This work presents a systematic investigation into the latent knowledge encoded within Network Foundation Models (NFMs). Different from existing efforts, we focus on hidden representations analysis rather than pure downstream task performance and analyze NFMs through a three-part evaluation: Embedding Geometry Analysis to assess representation space utilization, Metric Alignment Assessment to measure correspondence with domain-expert features, and Causal Sensitivity Testing to evaluate robustness to protocol perturbations. Using five diverse network datasets spanning controlled and real-world environments, we evaluate four stateof-the-art NFMs, revealing that they all exhibit significant anisotropy, inconsistent feature sensitivity patterns, an inability to separate the high-level context, payload dependency, and other properties. Our work identifies numerous limitations across all models and demonstrates that addressing them can significantly improve model performance (up to 0.35 increase in F1 scores without architectural changes).

As AI systems become more capable, deceptive behaviors can undermine evaluation and mislead users at deployment. Recent work has shown that lie detectors can accurately classify deceptive behavior, but they are not typically used in the training pipeline due to concerns around contamination and objective hacking. We examine these concerns by incorporating a lie detector into the labelling step of LLM post-training and evaluating whether the learned policy is genuinely more honest, or instead learns to fool the lie detector while remaining deceptive. Using DolusChat, a novel 65k-example dataset with paired truthful/deceptive responses, we identify three key factors that determine the honesty of learned policies: amount of exploration during preference learning, lie detector accuracy, and KL regularization strength. We find that preference learning with lie detectors and GRPO can lead to policies which evade lie detectors, with deception rates of over 85%. However, if the lie detector true positive rate (TPR) or KL regularization is sufficiently high, GRPO learns honest policies. In contrast, off-policy algorithms (DPO) consistently lead to deception rates under 25% for realistic TPRs. Our results illustrate a more complex picture than previously assumed: depending on the context, lie-detector-enhanced training can be a powerful tool for scalable oversight, or a counterproductive method encouraging undetectable misalignment.

Detecting structural change in dynamic network data has wide-ranging applications. Existing approaches typically divide the data into time bins, extract network features within each bin, and then compare these features over time. This introduces an inherent tradeoff between temporal resolution and statistical stability of the extracted features. Despite this tradeoff, reminiscent of time-frequency tradeoffs in signal processing, most methods rely on a fixed temporal resolution. Choosing an appropriate resolution parameter is typically difficult, and can be especially problematic in domains like cybersecurity, where anomalous behavior may emerge at multiple time scales.

Access a subscriber-only eBook of a collection of stories about how militaries are using Al models to make decisions. This ebook is available only for subscribers. A collection of stories about how militaries are using AI models to make decisions. Stories written by James O'Donnel by James O'Donnell A new US phone network for Christians aims to block porn and gender-related content James O'Donnell Musk v. Altman week 1: Elon Musk says he was duped, warns AI could kill us all, and admits that xAI distills OpenAI's models Michelle Kim Launching next week on T-Mobile's network, the cell plan takes a nuclear approach to online safety. Musk v. Altman week 1: Elon Musk says he was duped, warns AI could kill us all, and admits that xAI distills OpenAI's models Musk kept his cool, and OpenAI's lawyer bulldozed him with piercing questions about his motivations for suing the company. China has approved the world's first invasive brain-computer chip--here's what's next The country wants to become a global leader in brain implants.

The company also debuted a new platform for brands wanting to build their own AI glasses. High-end augmented reality and mixed reality devices are set to get a boost thanks to Qualcomm's latest XR chip. During a keynote at Augmented World Expo (AWE), the company unveiled its Snapdragon Reality Elite processor, which it says will allow the next generation of AR and mixed reality headsets to be smaller and more efficient. In terms of specs, the Snapdragon Reality Elite can support up to 4.4K resolution in each eye at 90 fps, a modest upgrade from the XR2+ Gen 2, but one that Qualcomm says will enable better image quality and lower latency. It also delivers significant improvements in terms of efficiency, with up to 20 percent boost in battery life while running up to 12 degrees Celsius (about 54 degrees Fahrenheit) cooler, compared with the XR2+ Gen 2. Performance-wise, Reality Elite comes with notable gains over the previous generation as well.

SoftBank's attempt to get $6 billion OpenAI margin loan stalls SoftBank Group's efforts to secure at least $6 billion through a margin loan backed by its OpenAI stake have stalled after the company lowered its fundraising target. SoftBank Group's talks with potential creditors to raise at least $6 billion from a margin loan backed by its OpenAI stake have stalled, people familiar with the matter said, just weeks after the Japanese conglomerate cut its initial target from $10 billion. The company is considering various fundraising options, according to the people, who asked not to be identified discussing private matters. It could still move forward with the margin loan at a later stage, they added. It's unclear why the margin loan discussions stalled. Borrowers and creditors can pause and revisit fundraising discussions for various reasons, and SoftBank hasn't elaborated on its plans, the people said.

As part of a growing anti-tech movement, startup dumb.co is pushing flip phones as a way for young people to find ‘social and spiritual freedom’

The Chinese company is adapting to the demise of Moore's Law, which guides chip production. It could complicate US chip dominance. Tingbo He, president of Huawei's chip-design subsidiary HiSilicon, says her company's engineers have developed a novel way to optimize semiconductors--and she believes it will close the performance gap between Chinese and Western chips over the next few years. Huawei's method, in short, focuses on speeding up computations across chips, circuits, and entire computing systems, rather than squeezing ever-more components onto a single piece of silicon. "We found a new path," He said at the IEEE International Symposium on Circuits and Systems in Shanghai last weekend.

We study the evolution of hidden-weight spectra in wide neural networks trained by (stochastic) gradient descent. We develop a two-level dynamical mean-field theory (DMFT) that jointly tracks bulk and outlier spectral dynamics for spiked ensembles whose spike directions remain statistically dependent on the random bulk. We apply this framework to two settings: (1) infinite-width nonlinear networks in mean-field/$μ$P scaling and (2) deep linear networks in the proportional high-dimensional limit, where width, input dimension, and sample size diverge with fixed ratios. Our theory predicts how outliers evolve with training time, width, output scale, and initialization variance. In deep linear networks, $μ$P yields width-consistent outlier dynamics and hyperparameter transfer, including width-stable growth of the leading NTK mode toward the edge of stability (EoS). In contrast, NTK parameterization exhibits strongly width-dependent outlier dynamics, despite converging to a stable large-width limit. We show that this bulk+outlier picture is descriptive of simple tasks with small output channels, but that tasks involving large numbers of outputs (ImageNet classification or GPT language modeling) are better described by a restructuring of the spectral bulk. We develop a toy model with extensive output channels that recapitulates this phenomenon and show that edge of the spectrum still converges for sufficiently wide networks.

SoftBank Group has reported a surge in quarterly profit due to valuation gains on its OpenAI investment, boosting confidence at the Japanese company to bet even more on the ChatGPT-maker. The gains on OpenAI outweighed lackluster investment gains elsewhere in the Tokyo-based technology group's portfolio while war in the Middle East roiled markets. That points to growing reliance on the U.S. startup, which faces rising competition from Anthropic and Google and is reportedly trailing its highest internal targets. SoftBank earned a net income of ¥1.83 trillion ($11.6 billion) in its fiscal fourth quarter, compared with the average analyst estimate of ¥295.2 billion. The profit could be attributed entirely to its booking $25 billion in valuation gains on OpenAI in the quarter, according to Bloomberg Intelligence analyst Kirk Boodry. In a time of both misinformation and too much information, quality journalism is more crucial than ever.