fine-tuning (SFT), few-shot prompting, and the Mixture of Agents (MoA) method on mathematical reasoning benchmarks. Our results show that incorporating weaker LLMs' opinions improves mathematical reasoning by an average of 5%, highlighting the value of diverse perspectives in reasoning tasks.

With the rapid advancement of Large Language Models (LLMs), the demand for robust instruction-following capabilities in code generation tasks has grown significantly. Code generation not only facilitates faster prototyping and automated testing, but also augments developer efficiency through improved maintainability and reusability of code. In this paper, we introduce CodeIF, the first benchmark specifically designed to assess the abilities of LLMs to adhere to task-oriented instructions within diverse code generation scenarios. CodeIF encompasses a broad range of tasks, including function synthesis, error debugging, algorithmic refactoring, and code explanation, thereby providing a comprehensive suite to evaluate model performance across varying complexity levels and programming domains. We conduct extensive experiments with LLMs, analyzing their strengths and limitations in meeting the demands of these tasks. The experimental results offer valuable insights into how well current models align with human instructions, as well as the extent to which they can generate consistent, maintainable, and contextually relevant code. Our findings not only underscore the critical role that instruction-following LLMs can play in modern software development, but also illuminate pathways for future research aimed at enhancing their adaptability, reliability, and overall effectiveness in automated code generation.

Private Transformer inference using cryptographic protocols offers promising solutions for privacy-preserving machine learning; however, it still faces significant runtime overhead (efficiency issues) and challenges in handling long-token inputs (scalability issues). We observe that the Transformer's operational complexity scales quadratically with the number of input tokens, making it essential to reduce the input token length. Notably, each token varies in importance, and many inputs contain redundant tokens. Additionally, prior private inference methods that rely on high-degree polynomial approximations for non-linear activations are computationally expensive. Therefore, reducing the polynomial degree for less important tokens can significantly accelerate private inference. Building on these observations, we propose CipherPrune, an efficient and scalable private inference framework that includes a secure encrypted token pruning protocol, a polynomial reduction protocol, and corresponding Transformer network optimizations. At the protocol level, encrypted token pruning adaptively removes unimportant tokens from encrypted inputs in a progressive, layer-wise manner. Additionally, encrypted polynomial reduction assigns lower-degree polynomials to less important tokens after pruning, enhancing efficiency without decryption. At the network level, we introduce protocol-aware network optimization via a gradient-based search to maximize pruning thresholds and polynomial reduction conditions while maintaining the desired accuracy. Our experiments demonstrate that CipherPrune reduces the execution overhead of private Transformer inference by approximately 6.1 for 128-token inputs and 10.6 for 512-token inputs, compared to previous methods, with only a marginal drop in accuracy. Transformers (Vaswani et al., 2017) have become the predominant approach for tackling a wide range of machine learning tasks, spanning Natural Language Processing (NLP) (Xue et al., 2024) and Computer Vision (CV) (Zheng et al., 2022) domains.

Most safety training methods for large language models (LLMs) based on fine-tuning rely on dramatically changing the output distribution of the model when faced with a harmful request, shifting it from an unsafe answer to a refusal to respond. These methods inherently compromise model capabilities and might make auto-regressive models vulnerable to attacks that make likely an initial token of affirmative response. To avoid that, we propose to expand the model's vocabulary with a special token we call red flag token () and propose to fine-tune the model to generate this token at any time harmful content is generated or about to be generated. This novel safety training method effectively augments LLMs into generative classifiers of harmfulness at all times during the conversation. This method offers several advantages: it enables the model to explicitly learn the concept of harmfulness while marginally affecting the generated distribution, thus maintaining the model's utility. It also evaluates each generated answer rather than just the input prompt and provides a stronger defence against sampling-based attacks. In addition, it simplifies the evaluation of the model's robustness and reduces correlated failures when combined with a classifier. We further show an increased robustness to long contexts, and supervised fine-tuning attacks.

Recent advances in code generation have illuminated the potential of employing large language models (LLMs) for general-purpose programming languages such as Python and C++, opening new opportunities for automating software development and enhancing programmer productivity. The potential of LLMs in software programming has sparked significant interest in exploring automated hardware generation and automation. Although preliminary endeavors have been made to adopt LLMs in generating hardware description languages (HDLs), several challenges persist in this direction. First, the volume of available HDL training data is substantially smaller compared to that for software programming languages. Second, the pre-trained LLMs, mainly tailored for software code, tend to produce HDL designs that are more error-prone. Third, the generation of HDL requires a significantly higher number of tokens compared to software programming, leading to inefficiencies in cost and energy consumption. To tackle these challenges, this paper explores leveraging LLMs to generate High-Level Synthesis (HLS)-based hardware design. Although code generation for domain-specific programming languages is not new in the literature, we aim to provide experimental results, insights, benchmarks, and evaluation infrastructure to investigate the suitability of HLS over low-level HDLs for LLM-assisted hardware design generation. To achieve this, we first finetune pre-trained models for HLS-based hardware generation, using a collected dataset with text prompts and corresponding reference HLS designs. An LLM-assisted framework is then proposed to automate end-to-end hardware code generation, which also investigates the impact of chain-of-thought and feedback loops promoting techniques on HLS-design generation. Limited by the timeframe of this research, we plan to evaluate more advanced reasoning models in the future.