Recent advancements in large language models (LLMs) have revolutionized their ability to handle single-turn tasks, yet real-world applications demand sophisticated multi-turn interactions. This survey provides a comprehensive review of recent advancements in evaluating and enhancing multi-turn interactions in LLMs. Focusing on task-specific scenarios, from instruction following in diverse domains such as math and coding to complex conversational engagements in roleplay, healthcare, education, and even adversarial jailbreak settings, we systematically examine the challenges of maintaining context, coherence, fairness, and responsiveness over prolonged dialogues. The paper organizes current benchmarks and datasets into coherent categories that reflect the evolving landscape of multi-turn dialogue evaluation. In addition, we review a range of enhancement methodologies under multi-turn settings, including model-centric strategies (contextual learning, supervised fine-tuning, reinforcement learning, and new architectures), external integration approaches (memory-augmented, retrieval-based methods, and knowledge graph), and agent-based techniques for collaborative interactions. Finally, we discuss open challenges and propose future directions for research to further advance the robustness and effectiveness of multi-turn interactions in LLMs. Related resources and papers are available at https://github.com/yubol-cmu/Awesome-Multi-Turn-LLMs.

Mamba extends earlier state space models (SSMs) by introducing input-dependent dynamics, and has demonstrated strong empirical performance across a range of domains, including language modeling, computer vision, and foundation models. However, a surprising weakness remains: despite being built on architectures designed for long-range dependencies, Mamba performs poorly on long-range sequential tasks. Understanding and addressing this gap is important for improving Mamba's universality and versatility. In this work, we analyze Mamba's limitations through three perspectives: expressiveness, inductive bias, and training stability. Our theoretical results show how Mamba falls short in each of these aspects compared to earlier SSMs such as S4D. To address these issues, we propose $\text{B}_2\text{S}_6$, a simple extension of Mamba's S6 unit that combines block-wise selective dynamics with a channel-specific bias. We prove that these changes equip the model with a better-suited inductive bias and improve its expressiveness and stability. Empirically, $\text{B}_2\text{S}_6$ outperforms S4 and S4D on Long-Range Arena (LRA) tasks while maintaining Mamba's performance on language modeling benchmarks.

We study online learning of feedforward neural networks with the sign activation function that implement functions from the unit ball in $\mathbb{R}^d$ to a finite label set $\{1, \ldots, Y\}$. First, we characterize a margin condition that is sufficient and in some cases necessary for online learnability of a neural network: Every neuron in the first hidden layer classifies all instances with some margin $γ$ bounded away from zero. Quantitatively, we prove that for any net, the optimal mistake bound is at most approximately $\mathtt{TS}(d,γ)$, which is the $(d,γ)$-totally-separable-packing number, a more restricted variation of the standard $(d,γ)$-packing number. We complement this result by constructing a net on which any learner makes $\mathtt{TS}(d,γ)$ many mistakes. We also give a quantitative lower bound of approximately $\mathtt{TS}(d,γ) \geq \max\{1/(γ\sqrt{d})^d, d\}$ when $γ\geq 1/2$, implying that for some nets and input sequences every learner will err for $\exp(d)$ many times, and that a dimension-free mistake bound is almost always impossible. To remedy this inevitable dependence on $d$, it is natural to seek additional natural restrictions to be placed on the network, so that the dependence on $d$ is removed. We study two such restrictions. The first is the multi-index model, in which the function computed by the net depends only on $k \ll d$ orthonormal directions. We prove a mistake bound of approximately $(1.5/γ)^{k + 2}$ in this model. The second is the extended margin assumption. In this setting, we assume that all neurons (in all layers) in the network classify every ingoing input from previous layer with margin $γ$ bounded away from zero. In this model, we prove a mistake bound of approximately $(\log Y)/ γ^{O(L)}$, where L is the depth of the network.

The RSF is trying to show that they don't need to reach Port Sudan by land in order to be able to have an impact there,

A chatbot developed by Elon Musk's multibillion-dollar artificial intelligence startup xAI appeared to be suffering from a glitch Wednesday when it repeatedly brought up white genocide in South Africa in response to user queries about unrelated topics on X. Grok, which competes with other chatbots like OpenAI's ChatGPT, is directly integrated into the social media platform that Musk also owns. Numerous examples of the phenomenon could be found by searching the official Grok profile for posts containing the term "boer," a word used to refer to people from South Africa of "Dutch, German, or Huguenot descent." It is sometimes used by Black South Africans as a pejorative against white Afrikaners, or people associated with the apartheid regime. In response to topics ranging from streaming platform HBO Max's name change to Medicaid cuts proposed by US lawmakers, the chatbot often seemed to initially stay on topic before veering back to white genocide in South Africa, completely unprompted. When asked to confirm the salary of Toronto Blue Jays player Max Scherzer, for example, the generative artificial intelligence chatbot launched into an explanation of white genocide and a controversial South African anti-apartheid song.

Three far-right extremists who amassed hundreds of weapons and planned to carry out attacks on targets including a mosque have been convicted of terrorism offences. Brogan Stewart, 25, from West Yorkshire, Christopher Ringrose, 34, from Staffordshire, and Marco Pitzettu, 25, from Derbyshire, were part of an online group who "idolised the Nazi regime". Sheffield Crown Court was told how Stewart had detailed torturing a Muslim leader using an "information extraction kit". All three were found guilty of terrorism offences at the same court on Wednesday and are due to be sentenced on 17 July.Counter Terrorism Policing North EastThe trio had amassed a cache of weapons as part of their planning During the nine-week trial, the court heard more than 200 weapons including machetes, hunting knives, swords and crossbows were found at their homes. Ringrose had also begun to build a 3D-printed semi-automatic firearm, which counter-terror police said would have been a "lethal weapon".

Early this year the Chinese company Deepseek revealed that it had developed a very powerful model mostly using Nvidia chips obtained before the Biden administration closed an export loophole in 2023, heightening the intensity of the race. Last week, the Trump administration ripped up those rules, with a spokesperson calling them "overly complex, bureaucratic" and saying they "would stymie American innovation." They then switched to a new tack: linking countries' access to AI chips with larger trade negotiations. Transitioning to a negotiation-based approach, the administration argued, could allow for more flexibility from country-to-country and allow Trump to secure key business concessions from Middle Eastern partners. Business and governments in the Middle East have massive ambitions for AI, aiming to position themselves at the forefront of this emerging technology.

The administration of U.S. President Donald Trump is clearing a path for two key Persian Gulf allies to pursue their artificial intelligence ambitions -- and some of the biggest U.S. tech companies are seizing on that opening with plans to spend billions of dollars in the region. Under agreements with the U.S. expected to be unveiled in coming days, Saudi Arabia and the United Arab Emirates are poised to win wider access to advanced AI chips from Nvidia and Advanced Micro Devices that are considered the gold standard for running AI models. The deals are taking shape while President Donald Trump visits the Middle East seeking to forge deeper business ties that put U.S. technology initiatives at center stage. Even before any formal announcement of accords between the U.S. and its partners, news began to emerge of American companies readying expanded projects in the region.

This paper introduces Codified Profiles for role-playing, a novel approach that represents character logic as structured, executable functions for behavioral decision-making. Each profile defines a set of functions parse_by_scene(scene) that outputs a list of logic-grounded assertions triggered_statements, using both explicit control structures (e.g., if-then-else) and condition checks like check_condition(scene, question), where each question is a semantically meaningful prompt about the scene (e.g., "Is the character in danger?") discriminated by the role-playing LLM as true, false, or unknown. This explicit representation offers three key advantages over traditional prompt-based profiles, which append character descriptions directly into text prompts: (1) Persistence, by enforcing complete and consistent execution of character logic, rather than relying on the model's implicit reasoning; (2) Updatability, through systematic inspection and revision of behavioral logic, which is difficult to track or debug in prompt-only approaches; (3) Controllable Randomness, by supporting stochastic behavior directly within the logic, enabling fine-grained variability that prompting alone struggles to achieve. To validate these advantages, we introduce a new benchmark constructed from 83 characters and 5,141 scenes curated from Fandom, using NLI-based scoring to compare character responses against ground-truth actions. Our experiments demonstrate the significant benefits of codified profiles in improving persistence, updatability, and behavioral diversity. Notably, by offloading a significant portion of reasoning to preprocessing, codified profiles enable even 1B-parameter models to perform high-quality role-playing, providing a scalable and efficient foundation for local deployment of role-play agents.

Large Language Models (LLMs) have the tendency to hallucinate, i.e., to sporadically generate false or fabricated information. This presents a major challenge, as hallucinations often appear highly convincing and users generally lack the tools to detect them. Uncertainty quantification (UQ) provides a framework for assessing the reliability of model outputs, aiding in the identification of potential hallucinations. In this work, we introduce pre-trained UQ heads: supervised auxiliary modules for LLMs that substantially enhance their ability to capture uncertainty compared to unsupervised UQ methods. Their strong performance stems from the powerful Transformer architecture in their design and informative features derived from LLM attention maps. Experimental evaluation shows that these heads are highly robust and achieve state-of-the-art performance in claim-level hallucination detection across both in-domain and out-of-domain prompts. Moreover, these modules demonstrate strong generalization to languages they were not explicitly trained on. We pre-train a collection of UQ heads for popular LLM series, including Mistral, Llama, and Gemma 2. We publicly release both the code and the pre-trained heads.