In-context learning (ICL) effectively conditions large language models (LLMs) for molecular tasks, such as property prediction and molecule captioning, by embedding carefully selected demonstration examples into the input prompt. This approach avoids the computational overhead of extensive pertaining and fine-tuning. However, current prompt retrieval methods for molecular tasks have relied on molecule feature similarity, such as Morgan fingerprints, which do not adequately capture the global molecular and atom-binding relationships. As a result, these methods fail to represent the full complexity of molecular structures during inference. Moreover, small-to-medium-sized LLMs, which offer simpler deployment requirements in specialized systems, have remained largely unexplored in the molecular ICL literature. To address these gaps, we propose a self-supervised learning technique, GAMIC (Graph-Aligned Molecular In-Context learning, which aligns global molecular structures, represented by graph neural networks (GNNs), with textual captions (descriptions) while leveraging local feature similarity through Morgan fingerprints. In addition, we introduce a Maximum Marginal Relevance (MMR) based diversity heuristic during retrieval to optimize input prompt demonstration samples. Our experimental findings using diverse benchmark datasets show GAMIC outperforms simple Morgan-based ICL retrieval methods across all tasks by up to 45%.

The challenge of formal proof generation has a rich history, but with modern techniques, we may finally be at the stage of making actual progress in real-life mathematical problems. This paper explores the integration of ChatGPT and basic searching techniques to simplify generating formal proofs, with a particular focus on the miniF2F dataset. We demonstrate how combining a large language model like ChatGPT with a formal language such as Lean, which has the added advantage of being verifiable, enhances the efficiency and accessibility of formal proof generation. Despite its simplicity, our best-performing Lean-based model surpasses all known benchmarks with a 31.15% pass rate. We extend our experiments to include other datasets and employ alternative language models, showcasing our models' comparable performance in diverse settings and allowing for a more nuanced analysis of our results. Our findings offer insights into AI-assisted formal proof generation, suggesting a promising direction for future research in formal mathematical proof.

Moving large objects, such as furniture, is a critical capability for robots operating in human environments. This task presents significant challenges due to two key factors: the need to synchronize whole-body movements to prevent collisions between the robot and the object, and the under-actuated dynamics arising from the substantial size and weight of the objects. These challenges also complicate performing these tasks via teleoperation. In this work, we introduce \method, a generalizable learning framework that leverages human-object interaction demonstrations to enable robots to perform large object manipulation tasks. Central to our approach is the Dynamic Chain, a novel representation that abstracts human-object interactions so that they can be retargeted to robotic morphologies. The Dynamic Chain is a spatial descriptor connecting the human and object root position via a chain of nodes, which encode the position and velocity of different interaction keypoints. We train policies in simulation using Dynamic-Chain-based imitation rewards and domain randomization, enabling zero-shot transfer to real-world settings without fine-tuning. Our approach outperforms both learning-based methods and teleoperation baselines across six evaluation metrics when tested on three distinct object types, both in simulation and on physical hardware. Furthermore, we successfully apply the learned policies to real-world tasks, such as moving a trash cart and rearranging chairs.

With the rapid advancement of large language models (LLMs), concerns about their privacy, safety, and trustworthiness, have become increasingly prominent (Liu et al., 2024d; Barez et al., 2025). However, retraining these models to eliminate the undesired data-model influence is often infeasible due to the significant computational and time costs involved. To address this challenge, LLM unlearning (Yao et al., 2024; Eldan & Russinovich, 2023; Maini et al., 2024; Liu et al., 2024b) has emerged as a post-pretraining strategy, which aims to mitigate the impact of undesirable data (e.g., sensitive, biased, unsafe, or illegal information) and suppress associated model capabilities, thereby preventing LLMs from generating harmful content while simultaneously preserving the model's utility post-unlearning. Despite the increasing importance of LLM unlearning, several recent studies (Łucki et al., 2024; Zhang et al., 2024e; Lynch et al., 2024; Hu et al., 2024; Deeb & Roger, 2024) have identified a critical issue: LLM unlearning often lacks robustness. Specifically, the susceptibility to quickly recovering'already-unlearned' knowledge post-unlearning is evident through so-called relearning attacks (Lynch et al., 2024; Hu et al., 2024). These attacks can effectively reverse the unlearning process by leveraging lightweight fine-tuning on the unlearned model using only a small number of data from the forget dataset.

As AI capabilities increasingly surpass human proficiency in complex tasks, current alignment techniques including SFT and RLHF face fundamental challenges in ensuring reliable oversight. These methods rely on direct human assessment and become untenable when AI outputs exceed human cognitive thresholds. In response to this challenge, we explore two hypotheses: (1) critique of critique can be easier than critique itself, extending the widely-accepted observation that verification is easier than generation to the critique domain, as critique itself is a specialized form of generation; (2) this difficulty relationship is recursively held, suggesting that when direct evaluation is infeasible, performing high-order critiques (e.g., critique of critique of critique) offers a more tractable supervision pathway. To examine these hypotheses, we perform Human-Human, Human-AI, and AI-AI experiments across multiple tasks. Our results demonstrate encouraging evidence supporting these hypotheses and suggest that recursive self-critiquing is a promising direction for scalable oversight.

Telegram is a popular platform for public communication, generating large amounts of messages through its channels. The library offers key features such as easy message retrieval, detailed channel information, engagement metrics, and topic identification using advanced modeling techniques. By simplifying data extraction and analysis, pytopicgram allows users to understand how content spreads and how audiences interact on Telegram. This paper describes the design, main features, and practical uses of pytopicgram, showcasing its effectiveness for studying public conversations on Telegram. Messaging platforms like Telegram have become critical spaces for information exchange, social mobilization, and digital communities. With features such as public channels, unlimited subscribers, and a degree of anonymity, Telegram has emerged as a valuable source of unstructured data reflecting various social, political, and cultural dynamics [1].

In this paper, we propose CodeSCM, a Structural Causal Model (SCM) for analyzing multi-modal code generation using large language models (LLMs). By applying interventions to CodeSCM, we measure the causal effects of different prompt modalities, such as natural language, code, and input-output examples, on the model. CodeSCM introduces latent mediator variables to separate the code and natural language semantics of a multi-modal code generation prompt. Using the principles of Causal Mediation Analysis on these mediators we quantify direct effects representing the model's spurious leanings. We find that, in addition to natural language instructions, input-output examples significantly influence code generation.

Free-text explanations are expressive and easy to understand, but many datasets lack annotated explanation data, making it challenging to train models for explainable predictions. To address this, we investigate how to use existing explanation datasets for self-rationalization and evaluate models' out-of-distribution (OOD) performance. We fine-tune T5-Large and OLMo-7B models and assess the impact of fine-tuning data quality, the number of fine-tuning samples, and few-shot selection methods. The models are evaluated on 19 diverse OOD datasets across three tasks: natural language inference (NLI), fact-checking, and hallucination detection in abstractive summarization. For the generated explanation evaluation, we conduct a human study on 13 selected models and study its correlation with the Acceptability score (T5-11B) and three other LLM-based reference-free metrics. Human evaluation shows that the Acceptability score correlates most strongly with human judgments, demonstrating its effectiveness in evaluating free-text explanations. Our findings reveal: 1) few annotated examples effectively adapt models for OOD explanation generation; 2) compared to sample selection strategies, fine-tuning data source has a larger impact on OOD performance; and 3) models with higher label prediction accuracy tend to produce better explanations, as reflected by higher Acceptability scores.

Large Language Models have proved surprisingly effective at solving a wide range of tasks from just a handful of examples. However, their lack of reliability and modularity limits their capacity to tackle large problems that require many steps of reasoning. In response, researchers have proposed advanced pipelines that leverage domain-specific knowledge to chain smaller prompts, provide intermediate feedback and improve performance through search. However, the current complexity of writing, tuning, maintaining and improving such pipelines has limited their sophistication. We propose oracular programming, a foundational paradigm for building LLM-enabled applications that lets domain experts express high-level problem-solving strategies as programs with unresolved choice points. These choice points are resolved at runtime by LLMs, which generalize from user-provided examples of correct and incorrect decisions. An oracular program is composed of three orthogonal components: a strategy that consists in a nondeterministic program with choice points that can be reified into a search tree, a policy that specifies how to navigate this tree with the help of LLM oracles, and a set of demonstrations that describe successful and unsuccessful search tree navigation scenarios across diverse problem instances. Each component is expressed in a dedicated programming language and can be independently improved or substituted. We address the key programming language design challenges of modularly composing oracular programs and enforcing consistency between their components as they evolve.

Could artificial intelligence ever become truly conscious in a functional sense; this paper explores that open-ended question through the lens of Life, a concept unifying classical biological criteria (Oxford, NASA, Koshland) with empirical hallmarks such as adaptive self maintenance, emergent complexity, and rudimentary self referential modeling. We propose a number of metrics for examining whether an advanced AI system has gained consciousness, while emphasizing that we do not claim all AI stems can become conscious. Rather, we suggest that sufficiently advanced architectures exhibiting immune like sabotage defenses, mirror self-recognition analogs, or meta-cognitive updates may cross key thresholds akin to life-like or consciousness-like traits. To demonstrate these ideas, we start by assessing adaptive self-maintenance capability, and introduce controlled data corruption sabotage into the training process. The result demonstrates AI capability to detect these inconsistencies and revert or self-correct analogous to regenerative biological processes. We also adapt an animal-inspired mirror self recognition test to neural embeddings, finding that partially trained CNNs can distinguish self from foreign features with complete accuracy. We then extend our analysis by performing a question-based mirror test on five state-of-the-art chatbots (ChatGPT4, Gemini, Perplexity, Claude, and Copilot) and demonstrated their ability to recognize their own answers compared to those of the other chatbots.