Jack Clark, co-founder and head of Anthropic's Public Benefit Fund, speaks at meetings of the International Monetary Fund and the World Bank Group in Bangkok on Oct. 16. Washington - A wave of American wealth created by artificial intelligence companies could produce some of the largest charitable sums since the era of Andrew Carnegie and John D. Rockefeller, raising questions about where the money goes. Anthropic, the maker of the Claude chatbot, has said it would commit 5% of its shares to a new Anthropic Public Benefit Fund, a stake that could be worth around $100 billion if the company reaches the $2 trillion valuation investors are eyeing for a stock market debut expected as early as November. Combined with pledges by its seven co-founders to give away 80% of their equity and a program matching employees' donations, Anthropic says roughly a fifth of its shares will be dedicated to public-benefit purposes. In a time of both misinformation and too much information, quality journalism is more crucial than ever.

The International Monetary Fund (IMF) has a responsibility to help poorer nations struggling with rising debt, high inflation and price shocks caused by war spillovers, the head of the global lender says. Speaking as the IMF's annual meetings in Bangkok wrapped up on Saturday, Managing Director Kristalina Georgieva said discussions with global policymakers showed both "optimism and worry". "But AI is also recognised to be a source of worry and so are inflation, high levels of debt, wars, climate risks." Georgieva said vulnerable nations are missing out on the AI investment boom that is driving growth in richer countries while at the same time dealing with higher borrowing costs. Higher global interest rates mean low-income countries now spend 8 to 14 percent of their revenue on debt repayments, up from previous years, she said.

Kristalina Georgieva says online fraud could lead people to opt out of system and turn to'shoeboxes under their bed' The heads of the Bank of England and the International Monetary Fund have issued a stark warning that rising cyber-attacks risk undermining trust in the financial system. Speaking on a panel about online fraud at the IMF and World Bank meetings in Bangkok, the governor of the Bank of England, Andrew Bailey, pointed to alarming headlines about self-replicating AI agents hijacking personal devices, and the French supermarket hack that heated fridges up to 50C . "If you try to say what's at the heart of this debate about AI - for me, it's about trust in technology and it's about human control to achieve that trust in technology," he said. Pointing to the key importance of technology in underpinning today's financial system, he added: "If we don't have trust in money, we don't operate as a society." Echoing his concerns, the IMF's managing director, Kristalina Georgieva, said the toll of cyber-fraud was now so great that people could decide to opt out of the formal financial system altogether.

International Monetary Fund Managing Director Kristalina Georgieva speaks during the IMF/World Bank 2026 Annual Meetings in Bangkok on Monday. The head of the International Monetary Fund has warned of the disruptive threat from artificial intelligence to the labor market and taxation, urging policymakers to rethink education and act now to help society adapt. "It is a slow-moving tsunami, but tsunami none the less," Managing Director Kristalina Georgieva said Monday during her first appearance at the annual meetings of the IMF and World Bank in Bangkok. Speaking at a session on AI's impact on employment, she said the technology is already hollowing out demand for many human workers by taking over routine, repetitive tasks. At the same time, it is increasing low-paying jobs in sectors like construction, which she said is "not a story to celebrate."

Protesters have gathered outside the Israeli embassy in Bangkok to protest against offensive behaviour by Israeli tourists, and alleged inappropriate land acquisition strategies. The Trump loyalty test: Will his supporters ever disagree with him?

One of Meta's Offices Was Briefly Overtaken by a Rogue Squirrel The animal escaped after apparently arriving inside a package at Meta's Bangkok office, injuring one employee before finally being caught. Meta's year so far hasn't exactly been a picnic. A squirrel apparently got loose inside a building in Bangkok, Thailand, where some of the tech giant's regional teams are based. The critter spent at least 20 minutes darting past staff, according to an internal memo seen by WIRED. It noted that the squirrel minorly injured a janitor before finally being caught.

Inference-time LLM alignment methods, particularly activation steering, offer an alternative to fine-tuning by directly modifying activations during generation. Existing methods, however, often rely on non-anticipative interventions that ignore how perturbations propagate through transformer layers and lack online error feedback, resulting in suboptimal, open-loop control. To address this, we show empirically that, despite the nonlinear structure of transformer blocks, layer-wise dynamics across multiple LLM architectures and scales are well-approximated by locally-linear models. Exploiting this property, we model LLM inference as a linear time-varying dynamical system and adapt the classical linear quadratic regulator to compute feedback controllers using layer-wise Jacobians, steering activations toward desired semantic setpoints in closed-loop with minimal computational overhead and no offline training. We also derive theoretical bounds on setpoint tracking error, enabling formal guarantees on steering performance. Using a novel adaptive semantic feature setpoint signal, our method yields robust, fine-grained behavior control across models, scales, and tasks, including state-of-the-art modulation of toxicity, truthfulness, refusal, and arbitrary concepts, surpassing baseline steering methods. Our code is available at: https://github.com/trustworthyrobotics/lqr-activation-steering

Supervised machine learning assumes that labeled data provide accurate measurements of the concepts models are meant to learn. Yet in practice, human labeling introduces systematic variation arising from ambiguous items, divergent interpretations, and simple mistakes. Machine learning research commonly treats all disagreement as noise, which obscures these distinctions and limits our understanding of what models actually learn. This paper reframes annotation as a measurement process and introduces a statistical framework for decomposing labeling outcomes into interpretable sources of variation: instance difficulty, annotator bias, situational noise, and relational alignment. The framework extends classical measurement-error models to accommodate both shared and individualized notions of truth, reflecting traditional and human label variation interpretations of error, and provides a diagnostic for assessing which regime better characterizes a given task. Applying the proposed model to a multi-annotator natural language inference dataset, we find empirical evidence for all four theorized components and demonstrate the effectiveness of our approach. We conclude with implications for data-centric machine learning and outline how this approach can guide the development of a more systematic science of labeling.

We propose a neural network model for contextual regression in which the regression model depends on contextual features that determine the active submodel and an algorithm to fit the model. The proposed simple contextual neural network (SCtxtNN) separates context identification from context-specific regression, resulting in a structured and interpretable architecture with fewer parameters than a fully connected feed-forward network. We show mathematically that the proposed architecture is sufficient to represent contextual linear regression models using only standard neural network components. Numerical experiments are provided to support the theoretical result, showing that the proposed model achieves lower excess mean squared error and more stable performance than feed-forward neural networks with comparable numbers of parameters, while larger networks improve accuracy only at the cost of increased complexity. The results suggest that incorporating contextual structure can improve model efficiency while preserving interpretability.

Knowledge distillation has emerged as a powerful technique for compressing large language models (LLMs) into efficient, deployable architectures while preserving their advanced capabilities. Recent advances in low-rank knowledge distillation, particularly methods like Low-Rank Clone (LRC), have demonstrated remarkable empirical success, achieving comparable performance to full-parameter distillation with significantly reduced training data and computational overhead. However, the theoretical foundations underlying these methods remain poorly understood. In this paper, we establish a rigorous theoretical framework for low-rank knowledge distillation in language models. We prove that under mild assumptions, low-rank projection preserves the optimization dynamics, yielding explicit convergence rates of $O(1/\sqrt{T})$. We derive generalization bounds that characterize the fundamental trade-off between model compression and generalization capability, showing that the generalization error scales with the rank parameter as $O(r(m+n)/\sqrt{n})$. Furthermore, we provide an information-theoretic analysis of the activation cloning mechanism, revealing its role in maximizing the mutual information between the teacher's and student's intermediate representations. Our theoretical results offer principled guidelines for rank selection, mathematically suggesting an optimal rank $r^* = O(\sqrt{n})$ where $n$ is the sample size. Experimental validation on standard language modeling benchmarks confirms our theoretical predictions, demonstrating that the empirical convergence, rank scaling, and generalization behaviors align closely with our bounds.