Discussion | Link

Discussion | Link

Wall Street Journal: A look at ADI Predictstreet's troubled launch, marked by tiny trading volumes, withdrawal bugs, and other issues, as it tried to challenge Kalshi and Polymarket  —  ADI Predictstreet bungled a World Cup ticket giveaway, users say  —  Vukašin Petrović was elated …

Altimeter Capital CEO Brad Gerstner said Trump wants to automatically create Trump Accounts for all American children. Anna Moneymaker/Getty Images Donald Trump wants children to auto-enroll into Trump Accounts, the accounts' lead architect said. Investor Brad Gerstner said Trump wants up to 70 million children to have the investment accounts. Think tanks and researchers agree that that's the best way to increase adoption. The man who spearheaded Trump Accounts said President Donald Trump wants to automatically open accounts for all American children. In an "All-In Podcast" interview released on Saturday, Altimeter Capital CEO Brad Gerstner spoke about the early success of Trump Accounts and how Trump wants to increase adoption. "In fact, the president suggested that we're going to auto-create accounts for all 50 million kids — or upwards of 70 million kids — under the age of 18," Gerstner said. "So he called on us to get the accounts open faster, for more people, to have more impact, and to make sure no child is left behind," he added. Gerstner was one of the architects of Trump Accounts, an investment vehicle for children under 18 that gives them early access to the stock market. Eligible newborns born between 2025 and 2028 will receive $1,000 deposited into their accounts by the US Treasury. Currently, the program is opt-in, requiring parents to create accounts for their children by submitting an IRS Form 4547 with their tax filings or online through the Trump Accounts website or app. Later in the interview, one of the podcast hosts, angel investor Jason Calacanis, asked if Gerstner would be following Trump's instructions to auto-create accounts using Social Security numbers of people under 18. Gerstner replied that it was the team's intention to get up to 70 million accounts created "over the course of the next 90 days using all of this data." But they needed to work with the Treasury, the White House, and the Social Security Administration to make it happen, he said. Trump Accounts were launched on July 4, and private donors like Michael and Susan Dell and SpaceX President Gwynne Shotwell have contributed millions to the accounts. Companies like Goldman Sachs, Micron, and Bank of America have pledged to contribute to their employees' children's accounts. In an X post on Thursday, the White House wrote that American families contributed nearly $125 million to Trump Accounts in the first five days since its launch. Gerstner said during the All-In interview that "over a billion dollars" have been deposited into the accounts in total. Researchers and think tanks have urged an auto-enrollment structure for the accounts since they were announced last year. The think tank Urban Institute said in a January article that opt-in programs have historically seen low participation rates, particularly among low-income families. And a September study published by the Washington University Brown School's Center for Social Development wrote that automatic enrollment would "enable the policy to avoid the high costs and low participation rates of programs with opt-in enrollment structures." The White House did not respond to a request for comment from Business Insider. Read the original article on Business Insider

Satya Nadella said that it was hypocritical for model makers to complain about distillation. Bloomberg/Getty Images Satya Nadella criticized AI labs for complaining about distillation attacks. Model distillation involves training AI using more advanced pre-existing AI models. Nadella suggested that reliance on leading AI models compromises data ownership and control. Satya Nadella took a quiet swipe at AI labs like Anthropic for how they train their models. In an X post on Sunday, the Microsoft CEO said that model makers complaining about distillation is hypocritical. Distillation is the process of training a less powerful model based on the outputs of a stronger one. "While the great innovation that comes from model providers having fair use rights to train models on public data is needed, I find it ironic that the status quo is to then turn around and impose restrictive terms on distillation, and to reserve the right to learn from customer usage and interaction data," Nadella wrote. He added that if learning only flows in one direction, owners of the learning infrastructure make all the money while creators of the knowledge get left out. Frontier AI model makers like Anthropic, OpenAI, and Google DeepMind rely on work created by others to train their own models. ChatGPT, Claude, and Gemini acquire their "intelligence" from publicly available writing, images, and other data. Numerous companies and individuals have sued the leading AI labs over nonconsensual content "scraping." Though the lab was not named, Nadella's comments seemed especially targeted toward Anthropic. Earlier this year, Anthropic CEO Dario Amodei complained that Chinese model makers are stealing his company's work, using Claude to train their own models. Last month, Anthropic wrote a letter to South Carolina Sen. Tim Scott and Massachusetts Sen. Elizabeth Warren saying that Alibaba had recently carried out "the largest known distillation attack" on it to date. "Competitors can use it to acquire powerful capabilities from other labs in a fraction of the time, and at a fraction of the cost, that it would take to develop them independently," Anthropic said in a lengthy statement on the subject in February. Alibaba did not publicly respond to Anthropic's accusations at the time. In Sunday's blog post, Nadella warned that companies relying on leading models are essentially handing over their proprietary data and then paying to use them. He said companies should own their AI infrastructure and institutional knowledge rather than rely on any single model vendor. They should also conduct their own evaluations and their own "learning loop," allowing their AI capabilities to improve continuously over time. "That is why enterprises need a real trust boundary for their human capital and token capital to compound," he said. "And it is a hard boundary across which nothing crosses, not even the intelligence exhaust, without consent." Elon Musk has also criticized Anthropic for how it collects data and trains its models. "Anthropic is guilty of stealing training data at massive scale and has had to pay multi-billion dollar settlements for their theft. This is just a fact," Musk wrote in a February X post, following Anthropic's complaint against Chinese models. Anthropic did not immediately respond to a request for comment from Business Insider. Read the original article on Business Insider

Wen-Yee Lee / Reuters: A Taiwanese minister says TSMC plans two more advanced chip packaging plants in Chiayi Science Park, with all four expected to generate $9.35B+ in annual output  —  Taiwan Semiconductor Manufacturing Co (TSMC) (2330.TW) will add two advanced chip packaging plants in the Chiayi Science Park …

Nathan Lambert / Interconnects AI: Open weight AI models are facing an existential policy test in the US, with Anthropic leading a campaign against Chinese models over distillation concerns  —  Staring down the barrel of policy action that could make open models a permanent second class citizen.  —  ∙ Paid

arXiv:2607.08839v1 Announce Type: new Abstract: Multimodal Large Language Models (MLLMs) are typically designed under the assumption that all modalities available during training will also be accessible at inference. However, many real-world settings violate this assumption, requiring models to operate under a privileged modality setting, where auxiliary modalities are available only during training. While these modalities contain valuable information, existing MLLMs largely fail to leverage them effectively, as they treat modalities as interchangeable inputs rather than sources of complementary supervision. We propose Mixture of Probes (MoP), a novel framework that disentangles modality-specific and modality-general signals within the MLLM, allowing the model to preserve modality-dependent structure while learning transferable representations across modalities. At its core, MoP achieves this through a structured probing mechanism that extracts and organizes information from intermediate representations of a shared modality encoder, rather than relying only on final-layer alignment as done in existing MLLMs. To support this disentanglement, we further introduce MoP Cross-modal Training (MoP-X), a training strategy for MoP centered around a probe disentanglement loss that prevents probe collapse and encourages cross-modal learning. We evaluate MoP across two domains spanning eight tasks and four modalities under a comprehensive evaluation protocol tailored to the privileged modality setting, where each modality is independently treated as the sole input at inference time. MoP consistently outperforms strong MLLM baselines, achieving up to 65% relative improvement, demonstrating that auxiliary modalities, even when unavailable at inference, can provide substantial gains when effectively leveraged during training. Code, model checkpoints, and evaluation protocols will be made available at https://github.com/Sony/MoP.

arXiv:2607.09094v1 Announce Type: new Abstract: Legal precedent retrieval is a fundamental task in legal case preparation, planning, litigation strategy, and legal research. Current approaches for automatic precedent retrieval map legal documents to a low-dimensional semantic space and compute similarity based on the proximity of their representations. These approaches treat legal documents as monolithic texts, ignoring the rhetorical organization of the legal technicalities. Ergo, they overlook nuanced legal meanings and fail to distinguish the contextual significance of legal entities and concepts that vary based on their rhetorical roles within the document. To address this insufficiency, we propose the PRecG pipeline that computes the similarity between pairs of legal judgments by hierarchically learning their representations. The process begins by decomposing each document into distinct semantic units (segments) based on the rhetorical roles of sentences. For each rhetorical segment, a knowledge graph is constructed to capture the legal entities and their relationships within the segment. Contextual representations of the entities are then learned and aggregated to derive segment-level embeddings. These embeddings are further integrated to produce a unified document-level representation, and finally, the semantic similarity between a pair of documents is computed. We validate the performance of the proposed approach through extensive experiments on a benchmark Indian legal dataset, comparing it against state-of-the-art baselines to demonstrate its effectiveness.

arXiv:2607.08776v1 Announce Type: new Abstract: Despite the success of knowledge distillation (KD) in Large Language Models (LLMs), the underlying mechanism behind its efficacy remains unclear. In this paper, we propose a unified approach to explore the common mechanism of various KD methods using interactions. Specifically, we decompose the output score of the LLM into the sum of numerous interactions. Each interaction represents a nonlinear relationship involving a set of input variables (e.g., words). Based on the decomposed interactions, we discover that the common mechanism underlying various KD methods is the sparsification of interactions, i.e., student models retain fewer interactions for inference while suppressing other interactions to zero effects. Furthermore, we discover that the performance variance across different KD methods arises from their capabilities in handling complex interactions. A KD method typically yields better performance if it enables the student model to achieve higher sparsity of complex interactions. Motivated by these insights, we propose a plug-and-play loss function called Complex Interaction Penalty (CIP) to explicitly enforce the sparsity of complex interactions during the distillation process. Extensive experiments demonstrate that integrating CIP consistently improves the performance of diverse KD methods on both in-domain and out-of-distribution benchmarks.

arXiv:2607.08778v1 Announce Type: new Abstract: Alzheimer's Disease (AD) is a complex neurodegenerative disorder that continues to impact millions of people worldwide. Predicting AD conversion during the prodromal stage remains critical for disease understanding and patient care. As such, survival models are widely used for AD risk prediction, yet they are typically static predictors with limited interpretability and no capacity for natural language reasoning. In this work, we propose iLENS, an interpretable large language model (LLM) guided framework based on mixture-of-experts (MoE) for survival prediction in AD conversion. Our approach uses LLM to synthesize structured neuroimaging measurements and unstructured information to guide expert routing. Our framework demonstrates competitive predictive performance and capability in patient subtyping. Furthermore, our framework provides transparent, biologically grounded rationales for its routing decisions, bridging the gap between high-performance survival analysis and interpretable clinical decision support.

arXiv:2607.09121v1 Announce Type: new Abstract: In this study, we examine the opportunities brought by Large Language Models (LLMs) to various aspects of fundamental analysis of companies based on their reports as well as data and documents describing macroeconomic situation like GDP and inflation changes as well as documents filled to the U.S. Securities and Exchange Commission (SEC) which can be found in EDGAR. We were preprocessing those data and than sending via API to gpt-4o model in a Retrieval-Augmented Generation (RAG) like regime. We prepared as well a document describing an exemplar investor knowledge based on Kitchin cycles. We were scanning data important for analysis of 9 companies for 4 weeks. Using LLM we were producing automatic briefs about them. They were sent to nine participants who are individual investors to evaluate usefulness of such approach to data analysis.

arXiv:2607.09204v1 Announce Type: new Abstract: Pretrained language models often exhibit structured weight spectra, suggesting that training may repeatedly produce similar layerwise and component-wise organization. We ask whether these recurring spectral patterns can be reused as an initialization signal for GPT-2-style language-model pretraining. First, we analyze eleven pretrained GPT-2-style checkpoints that vary in size, language, tokenizer, and training corpus, measuring Frobenius norm and effective-rank entropy across layers and Transformer subcomponents. The checkpoints show shared depth trends, especially increasing scale and stronger spectral concentration in residual-writing matrices. We then construct initialization schemes that imitate the component-wise magnitudes and spectral profiles of pretrained models, and compare them with several weight initialization methods. These initializers visibly change the model's structural spectral patterns, but the evaluation results do not show a corresponding performance advantage. Pretrained-weight reuse remains competitive, while coarse spectral matching alone is not a reliable optimization strategy. Our results suggest that pretrained spectra are useful diagnostics of trained model structure, but that effective reuse likely requires preserving richer information than component-wise scale and singular-value shape.

arXiv:2607.09291v1 Announce Type: new Abstract: Medieval document transcribers have very different practices; on top of that, heterogeneous digitization policies have resulted in corpora where the character-set must be viewed as fluid. In this paper we address the problem of changing between character-sets in a flexible manner. We focus on one-to-one character mappings and train characterlevel one-to-one RNNs to undo them with self-supervision; recovering half the CER even with 20 text lines. We analyse the use of these one-to-one networks for HTR post-correction and we see that they obtain significant improvements while totally ignoring ins-dels. We then use the exact same networks with character-level alignment groundtruth compiled from parallel corpora in a training and inference mode we call Banded RNNs. We use such networks to successfully expand abbreviations in medieval charter transcriptions. Finally we introduce an elaborate heuristic which takes the characters of two arbitrary character-sets and defines a metric encapsulating what we consider to be semantic similarity of characters. We call the construction of such mappings letter lemmatization and present a rich Python library that efficiently performs all presented methods.

arXiv:2607.09306v1 Announce Type: new Abstract: Language models are optimised for scale, yet remain functional rather than companionable, and as an assistant personalises into a companion, accumulating memory of one user, it quietly becomes someone, and can silently acquire traits that harm that user. What a companion is becoming, and what would make it worth becoming, has no reliable instrument: trained human raters cannot agree on the answer (Fleiss kappa = 0.074). Here we show that three small language models (146 M to 3 B parameters) sharing a hyperbolic substrate answer both halves of that question. A 146 M behavioural auditor, trained from scratch, detects the compliance gap that those raters cannot (90.7% binary-compliance accuracy); a linear read-out of its frozen representation further detects companion-induced sycophancy, dependence-fostering and confabulated memories on generator families unseen in training (AUROC 0.804 under style-controlled, leave-one-generator-out evaluation, versus 0.721 for a frontier zero-shot judge on the same items). A creative frame-seeder is preferred in 100% of 311 decided pairwise comparisons over four prompting baselines. A memory operating system implements designed forgetting, M(t) = S*exp(-lambda*t), whose predicted skeleton-wallpaper partition emerges only under selective retrieval gating in a four-condition pilot. Creativity, honesty and designed forgetting constitute a small-model route to trustworthy companion AI.

arXiv:2607.08808v1 Announce Type: new Abstract: Recent advances in 3D Gaussian Splatting (3DGS) have enabled high-quality, render-ready scene representations for novel-view synthesis. However, most existing 3DGS pipelines rely on multi-view observations (or non-causal access to future frames) to achieve sufficient coverage, which is often unavailable in on-device robotics and AR settings where sensing is restricted to a single stereo rig. Recovering a high-quality 3DGS scene from one stereo observation, therefore, remains challenging due to occlusions, limited field of view, and missing geometry. We present StereoSplat+, a diffusion-enhanced feed-forward framework that enables causal reconstruction from a single stereo pair. Our method builds on two key components. First, we propose StereoSplat, an input-invariant feed-forward 3D Gaussian estimator that takes a variable number of posed stereo pairs as input and predicts high-quality 3D Gaussians. StereoSplat fuses complementary geometry cues via a cost-volume branch and a triplane-based 3D volume branch and leverages continuous pose encoding to generalize across view counts and camera configurations. Second, since multiple posed stereo pairs are typically unavailable at inference time, we introduce a diffusion-enhanced one-shot progressive inference scheme called StereoSplat+: starting from one stereo pair, we render novel stereo views from the predicted 3DGS, refine them with a one-step diffusion enhancer, and feed them back as additional inputs to update the 3DGS. Experiments on the KITTI-360 dataset show that StereoSplat+ improves novel-view rendering quality and geometry accuracy, especially in occluded regions and under strong view extrapolation, outperforming recent feed-forward 3DGS baselines.

arXiv:2607.08867v1 Announce Type: new Abstract: Cloud-based deep learning enables large-scale medical image analysis but raises significant privacy concerns when sensitive patient images are outsourced for model development. Image disguising has recently emerged as a promising privacy-enhancing technology (PET) that transforms images into visually unintelligible representations while preserving information for downstream learning. We established a unified framework to evaluate representative methods, DisguisedNets and NeuraCrypt, across four datasets involving classification and semantic segmentation tasks. Our analysis assessed predictive utility, efficiency, and robustness against reconstruction attacks. Results showed that image disguising performance varies significantly between tasks; while methods preserved utility for medical image classification, they caused substantial degradation in dense semantic segmentation. Specifically, Randomized Multidimensional Transformation (RMT) offered the optimal balance of performance and security, whereas AES-based disguising severely impacted utility. Furthermore, regression-based reconstruction attacks effective on natural images proved considerably less successful on realistic medical images. These findings provide a systematic assessment of PET suitability for confidential medical AI applications.

arXiv:2607.08879v1 Announce Type: new Abstract: Indoor scene relighting demands photorealism, precise spatial control, and strict multi-view consistency. While diffusion-based image editing models enable semantic lighting manipulation via text prompts, enforcing exact 3D light placement often disrupts their generative priors. We propose Lume-Palette, a progressive framework that leverages semantic lighting priors for spatially controllable multi-view indoor relighting. The approach decouples relighting into two stages: (1) illumination distillation, which extracts canonical illumination palettes from a pretrained diffusion model to preserve realistic material-light interactions, and (2) illumination casting, which explicitly maps target spatial lighting conditions defined from coarse 3D geometry. To efficiently handle dense multi-view and multi-modal inputs, we introduce an asymmetric multi-view conditioning strategy that selectively injects essential spatial context. Experiments on diverse synthetic scenes and real-world scenes demonstrate that Lume-Palette produces photorealistic, spatially controllable, and multi-view consistent relighting results. Project Page: https://cjeen.github.io/lumepalette

arXiv:2607.08781v1 Announce Type: new Abstract: The coupling in flow matching -- the rule pairing noise vectors with data points -- is typically treated as a computational choice. We show that this coupling can instead serve as an alignment interface: by matching noise and data according to a target molecular property, it embeds controllable structure directly into the learned flow field. Building on this view, we introduce Reward Transport, which uses optimal transport coupling at training time to align a scalar noise-space coordinate with molecular rewards; at inference, varying this coordinate steers the generated distribution without requiring an oracle, reward model, gradient guidance, or additional computation. In the coupling-preserving limit, thresholding this coordinate recovers the Cross-Entropy Method's truncated reward distribution, providing a principled, continuously adjustable distribution-level control knob. Empirically, on ZINC-250K and GuacaMol, sweeping the scalar induces monotone control of logP and consistent QED control over its operating range; most tellingly, the same knob produces opposite structural responses for different targets, growing molecules for logP but shrinking them for QED, which rules out a generic size bias. The interface is complementary to classifier-free guidance and conditional flow matching, while a negative result under epsilon-prediction diffusion clarifies where coupling-level alignment is structurally absent. Code: https://github.com/KehanGuo2/reward-transport

arXiv:2607.08780v1 Announce Type: new Abstract: Mixture-of-Experts (MoE) models activate only a sparse subset of experts per token, yet consecutive tokens frequently activate different experts -- causing constant weight swapping between slow storage and fast memory on edge devices. Existing remedies are either system-level (caching heuristics) or post-hoc (router fine-tuning), leaving the root cause unchanged during pretraining. We propose StickyMoE, a differentiable routing consistency loss that penalises abrupt expert switches between adjacent tokens, encouraging the router to maintain the same expert assignment across semantically coherent spans. StickyMoE requires no architectural changes, adds a single hyperparameter lambda, and unlike post-hoc methods, allows expert representations and routing decisions to co-adapt from the first training step. Experiments on small-scale MoE language models show that StickyMoE reduces the expert switch rate by up to 60% with less than 4% perplexity degradation, Pareto-dominating post-hoc fine-tuning on the quality-locality frontier. Routing temporal locality is most efficiently instilled at training time.