In-context learning (ICL) has emerged as a powerful capability of large language models (LLMs), enabling them to perform new tasks based on a few provided examples without explicit fine-tuning. Despite their impressive adaptability, these models remain vulnerable to subtle adversarial perturbations and exhibit unpredictable behavior when faced with linguistic variations. Inspired by software testing principles, we introduce a software testing-inspired framework, called MMT4NL, for evaluating the trustworthiness of in-context learning by utilizing adversarial perturbations and software testing techniques. It includes diverse evaluation aspects of linguistic capabilities for testing the ICL capabilities of LLMs. MMT4NL is built around the idea of crafting metamorphic adversarial examples from a test set in order to quantify and pinpoint bugs in the designed prompts of ICL. Our philosophy is to treat any LLM as software and validate its functionalities just like testing the software. Finally, we demonstrate applications of MMT4NL on the sentiment analysis and question-answering tasks. Our experiments could reveal various linguistic bugs in state-of-the-art LLMs.

Large Language Models (LLMs) as stochastic systems may generate numbers that deviate from available data, a failure known as \emph{numeric hallucination}. Existing safeguards -- retrieval-augmented generation, citations, and uncertainty estimation -- improve transparency but cannot guarantee fidelity: fabricated or misquoted values may still be displayed as if correct. We propose \textbf{Proof-Carrying Numbers (PCN)}, a presentation-layer protocol that enforces numeric fidelity through mechanical verification. Under PCN, numeric spans are emitted as \emph{claim-bound tokens} tied to structured claims, and a verifier checks each token under a declared policy (e.g., exact equality, rounding, aliases, or tolerance with qualifiers). Crucially, PCN places verification in the \emph{renderer}, not the model: only claim-checked numbers are marked as verified, and all others default to unverified. This separation prevents spoofing and guarantees fail-closed behavior. We formalize PCN and prove soundness, completeness under honest tokens, fail-closed behavior, and monotonicity under policy refinement. PCN is lightweight and model-agnostic, integrates seamlessly into existing applications, and can be extended with cryptographic commitments. By enforcing verification as a mandatory step before display, PCN establishes a simple contract for numerically sensitive settings: \emph{trust is earned only by proof}, while the absence of a mark communicates uncertainty.

Instruction Fine-Tuning (IFT) has been widely adopted as an effective post-training strategy to enhance various abilities of Large Language Models (LLMs). However, prior studies have shown that IFT can significantly compromise LLMs' safety, particularly their ability to refuse malicious instructions, raising significant concerns. Recent research into the internal mechanisms of LLMs has identified the refusal direction (r-direction) in the hidden states, which plays a pivotal role in governing refusal behavior. Building on this insight, our study reveals that the r-direction tends to drift during training, which we identify as one of the causes of the associated safety risks. To mitigate such drift, our proposed ProCon method introduces a projection-constrained loss term that regularizes the projection magnitude of each training sample's hidden state onto the r-direction. Our initial analysis shows that applying an appropriate constraint can effectively mitigate the refusal direction drift and associated safety risks, but remains limited by overall performance barriers. To overcome this barrier, informed by our observation of early-stage sharp drift and a data-driven perspective, we introduce a warm-up strategy that emphasizes early-stage strong constraints and broaden the data distribution to strengthen constraint signals, leading to an enhanced ProCon method. Experimental results under various datasets, scenarios, and LLMs demonstrate that our method can significantly mitigate safety risks posed by IFT while preserving task performance gains. Even compared with strong baselines, our method consistently delivers superior overall performance. Crucially, our analysis indicates that ProCon can contribute to stabilizing the r-direction during training, while such an interpretability-driven exploration of LLMs' internal mechanisms lays a solid foundation for future safety research.

Identifying relationships between variables is a fundamental task in science. Among various approaches, linear regression plays a central role in linking explanatory variables to dependent variables in statistical modeling [1, 2]. Linear regression is useful in physics [3, 4] for extracting equations of motion from time series data [5] and for predicting trends in dynamical systems [6], but its simplicity, interpretability, and predictive power make it a cornerstone of data analysis [7], forecasting [8], and decision-making [9] in many fields. Moreover, linear regression forms the foundation for many advanced statistical and machine learning models [10], including logistic regression [11], support vector machines [12], and generalized linear models [13]. Extensions of linear regression often aim to capture more complex relationships by introducing higher-order polynomial terms or additional nonlinear transformations. Modern developments in machine learning have enabled the training of deep and highly overparameterized models capable of modeling intricate patterns far beyond the reach of simple linear approaches. In particular, deep learning models can be interpreted as sophisticated forms of nonlinear regression [14], capable of approximating complex functions with high flexibility. Despite its utility, linear regression struggles with modern high-dimensional datasets where only a small subset of variables is truly informative.

Trump's Epstein crisis explodes as lewd birthday letter showing president's signature is revealed Judge's'promise' let career criminal walk free to butcher Ukrainian refugee after his MOM said he should be locked up'She was so f***ed up': Carolyn Bessette's friends tell MAUREEN CALLAHAN of her secret Daddy issue, JFK Jr's murder brag that drove her mad... and why everything we know about her is a lie The chaos behind when Meghan Markle was told not to be at Queen Elizabeth II's deathbed They were locked in a dungeon inside a house of horrors. But incredible footage shows five kids' daring acts while their parents were out... and it left neighbors speechless Turn back the clock with the K-beauty retinol cream Amazon shoppers say leaves their skin'silky smooth' - and it's now $10 Scientists crack the ultimate answer to the meaning of life... and it's hidden among 38M obituaries CBS News hires a CONSERVATIVE to police interviews after Trump and Noem'deceptive' editing fury Scientist claims life on Earth was not random... but engineered Supreme Court LIFTS restrictions on Trump's immigration raids despite claims agents targeted people by race I was 52 with a collapsed'turkey neck'. Here's how I turned back the clock 10 years Plastic surgeons weigh in on Jessica Simpson's dramatic new look at VMAs as fans declare her'unrecognizable' Billionaire turns his back on Trump as he blasts President's'risky' financial move that could cost Americans their savings Trump loses appeal and must pay $83 million to E. Jean Carroll AMANDA PLATELL: Harry is'desperate' to come back to Britain and reclaim his royal role - but this fresh snub from William makes it clear why it will never happen... and why he'll never forgive his brother Scientists crack the ultimate answer to the meaning of life... and it's hidden among 38million obituaries Scientists on a mission to uncover what constitutes a life well lived found the answer after analyzing 38 million obituaries from the US spanning 30 years. Using automated text analysis tools, the team found that the most commonly celebrated values were tradition and benevolence. Nearly 80 percent of obituaries highlighted respect for customs or religion, while 76 percent emphasized caring, reliability and trustworthiness.

The effective altruism movement, led by RAND CEO Jason Matheny, may challenge Trump's plans for American artificial intelligence dominance through regulatory advocacy.

Technology Engineering The'Star Trek' technology that came to real life Breakthroughs, discoveries, and DIY tips sent every weekday. To celebrate Star Trek Day on September 8, the European Space Agency (ESA) released a video of the Star Trek technology that's made it real-life space. So while we still don't have teleporters or deflector shields, ISS astronauts kind of have tricorders like the one used by Captain Christopher Pike in the first episode of the original series. We've also seen the development of technology that resembles Replicators, VISOR, and PADDs. The original premiered on network television in the United States on September 8, 1966.

Turkish police stormed the Istanbul offices of the main opposition Republican People's Party's (CHP), using shields and pepper spray to enforce a court ruling ousting provincial head Ozgur Celik. Party members barricaded themselves inside with furniture to block police. Nepal'Gen Z' protest death toll climbs, parliament stormed Israel wants to'destroy Gaza City, not occupy it'

The Great Canadian Treasure Hunt's first clue is a 13-stanza poem. Breakthroughs, discoveries, and DIY tips sent every weekday. An actual treasure chest filled with around $1 million in gold coins is hidden somewhere in Canada. However, the mystery isn't tied to a centuries' old pirate bounty or unsolved bank heist, however. These riches were instead intentionally hidden by a mining consortium to celebrate the country's "rich mining heritage and spirit of adventure."

Nepal'Gen Z' protest death toll climbs, parliament stormed NewsFeed Nepal'Gen Z' protest death toll climbs, parliament stormed At least 19 people have been killed in clashes between security forces and protesters in Nepal. Mostly young'Gen Z' demonstrators took to the streets and stormed parliament amid anger over a social media ban and corruption. Israel wants to'destroy Gaza City, not occupy it'