Identifying bias in LLMs is ongoing. Because they are still in development, what is true today may be false tomorrow. We therefore need general strategies for debiasing that will outlive current models. Strategies developed for debiasing human decision making offer one promising approach as they incorporate an LLM-style prompt intervention designed to bring latent knowledge into awareness during decision making. LLMs trained on vast amounts of information contain information about potential biases, counter-arguments, and contradictory evidence, but that information may only be brought to bear if prompted. Metacognitive prompts developed in the human decision making literature are designed to achieve this, and as I demonstrate here, they show promise with LLMs. The prompt I focus on here is "could you be wrong?" Following an LLM response, this prompt leads LLMs to produce additional information, including why they answered as they did, errors, biases, contradictory evidence, and alternatives, none of which were apparent in their initial response. Indeed, this metaknowledge often reveals that how LLMs and users interpret prompts are not aligned. Here I demonstrate this prompt using a set of questions taken from recent articles about LLM biases, including implicit discriminatory biases and failures of metacognition. "Could you be wrong" prompts the LLM to identify its own biases and produce cogent metacognitive reflection. I also present another example involving convincing but incomplete information, which is readily corrected by the metacognitive prompt. In sum, this work argues that human psychology offers a new avenue for prompt engineering, leveraging a long history of effective prompt-based improvements to human decision making.

As AI systems become increasingly capable and ubiquitous, ensuring the safety of these systems is critical. However, existing safety tools often target different aspects of model safety and cannot provide full assurance in isolation, highlighting a need for integrated and composite methodologies. This paper introduces BlueGlass, a framework designed to facilitate composite AI safety workflows by providing a unified infrastructure enabling the integration and composition of diverse safety tools that operate across model internals and outputs. Furthermore, to demonstrate the utility of this framework, we present three safety-oriented analyses on vision-language models for the task of object detection: (1) distributional evaluation, revealing performance trade-offs and potential failure modes across distributions; (2) probe-based analysis of layer dynamics highlighting shared hierarchical learning via phase transition; and (3) sparse autoencoders identifying interpretable concepts. More broadly, this work contributes foundational infrastructure and findings for building more robust and reliable AI systems.

Recently, large language models (LLMs) have demonstrated powerful capabilities in performing various tasks and thus are applied by recent studies to time series forecasting (TSF) tasks, which predict future values with the given historical time series. Existing LLM-based approaches transfer knowledge learned from text data to time series prediction using prompting or fine-tuning strategies. However, LLMs are proficient at reasoning over discrete tokens and semantic patterns but are not initially designed to model continuous numerical time series data. The gaps between text and time series data lead LLMs to achieve inferior performance to a vanilla Transformer model that is directly trained on TSF data. However, the vanilla Transformers often struggle to learn high-level semantic patterns. In this paper, we design a novel Transformer-based architecture that complementarily leverages LLMs and vanilla Transformers, so as to integrate the high-level semantic representations learned by LLMs into the temporal information encoded by time series Transformers, where a hybrid representation is obtained by fusing the representations from the LLM and the Transformer. The resulting fused representation contains both historical temporal dynamics and semantic variation patterns, allowing our model to predict more accurate future values. Experiments on benchmark datasets demonstrate the effectiveness of the proposed approach.

Representation Fine-tuning (ReFT), a recently proposed Parameter-Efficient Fine-Tuning (PEFT) method, has attracted widespread attention for significantly improving parameter efficiency by editing representation space alone. In this work, we investigate applying ReFT to complex reasoning tasks. However, directly using the native ReFT method, which modifies fixed representations at the beginning and end of each layer, yields suboptimal performance, as these fixed-position representations have uncertain impact on the outputs. We observe that, in complex reasoning tasks, there often exist certain critical representations. These representations either integrate significant information from preceding layers or regulate subsequent layer representations. Through layer-by-layer propagation, they exert a substantial influence on the final output. Naturally, fine-tuning these critical representations has the potential to greatly enhance reasoning performance. Building upon these insights, we propose Critical Representation Fine-Tuning (CRFT), a novel method that identifies and optimizes these critical representations through information flow analysis. CRFT operates within a supervised learning framework, dynamically optimizing critical representations in a low-rank linear subspace while freezing the base model. The effectiveness and efficiency of our method are validated across eight benchmarks for arithmetic and commonsense reasoning, using LLaMA and Mistral model families. Furthermore, our method also adapts effectively to few-shot settings, boosting one-shot accuracy by 16.4%. Our work highlights the untapped potential of representation-level optimization for CoT reasoning, offering a lightweight yet powerful alternative to traditional PEFT methods.

Large Language Models (LLM) have achieved remarkable performance across a large number of tasks, but face critical deployment and usage barriers due to substantial computational requirements. Model compression methods, which aim to reduce model size while preserving its capacity, are an important means to mitigate these issues. Promising approaches along these lines, such as structured pruning, typically require costly empirical search for optimal variants and may run the risk of ignoring better solutions. In this work we introduce GeLaCo, an evolutionary approach to LLM compression via layer collapse. Our approach supports an efficient exploration of the compression solution space via population-based search and a module-wise similarity fitness function capturing attention, feed-forward, and hidden state representations. GeLaCo also supports both single and multi-objective evolutionary compression search, establishing the first Pareto frontier along compression and quality axes. We evaluate GeLaCo solutions via both perplexity-based and generative evaluations over foundational and instruction-tuned models, outperforming state-of-the-art alternatives.

This paper investigates whether state-of-the-art Large Language Models (LLMs) can automatically translate SPARQL between popular Knowledge Graph (KG) schemas. We focus on translations between the DBpedia and Wikidata KG, and later on DBLP and OpenAlex KG. This study addresses a notable gap in KG interoperability research by rigorously evaluating LLM performance on SPARQL-to-SPARQL translation. Two benchmarks are assembled, where the first align 100 DBpedia-Wikidata queries from QALD-9-Plus; the second contains 100 DBLP queries aligned to OpenAlex, testing generalizability beyond encyclopaedic KGs. Three open LLMs: Llama-3-8B, DeepSeek-R1-Distill-Llama-70B, and Mistral-Large-Instruct-2407 are selected based on their sizes and architectures and tested with zero-shot, few-shot, and two chain-of-thought variants. Outputs were compared with gold answers, and resulting errors were categorized. We find that the performance varies markedly across models and prompting strategies, and that translations for Wikidata to DBpedia work far better than translations for DBpedia to Wikidata.

Single-cell foundation models such as scGPT represent a significant advancement in single-cell omics, with an ability to achieve state-of-the-art performance on various downstream biological tasks. However, these models are inherently limited in that a vast amount of information in biology exists as text, which they are unable to leverage. There have therefore been several recent works that propose the use of LLMs as an alternative to single-cell foundation models, achieving competitive results. However, there is little understanding of what factors drive this performance, along with a strong focus on using LLMs as an alternative, rather than complementary approach to single-cell foundation models. In this study, we therefore investigate what biological insights contribute toward the performance of LLMs when applied to single-cell data, and introduce scMPT; a model which leverages synergies between scGPT, and single-cell representations from LLMs that capture these insights. scMPT demonstrates stronger, more consistent performance than either of its component models, which frequently have large performance gaps between each other across datasets. We also experiment with alternate fusion methods, demonstrating the potential of combining specialized reasoning models with scGPT to improve performance. This study ultimately showcases the potential for LLMs to complement single-cell foundation models and drive improvements in single-cell analysis.

--Suicide is a critical global health issue that requires urgent attention. Even though prior work has revealed valuable insights into detecting current suicide risk on social media, little attention has been paid to developing models that can predict subsequent suicide risk over time, limiting their ability to capture rapid fluctuations in individuals' mental state transitions. In addition, existing work ignores protective factors that play a crucial role in suicide risk prediction, focusing predominantly on risk factors alone. Protective factors such as social support and coping strategies can mitigate suicide risk by moderating the impact of risk factors. Therefore, this study proposes a novel framework for predicting subsequent suicide risk by jointly learning the dynamic influence of both risk factors and protective factors on users' suicide risk transitions. We propose a novel Protective Factor-A ware Dataset, which is built from 12 years of Reddit posts along with comprehensive annotations of suicide risk and both risk and protective factors. We also introduce a Dynamic Factors Influence Learning approach that captures the varying impact of risk and protective factors on suicide risk transitions, recognizing that suicide risk fluctuates over time according to established psychological theories. Our thorough experiments demonstrate that the proposed model significantly outperforms state-of-the-art models and large language models across three datasets. In addition, the proposed Dynamic Factors Influence Learning provides interpretable weights, helping clinicians better understand suicidal patterns and enabling more targeted intervention strategies. Suicide issue remains one of society's most urgent challenges, with each case representing not only the loss of an individual life but also creating profound ripple effects across families, healthcare systems, and communities that persist for generations. Jun Li, X. Wang, H. Li, H. Leong, and Qing Li are with Department of Computing, The Hong Kong Polytechnic University, Hong Kong SAR, China (e-mail: hialex.li@connect.polyu.hk;

Touch is recognized as a vital sense for humans and an equally important modality for robots, especially for dexterous manipulation, material identification, and scenarios involving visual occlusion. Building upon very recent work in touch foundation models, this demonstration will feature Octopi-1.5, our latest visual-tactile-language model. Compared to its predecessor, Octopi-1.5 introduces the ability to process tactile signals from multiple object parts and employs a simple retrieval-augmented generation (RAG) module to improve performance on tasks and potentially learn new objects on-the-fly. The system can be experienced live through a new handheld tactile-enabled interface, the TMI, equipped with GelSight and TAC-02 tactile sensors. This convenient and accessible setup allows users to interact with Octopi-1.5 without requiring a robot. During the demonstration, we will showcase Octopi-1.5 solving tactile inference tasks by leveraging tactile inputs and commonsense knowledge. For example, in a Guessing Game, Octopi-1.5 will identify objects being grasped and respond to follow-up queries about how to handle it (e.g., recommending careful handling for soft fruits). We also plan to demonstrate Octopi-1.5's RAG capabilities by teaching it new items. With live interactions, this demonstration aims to highlight both the progress and limitations of VTLMs such as Octopi-1.5 and to foster further interest in this exciting field. Code for Octopi-1.5 and design files for the TMI gripper are available at https://github.com/clear-nus/octopi-1.5.

Large decoder-based language models have become the dominant architecture for reward modeling in reinforcement learning from human feedback (RLHF). However, as reward models are increasingly deployed in test-time strategies, their inference costs become a growing concern. We present TinyRM, a family of small, bidirectional masked language models (MLMs) with as few as 400 million parameters, that rival the capabilities of models over 175 times larger on reasoning and safety preference modeling tasks. TinyRM combines FLAN-style prompting, Directional Low-Rank Adaptation (DoRA), and layer freezing to achieve strong performance on RewardBench, despite using significantly fewer resources. Our experiments suggest that small models benefit from domain-specific tuning strategies, particularly in reasoning, where lightweight finetuning methods are especially effective. While challenges remain in building generalist models and conversational preference modeling, our preliminary results highlight the promise of lightweight bidirectional architectures as efficient, scalable alternatives for preference modeling.