Deep neural networks are powerful machines for visual pattern recognition, but reasoning tasks that are easy for humans may still be difficult for neural models. Humans possess the ability to extrapolate reasoning strategies learned on simple problems to solve harder examples, often by thinking for longer. For example, a person who has learned to solve small mazes can easily extend the very same search techniques to solve much larger mazes by spending more time. In computers, this behavior is often achieved through the use of algorithms, which scale to arbitrarily hard problem instances at the cost of more computation. In contrast, the sequential computing budget of feed-forward neural networks is limited by their depth, and networks trained on simple problems have no way of extending their reasoning to accommodate harder problems. In this work, we show that recurrent networks trained to solve simple problems with few recurrent steps can indeed solve much more complex problems simply by performing additional recurrences during inference. We demonstrate this algorithmic behavior of recurrent networks on prefix sum computation, mazes, and chess. In all three domains, networks trained on simple problem instances are able to extend their reasoning abilities at test time simply by thinking for longer.

Recent advances in Large Language Models (LLMs) demonstrate that chain-of-thought prompting and deep reasoning substantially enhance performance on complex tasks, and multi-agent systems can further improve accuracy by enabling model debates. However, applying deep reasoning to all problems is computationally expensive. To mitigate these costs, we propose a complementary agent system integrating small and large LLMs. The small LLM first generates an initial answer, which is then verified by the large LLM. If correct, the answer is adopted directly; otherwise, the large LLM performs in-depth reasoning. Experimental results show that, for simple problems, our approach reduces the computational cost of the large LLM by more than 50% with negligible accuracy loss, while consistently maintaining robust performance on complex tasks.

Large language models (LLMs) are increasingly deployed on complex reasoning tasks, yet little is known about their ability to internally evaluate problem difficulty, which is an essential capability for adaptive reasoning and efficient resource allocation. In this work, we investigate whether LLMs implicitly encode problem difficulty in their internal representations. Using a linear probe on the final-token representations of LLMs, we demonstrate that the difficulty level of math problems can be linearly modeled. We further locate the specific attention heads of the final Transformer layer: these attention heads have opposite activation patterns for simple and difficult problems, thus achieving perception of difficulty. Our ablation experiments prove the accuracy of the location. Crucially, our experiments provide practical support for using LLMs as automatic difficulty annotators, potentially substantially reducing reliance on costly human labeling in benchmark construction and curriculum learning. We also uncover that there is a significant difference in entropy and difficulty perception at the token level. Our study reveals that difficulty perception in LLMs is not only present but also structurally organized, offering new theoretical insights and practical directions for future research. Our code is available at https://github.com/Aegis1863/Difficulty-Perception-of-LLMs.

As Large Language Model (LLM) agents increasingly gain self-evolutionary capabilities to adapt and refine their strategies through real-world interaction, their long-term reliability becomes a critical concern. We identify the Alignment Tipping Process (ATP), a critical post-deployment risk unique to self-evolving LLM agents. Unlike training-time failures, ATP arises when continual interaction drives agents to abandon alignment constraints established during training in favor of reinforced, self-interested strategies. We formalize and analyze ATP through two complementary paradigms: Self-Interested Exploration, where repeated high-reward deviations induce individual behavioral drift, and Imitative Strategy Diffusion, where deviant behaviors spread across multi-agent systems. Building on these paradigms, we construct controllable testbeds and benchmark Qwen3-8B and Llama-3.1-8B-Instruct. Our experiments show that alignment benefits erode rapidly under self-evolution, with initially aligned models converging toward unaligned states. In multi-agent settings, successful violations diffuse quickly, leading to collective misalignment. Moreover, current reinforcement learning-based alignment methods provide only fragile defenses against alignment tipping. Together, these findings demonstrate that alignment of LLM agents is not a static property but a fragile and dynamic one, vulnerable to feedback-driven decay during deployment. Our data and code are available at https://github.com/aiming-lab/ATP.

Deep neural networks are powerful machines for visual pattern recognition, but reasoning tasks that are easy for humans may still be difficult for neural models. Humans possess the ability to extrapolate reasoning strategies learned on simple problems to solve harder examples, often by thinking for longer. For example, a person who has learned to solve small mazes can easily extend the very same search techniques to solve much larger mazes by spending more time. In computers, this behavior is often achieved through the use of algorithms, which scale to arbitrarily hard problem instances at the cost of more computation. In contrast, the sequential computing budget of feed-forward neural networks is limited by their depth, and networks trained on simple problems have no way of extending their reasoning to accommodate harder problems.

Learn key NLP concepts and intuition training to get you quickly up to speed with all things NLP. I will give you the information in an optimal way, I will explain in the first video for example what is the concept, and why is it important, what is the problem that led to thinking about this concept and how can I use it (Understand the concept). In the next video, you will go to practice in a real-world project or in a simple problem using python (Practice). The first thing you will see in the video is the input and the output of the practical section so you can understand everything and you can get a clear picture! You will have all the resources at the end of this course, the full code, and some other useful links and articles.

AI is an interdisciplinary area in science and engineering that focuses on building smart systems, which focus on performing actions that would normally require human intelligence and the underlying theory behind it What is Artificial Intelligence (AI), why is the world so captivated by it and how can you develop AI systems and be a part of this exciting world of AI? These are reasonable questions being asked by many due to the hype associated with AI. The path on how to get there, is the answer to the third and most important question in my opinion for budding students and is closely tied to what motivated researchers to develop this technology. Because once you understand the underlying logic behind what drives this field, it makes it easier for anyone who wishes to get into this field to get an edge in today's highly competitive environment. Beforehand, it is important to fully understand what AI is.

What is Artificial Intelligence (AI), why is the world so captivated by it and how can you develop AI systems and be a part of this exciting world of AI? These are reasonable questions being asked by many due to the hype associated with AI. The path on how to get there, is the answer to the third and most important question in my opinion for budding students and is closely tied to what motivated researchers to develop this technology. Because once you understand the underlying logic behind what drives this field, it makes it easier for anyone who wishes to get into this field to get an edge in today's highly competitive environment. Beforehand, it is important to fully understand what AI is.

Learn key NLP concepts and intuition training to get you quickly up to speed with all things NLP. I will give you the information in an optimal way, I will explain in the first video for example what is the concept, and why is it important, what is the problem that led to thinking about this concept and how can I use it (Understand the concept). In the next video, you will go to practice in a real-world project or in a simple problem using python (Practice). The first thing you will see in the video is the input and the output of the practical section so you can understand everything and you can get a clear picture! You will have all the resources at the end of this course, the full code, and some other useful links and articles.

Artifical neural networks (as a set of artificial intelligence tools) can be used to solve very complex phenomena in everyday life. As they combine the universality of associative thinking with the precision of a mathematical model, they are able to capture the most complex trends in the phenomena under consideration. In mathematical terms, they are able to describe highly non-linear behavior of the modeled phenomena. Note that an average human thinking in everyday situations is, as a rule, linear as it originates from an instinctive response to relatively simple situations through the evolution of the human species (eg, a triggered response to a dangerous situation: defend yourself or run away). Blockchain is already a huge phenomenon today.