As large language models become increasingly capable, there is growing concern that they may develop reasoning processes that are encoded or hidden from human oversight. To investigate whether current interpretability techniques can penetrate such encoded reasoning, we construct a controlled testbed by fine-tuning a reasoning model (DeepSeek-R1-Distill-Llama-70B) to perform chain-of-thought reasoning in ROT-13 encryption while maintaining intelligible English outputs. We evaluate mechanistic interpretability methods--in particular, logit lens analysis--on their ability to decode the model's hidden reasoning process using only internal activations. We show that logit lens can effectively translate encoded reasoning, with accuracy peaking in intermediate-to-late layers. Finally, we develop a fully unsupervised decoding pipeline that combines logit lens with automated paraphrasing, achieving substantial accuracy in reconstructing complete reasoning transcripts from internal model representations. These findings suggest that current mechanistic interpretability techniques may be more robust to simple forms of encoded reasoning than previously understood. Our work provides an initial framework for evaluating interpretability methods against models that reason in non-human-readable formats, contributing to the broader challenge of maintaining oversight over increasingly capable AI systems.

RFK Jr taunts Donald Trump as he shares pointed'Thanksgiving dinner' photo with the president, Elon Musk and Don Jr Fans hail Cece Winans' 'best ever' rendition of the national anthem on Thanksgiving and beg the NFL to get her to the Super Bowl I've seen it too many times - I have to speak up: KENNEDY Trump plunged into security scandal over Afghan shooter's asylum - after president blamed Biden Bryan Kohberger becomes nightmare prison diva... as he throws huge tantrum over BANANAS behind bars My wife was blindsided when I asked for a divorce. There was no foul play or'other woman' but this is why I did it... and the six subtle signs your partner is planning on leaving you too: RICHARD WARNER My book on the Kennedys was used as a'mistress manual' by Olivia Nuzzi... then this wannabe Carolyn Bessette had the nerve to hound me with these outrageous texts: MAUREEN CALLAHAN Americans are finally realizing why we don't eat turkey eggs Plastic surgeon reveals secrets of Tom Brady's changing face, including'unnatural' procedure... and truth about Ozempic use Lilibet's locks steal the show! Meghan's daughter is every inch the little Princess with her fiery red locks in a neat plait at Thanksgiving outing Kimberly Guilfoyle leaves little to the imagination in a figure-hugging sheer lace gown for Thanksgiving dinner in Athens in her role as US Ambassador - after admitting she's'husband hunting' Hollywood stars who REFUSE to celebrate Thanksgiving over animal cruelty and its'blood-soaked' history A strong winter storm is set to hit parts of the US Midwest and Great Lakes region this weekend, threatening flight delays and road safety for millions after the holiday . Winter Storm Watches are now in effect across Illinois, Wisconsin, Iowa, Missouri, Indiana, Michigan, Nebraska, South Dakota and Minnesota, impacting around 50 million Americans. Forecasters warned of potentially heavy snowfall, with accumulations of six to 12 inches or more possible in many areas, especially north of Interstate 70 and along and south of Interstate 90.

In the past few years LLMs have emerged as a tool that can aid programmers by taking natural language descriptions and generating code based on it. However, the reliability of LLM code generation and current validation techniques for it are far from strong enough to be used for mission-critical or safety-critical applications. In this work we explore ways to offer formal guarantees of correctness to LLM generated code; such guarantees could improve the quality of general AI Code Assistants and support their use for critical applications. To address this challenge we propose to incorporate a Formal Query Language that can represent a user's intent in a formally defined but natural language-like manner that a user can confirm matches their intent. We then have a formal specification of the user intent which we can use to verify that LLM-generated code matches the user's intent. We implement these ideas in our system, Astrogator, for the Ansible programming language, widely used for system administration, including for critical systems. The system includes an intuitive formal query language, a calculus for representing the behavior of Ansible programs, and a symbolic interpreter and a unification algorithm which together are used for the verification. A key innovation in Astrogator is the use of a Knowledge Base to capture system-specific implementation dependencies that greatly reduce the need for system knowledge in expressing formal queries. On a benchmark suite of 21 code-generation tasks, our verifier is able to verify correct code in 83% of cases and identify incorrect code in 92%.

This material may not be published, broadcast, rewritten, or redistributed. Quotes displayed in real-time or delayed by at least 15 minutes. Market data provided by Factset . Powered and implemented by FactSet Digital Solutions . Mutual Fund and ETF data provided by Refinitiv Lipper .

Constructing abstract interpreters that provide global soundness guarantees remains a major obstacle in abstract interpretation. We investigate whether modern LLMs can reduce this burden by leveraging them to synthesize sound, non-trivial abstract interpreters across multiple abstract domains in the setting of neural network verification. We formulate synthesis as a constrained optimization problem and introduce a novel mathematically grounded cost function for measuring unsoundness under strict syntactic and semantic constraints. Based on this formulation, we develop a unified framework that unifies LLM-based generation with syntactic and semantic validation and a quantitative cost-guided feedback mechanism. Empirical results demonstrate that our framework not only matches the quality of handcrafted transformers, but more importantly, discovers sound, high-precision transformers for complex nonlinear operators that are absent from existing literature.

Breakthroughs, discoveries, and DIY tips sent every weekday. Illinois has become the first state to enact legislation banning the use of AI tools like ChatGPT for providing therapy. The bill, signed into law by Governor J.B. Pritzker last Friday, comes amid growing research showing an increase in people experimenting with AI for mental health as the country faces a shortage of access to professional therapy services. The Wellness and Oversight for Psychological Resources Act, officially called HB 1806, prohibits healthcare providers from using AI for therapy and psychotherapy services. Specifically, it prevents AI chatbots or other AI-powered tools from interacting directly with patients, making therapeutic decisions, or creating treatment plans.

Data is fundamental to the training of language models (LM). Recent research has been dedicated to data efficiency, which aims to maximize performance by selecting a minimal or optimal subset of training data. Techniques such as data filtering, sampling, and selection play a crucial role in this area. To complement it, we define Data Efficacy, which focuses on maximizing performance by optimizing the organization of training data and remains relatively underexplored. This work introduces a general paradigm, DELT, for considering data efficacy in LM training, which highlights the significance of training data organization. DELT comprises three components: Data Scoring, Data Selection, and Data Ordering. Among these components, we design Learnability-Quality Scoring (LQS), as a new instance of Data Scoring, which considers both the learnability and quality of each data sample from the gradient consistency perspective. We also devise Folding Ordering (FO), as a novel instance of Data Ordering, which addresses issues such as model forgetting and data distribution bias. Comprehensive experiments validate the data efficacy in LM training, which demonstrates the following: Firstly, various instances of the proposed DELT enhance LM performance to varying degrees without increasing the data scale and model size. Secondly, among these instances, the combination of our proposed LQS for data scoring and Folding for data ordering achieves the most significant improvement. Lastly, data efficacy can be achieved together with data efficiency by applying data selection. Therefore, we believe that data efficacy is a promising foundational area in LM training.

Meta on Tuesday said it had struck an agreement to keep one nuclear reactor of a US utility company in Illinois operating for 20 years. Meta's deal with Constellation Energy is the social networking company's first with a nuclear power plant. Other large tech companies are looking to secure electricity as US power demand rises significantly in part due to the needs of artificial intelligence and datacenters. Google has reached agreements to supply its datacenters with nuclear power via a half-dozen small reactors built by a California utility company. Microsoft's similar contract will restart the Three Mile Island nuclear plant, the site of the most serious nuclear accident and radiation leak in US history.

Whether a large and fuzzy social media sensation or deep-sea slug slunking around the ocean's Midnight Zone, there are still so many exciting animals on Earth just waiting for their close-up. In that spirit, here are the 11 of the most exciting animal stories that Popular Science covered this year. A wildlife filmmaker and biology doctoral student took what could be the first picture of a newborn great white shark. Filmmaker Carlos Gauna and University of California, Riverside biology doctoral student Phillip Sternes were looking for sharks near Santa Barbara on California's central coast. Most great whites are gray on top with white bellies, but Gauana's drone camera showed a roughly 5-foot-long shark pup that had more white on its body than normal.

Measures of textual similarity and divergence are increasingly used to study cultural change. But which measures align, in practice, with social evidence about change? We apply three different representations of text (topic models, document embeddings, and word-level perplexity) to three different corpora (literary studies, economics, and fiction). In every case, works by highly-cited authors and younger authors are textually ahead of the curve. We don't find clear evidence that one representation of text is to be preferred over the others. But alignment with social evidence is strongest when texts are represented through the top quartile of passages, suggesting that a text's impact may depend more on its most forward-looking moments than on sustaining a high level of innovation throughout.