Abstract: Hallucination is a persistent challenge in large language models (LLMs), where even with rigorous quality control, models often generate distorted facts. This paradox, in which error generation continues despite high-quality training data, calls for a deeper understanding of the underlying LLM mechanisms. To address it, we propose a novel concept: knowledge overshadowing, where model’s dominant knowledge can obscure less prominent knowledge during text generation, causing the model to fabricate inaccurate details. Building on this idea, we introduce a novel framework to quantify factual hallucinations by modeling knowledge overshadowing. Central to our approach is the log-linear law, which predicts that the rate of factual hallucination increases linearly with the logarithmic scale of Knowledge Popularity, Knowledge Length, and Model Size. The law provides a means to preemptively quantify hallucinations, offering foresight into their occurrence even before model training or inference. Built on overshadowing effect, we propose a new decoding strategy CoDa, to mitigate hallucinations, which notably enhance model factuality on Overshadow (27.9%), MemoTrap (13.1%) and NQ-Swap (18.3%). Our findings not only deepen understandings of the underlying mechanisms behind hallucinations but also provide actionable insights for developing more predictable and controllable language models.
From: Yuji Zhang [view email].
Have you ever questioned the deep nature of time? While some physicists argue that time is just an illusion, dismissing it outright contradicts our lived experience. In my latest work, Temporal Mechanics: D-Theory as a Critical Upgrade to Our Understanding of the Nature of Time (2025), I explore how time is deeply rooted in the computational nature of reality and information processing by conscious systems. This paper tackles why the “now” is all we have.
In the absence of observers, the cosmic arrow of time doesn’t exist. This statement is not merely philosophical; it is a profound implication of the problem of time in physics. In standard quantum mechanics, time is an external parameter, a backdrop against which events unfold. However, in quantum gravity and the Wheeler-DeWitt equation, the problem of time emerges because there is no preferred universal time variable—only a timeless wavefunction of the universe. The flow of time, as we experience it, arises not from any fundamental law but from the interaction between observers and the informational structure of reality.
Mushrooms are probably the most miraculous entities because each mushroom can aid in a different way to cure each illness in the human biology. Much like cannabis is actually a cure all for so many ailments in humans so in turn are mushrooms able to do the same.
Alzheimer’s disease (AD) stands as a formidable neurodegenerative ailment and a prominent contributor to dementia. The scarcity of available therapies for AD accentuates the exigency for innovative treatment modalities. Psilocybin, a psychoactive alkaloid intrinsic to hallucinogenic mushrooms, has garnered attention within the neuropsychiatric realm due to its established safety and efficacy in treating depression. Nonetheless, its potential as a therapeutic avenue for AD remains largely uncharted. This comprehensive review endeavors to encapsulate the pharmacological effects of psilocybin while elucidating the existing evidence concerning its potential mechanisms contributing to a positive impact on AD. Specifically, the active metabolite of psilocybin, psilocin, elicits its effects through the modulation of the 5-hydroxytryptamine 2A receptor (5-HT2A receptor). This modulation causes heightened neural plasticity, diminished inflammation, and improvements in cognitive functions such as creativity, cognitive flexibility, and emotional facial recognition. Noteworthy is psilocybin’s promising role in mitigating anxiety and depression symptoms in AD patients. Acknowledging the attendant adverse reactions, we proffer strategies aimed at tempering or mitigating its hallucinogenic effects. Moreover, we broach the ethical and legal dimensions inherent in psilocybin’s exploration for AD treatment. By traversing these avenues, We propose therapeutic potential of psilocybin in the nuanced management of Alzheimer’s disease.
Alzheimer’s disease (AD) is a progressive neurodegenerative disease that is the leading cause of dementia in the elderly population (Anonymous, 2021). It is characterized by the deposition of amyloid-beta (Aβ) plaques, tau neurofibrillary tangles, and neuroinflammation (Scheltens et al., 2021). The prevalence of dementia is expected to rise as the global population grows and ages, with projections estimating a significant increase in the number of cases (Anonymous, 2022b). In 2019, the total cost of healthcare, long-term care, and hospice services for individuals aged 65 years and older with dementia in the United States was estimated at $2.2billion, so AD imposes a substantial burden on individuals, families, society, and the economy (Anonymous, 2022a). The U.S. Food and Drug Administration (FDA) has approved seven drugs for the treatment of AD, including rivastigmine, donepezil, galantamine, memantine, memantine combined with donepezil, aducanumab and lecanemab.
In today’s AI news, Elon Musk’s AI company, xAI, has officially launched its latest flagship AI model, Grok 3. Released late on February 17, 2025, Grok 3 introduces significant advancements over its predecessor, Grok 2, and aims to compete with leading AI models such as OpenAI’s GPT-4o and Google’s Gemini.
In other advancements, Replit has transformed non-technical employees at Zillow into software developers. The real estate giant now routes over 100,000 home shoppers to agents using applications built by team members who had never written code before. This breakthrough stems from Replit’s new partnership with Anthropic and Google Cloud, which has enabled over 100,000 applications on Google Cloud Run.
Then, Wu Yonghui, a prestigious “Google Fellow” who worked at the US tech giant for 17 years, recently joined TikTok owner ByteDance to lead foundational research on artificial intelligence (AI), as the firm seeks to “explore the upper limit of intelligence”. Wu now works at ByteDance’s Seed department, which the Beijing-based company started in early 2023.
Meanwhile, large companies are not adopting AI as quickly as start-ups, AWS managing director Tanuja Randery says. The gap is leading to a “two-tier” AI economy as startups outpace corporations. Citing a new report from AWS, Randery said that European startups had integrated AI at pace over the last year while larger enterprises in the region were falling behind.
In videos, join Sara Bacha from Converge Technology Solutions as she delves into how GraphRAG outperforms traditional RAG by leveraging knowledge graphs and LLM to enhance data relationships and accuracy. Learn the benefits in development, production, and governance, making maintenance easier with better explainability and traceability.
In today’s AI news, Galileo launched an Agent Leaderboard on Hugging Face, an open-source AI platform where users can build, train, access, and deploy AI models. The leaderboard is meant to help people learn how AI agents perform in real-world business applications and help teams determine which agent best fits their needs.
In other advancements, Bloomberg reported Friday that xAI is canvassing existing investors, including Sequoia Capital, Andreessen Horowitz, and Valor Equity Partners for the round, which would bring xAI’s total raised to $22.4 billion, according to Crunchbase. Bloomberg also noted that discussions are ongoing and that the terms of the fundraising round may change.
Ve done mobile app development will know how challenging it can be to deliver the right kind of experience on a smartphone. + And, while speaking with former U.K. Prime Minister Tony Blair at the World Governments Summit in Dubai on Wednesday, Oracle cofounder and executive chairman, Larry Ellison said that while government organizations collect massive amounts of data, it is highly fragmented, making it hard to feed it into an AI model.
In videos, the Imagination in Action video series from Davos 2025 is being uploaded and we’re featuring the sessions in today’s newsletter. First we dive into an in-depth panel discussion featuring AI visionaries Max Tegmark, Demis Hassabis, Yoshua Bengio, Dawn Song, and Ya-Qin Zhang. In this engaging conversation, the experts unpack the distinctions between narrow AI, AGI, and super intelligence …
And, an expert panel explores how regulation can drive innovation in AI, featuring perspectives from panelists: Robert Mahari, JD-PhD at Massachusetts Institute of Technology and Harvard Law School, Pablo Arredondo, Vice President of CoCounsel at Thomson Reuters and Founder of Casetext, Part of Thomson Reuters, Julia Apostle, Partner at Orrick, Herrington & Sutcliffe LLP, Gabriele Mazzini, Fellow at MIT Connection Science and Architect of EU AI Act.
Divorce, the legal dissolution of marriage, can be driven by a variety of factors, ranging from changes in the economic status or health conditions of spouses to contrasting values. The end of a marriage can often be challenging to process. Thus, it can have adverse effects on the well-being and mental health of ex-spouses.
On average, the rates of divorce worldwide have increased over the past century. Improved understanding of the primary factors that prompt people to dissolve a marriage could help to devise more effective couples and marriage counseling strategies, potentially contributing to a reduction in divorce rates.
Sari Mentser and Lilach Sagiv, two researchers at the Hebrew University of Jerusalem, recently carried out a study specifically exploring the relationship between people’s values and divorce. Their findings, published in Communications Psychology, suggest that interaction between spouses’ cultural and personal values can predict divorce.