The recent trend towards utilisation of reasoning models has improved the performance of Large Language Models (LLMs) across many tasks which involve logical steps. One linguistic task that could benefit from this framing is idiomaticity detection, as a potentially idiomatic expression must first be understood before it can be disambiguated and serves as a basis for reasoning. In this paper, we explore how reasoning capabilities in LLMs affect idiomaticity detection performance and examine the effect of model size. We evaluate, as open source representative models, the suite of DeepSeek-R1 distillation models ranging from 1.5B to 70B parameters across four idiomaticity detection datasets. We find the effect of reasoning to be smaller and more varied than expected. For smaller models, producing chain-of-thought (CoT) reasoning increases performance from Math-tuned intermediate models, but not to the levels of the base models, whereas larger models (14B, 32B, and 70B) show modest improvements. Our in-depth analyses reveal that larger models demonstrate good understanding of idiomaticity, successfully producing accurate definitions of expressions, while smaller models often fail to output the actual meaning. For this reason, we also experiment with providing definitions in the prompts of smaller models, which we show can improve performance in some cases.

We propose STONK (Stock Optimization using News Knowledge), a multimodal framework integrating numerical market indicators with sentiment-enriched news embeddings to improve daily stock-movement prediction. By combining numerical & textual embeddings via feature concatenation and cross-modal attention, our unified pipeline addresses limitations of isolated analyses. Backtesting shows STONK outperforms numeric-only baselines. A comprehensive evaluation of fusion strategies and model configurations offers evidence-based guidance for scalable multimodal financial forecasting. Source code is available on GitHub

Analog IC design is a bottleneck due to its reliance on experience and inefficient simulations, as traditional formulas fail in advanced nodes. Applying Large Language Models (LLMs) directly to this problem risks mere "guessing" without engineering principles. We present a "synergistic reasoning" framework that integrates an LLM's strategic reasoning with the physical precision of the gm/Id methodology. By empowering the LLM with gm/Id lookup tables, it becomes a quantitative, data-driven design partner. We validated this on a two-stage op-amp, where our framework enabled the Gemini model to meet all TT corner specs in 5 iterations and extended optimization to all PVT corners. A crucial ablation study proved gm/Id data is key for this efficiency and precision; without it, the LLM is slower and deviates. Compared to a senior engineer's design, our framework achieves quasi-expert quality with an order-of-magnitude improvement in efficiency. This work validates a path for true analog design automation by combining LLM reasoning with scientific circuit design methodologies.

Large language models (LLMs) exhibit impressive in-context learning (ICL) abilities, enabling them to solve wide range of tasks via textual prompts alone. As these capabilities advance, the range of applicable domains continues to expand significantly. However, identifying the precise mechanisms or internal structures within LLMs that allow successful ICL across diverse, distinct classes of tasks remains elusive. Physics-based tasks offer a promising testbed for probing this challenge. Unlike synthetic sequences such as basic arithmetic or symbolic equations, physical systems provide experimentally controllable, real-world data based on structured dynamics grounded in fundamental principles. This makes them particularly suitable for studying the emergent reasoning behaviors of LLMs in a realistic yet tractable setting. Here, we mechanistically investigate the ICL ability of LLMs, especially focusing on their ability to reason about physics. Using a dynamics forecasting task in physical systems as a proxy, we evaluate whether LLMs can learn physics in context. We first show that the performance of dynamics forecasting in context improves with longer input contexts. To uncover how such capability emerges in LLMs, we analyze the model's residual stream activations using sparse autoencoders (SAEs). Our experiments reveal that the features captured by SAEs correlate with key physical variables, such as energy. These findings demonstrate that meaningful physical concepts are encoded within LLMs during in-context learning. In sum, our work provides a novel case study that broadens our understanding of how LLMs learn in context.

The overall architecture of CFDT is shown in Figure 1. The base backbone is consistent with the network illustrated in the section of 3.1 Local-Global Cross Fusion. As shown by the red dotted lines in Figure 1, we use 100 det tokens as the additional input to perform self attention in the backbone. The det tokens dimension is also set as 256. Neck is a decoder-only modules, and there are 6 decoder layers in this neck.

However, in these domains, the number of available problems typically drops rapidly as a function of problem length (e.g. Figure 2).

Prompted by these insights, we reviewed over 1.5 million scientific papers to understand the origin of this invalid

The large number of ways to phrase a statement in natural language, combined with the core trained imperative to continue text the way a human would, means that nuances of human vulnerability to error and misinterpretation are also reproduced by these models. Hallucination, the tendency of LLMs to generate content that is factually incorrect or nonsensical. For example, a model might recall a fact from its training data or from its prompt with 99% probability (taken over the distribution of the decoding process) but miserably fail to recall it 1% of the time. Or, ignoring for a moment the stochasticity of decoding, it might recall the fact for 99% of the plausible prompts asking to do it but not for the remaining 1%. Indirect prompt injection, the potential for malicious instructions to be embedded within input data not under the user's direct control (such as emails), potentially altering the model's behavior in unexpected ways.

About two weeks ago, OpenAI released GPT-5. The newest AI model in the GPT line, GPT-5 was put forth as the company's "smartest, fastest, most useful model yet" with "built-in thinking" and "expert-level intelligence." But the release backfired for one important reason. Part of the changes in GPT-5 involved addressing the sycophantic positivity found in previous models, where the AI chatbot would incessantly praise the user to an undo degree and emphatically agree to make the user feel better. Lots of users disliked this, so GPT-5 was made to be "less effusively agreeable" and "use fewer unnecessary emojis."