Goto

Collaborating Authors

 James Bay


ICR Probe: Tracking Hidden State Dynamics for Reliable Hallucination Detection in LLMs

Zhang, Zhenliang, Hu, Xinyu, Zhang, Huixuan, Zhang, Junzhe, Wan, Xiaojun

arXiv.org Artificial Intelligence

Large language models (LLMs) excel at various natural language processing tasks, but their tendency to generate hallucinations undermines their reliability. Existing hallucination detection methods leveraging hidden states predominantly focus on static and isolated representations, overlooking their dynamic evolution across layers, which limits efficacy. To address this limitation, we shift the focus to the hidden state update process and introduce a novel metric, the ICR Score (Information Contribution to Residual Stream), which quantifies the contribution of modules to the hidden states' update. We empirically validate that the ICR Score is effective and reliable in distinguishing hallucinations. Building on these insights, we propose a hallucination detection method, the ICR Probe, which captures the cross-layer evolution of hidden states. Experimental results show that the ICR Probe achieves superior performance with significantly fewer parameters. Furthermore, ablation studies and case analyses offer deeper insights into the underlying mechanism of this method, improving its interpretability.


Russian drones hit Ukraine power plant, leaving residents in the cold

Al Jazeera

Russian drone strikes have damaged a thermal power plant in Mykolaiv in southern Ukraine overnight, leaving 46,000 consumers without heating as temperatures plunge below freezing, Ukrainian Prime Minister Denys Shmyhal said. "This was done deliberately to leave people without heat in sub-zero temperatures and create a humanitarian catastrophe," Shmyhal said on the Telegram messenger app. Russia attacked Ukraine with 143 drones overnight, but the Ukrainian military said it shot down 95 of them, while 46 did not reach their targets, likely thanks to the use of electromagnetic countermeasures that disrupt drone attacks. At least one person was injured in the overnight attacks which also damaged houses in the Kyiv region, Ukrainian officials said. The temperature in Mykolaiv is expected to fall to minus 7 degrees Celsius (19.4 Fahrenheit) on Sunday night.


Can simplifying AI rules in Europe create competition for US and China?

Al Jazeera

Can simplifying AI rules in Europe create competition for US and China? Can simplifying AI rules in Europe create competition for US and China? Europe to cut red tape to make artificial intelligence advancements easier.Read more The Artificial Intelligence Action Summit in Paris has drawn nearly 100 world leaders and tech firms, and the consensus is that 2025 is not the year for new AI regulations. France says it is time to simplify the rules in Europe to allow AI advances – or risk being left behind. Which countries have banned DeepSeek and why? list 2 of 3 Elon Musk-led group makes 97.4bn bid for OpenAI list 3 of 3 In January, Chinese start-up DeepSeek disrupted Wall Street and Silicon Valley.


Ukraine gets green light to use US long-range missiles: What's next?

Al Jazeera

United States President Joe Biden has reportedly lifted restrictions on Kyiv on the use of long-range missiles, which means Ukrainian forces may fire American-made missiles inside Russian territory for the first time. The move, which comes weeks before Biden leaves office and hours after massive Russian missile and drone attacks, has angered the Kremlin, which accused Washington of "throwing oil on the fire". Kremlin spokesman Dmitry Peskov said the decision would mean Washington's direct involvement in the conflict, echoing a similar sentiment expressed by President Vladimir Putin in September. The White House and President-elect Donald Trump have not commented yet, but Trump's eldest son, Donald Trump Jr, said: "The military industrial complex seems to want to make sure they get World War III going before my father has a chance to create peace and save lives." The elder Trump, who takes office on January 20, repeatedly pledged during his campaign to negotiate an end to the Ukraine war.


How far can Ukraine's military go inside Russia?

Al Jazeera

Moscow has come under one of the largest drone attacks of the war.Read more When President Vladimir Putin launched Russia's so-called "special military operation" in Ukraine two-and-a-half years ago, he expected a speedy victory. Not only did that not happen, but Ukraine has now brought the war home to Russia. Russia faces manpower woes after failing to stop Ukraine's Kursk incursion list 2 of 4 Russians flock to evacuation centre to flee Ukraine's incursion in Kursk list 4 of 4 The capital has faced one of its biggest drone attacks of the war – according to the mayor of Moscow. Meanwhile, Ukraine's incursion into the Kursk region caught Russia by surprise. Has Ukraine's bold move put on hold discussions about a stalemate and possible negotiations involving concessions to Russia? What are the prospects for a Gaza ceasefire deal?


Biden vows G7 response, 'ironclad' US support for Israel after Iran attacks

Al Jazeera

US President Joe Biden has condemned the Iranian drone attacks on military facilities in Israel, reiterating Washington DC's "ironclad" support and a coordinated Group of Seven (G7) diplomatic response, even as reports started to emerge that he is also seeking to de-escalate the situation. Biden cut short a trip to Delaware and returned to the US capital to meet advisers following the late Saturday night attack, the White House said in a statement. The statement said that US forces and facilities had not been hit, adding that the US helped Israel in taking down "nearly all" of the attacking drones and missiles. The US president also he reiterated the "ironclad" support for Israel's security in a call with Prime Minister Benjamin Netanyahu, with whom he has had strained relations over Israel's handling of the war in Gaza. "I told him that Israel demonstrated a remarkable capacity to defend against and defeat even unprecedented attacks – sending a clear message to its foes that they cannot effectively threaten the security of Israel," the White House quoted Biden as saying.


Interactions in information spread: quantification and interpretation using stochastic block models

Poux-Médard, Gaël, Velcin, Julien, Loudcher, Sabine

arXiv.org Machine Learning

In most real-world applications, it is seldom the case that a given observable evolves independently of its environment. In social networks, users' behavior results from the people they interact with, news in their feed, or trending topics. In natural language, the meaning of phrases emerges from the combination of words. In general medicine, a diagnosis is established on the basis of the interaction of symptoms. Here, we propose a new model, the Interactive Mixed Membership Stochastic Block Model (IMMSBM), which investigates the role of interactions between entities (hashtags, words, memes, etc.) and quantifies their importance within the aforementioned corpora. We find that interactions play an important role in those corpora. In inference tasks, taking them into account leads to average relative changes with respect to non-interactive models of up to 150\% in the probability of an outcome. Furthermore, their role greatly improves the predictive power of the model. Our findings suggest that neglecting interactions when modeling real-world phenomena might lead to incorrect conclusions being drawn.