Vision-Language Models (VLMs) have shown strong performance in tasks like visual question answering and multimodal text generation, but their effectiveness in scientific domains such as materials science remains limited. While some machine learning methods have addressed specific challenges in this field, there is still a lack of foundation models designed for broad tasks like polymer property prediction using multimodal data. In this work, we present a multimodal polymer dataset to fine-tune VLMs through instruction-tuning pairs and assess the impact of multimodality on prediction performance. Our fine-tuned models, using LoRA, outperform unimodal and baseline approaches, demonstrating the benefits of multimodal learning. Additionally, this approach reduces the need to train separate models for different properties, lowering deployment and maintenance costs.

Translating the internal representations and computations of models into concepts that humans can understand is a key goal of interpretability. While recent dictionary learning methods such as Sparse Autoencoders (SAEs) provide a promising route to discover human-interpretable features, they suffer from a variety of problems, including a systematic failure to capture the rich conceptual information that drives linguistic understanding. Instead, they exhibit a bias towards shallow, token-specific, or noisy features, such as "the phrase 'The' at the start of sentences". In this work, we propose that this is due to a fundamental issue with how dictionary learning methods for LLMs are trained. Language itself has a rich, well-studied structure spanning syntax, semantics, and pragmatics; however, current unsupervised methods largely ignore this linguistic knowledge, leading to poor feature discovery that favors superficial patterns over meaningful concepts. We focus on a simple but important aspect of language: semantic content has long-range dependencies and tends to be smooth over a sequence, whereas syntactic information is much more local. Building on this insight, we introduce Temporal Sparse Autoencoders (T-SAEs), which incorporate a novel contrastive loss encouraging consistent activations of high-level features over adjacent tokens. This simple yet powerful modification enables SAEs to disentangle semantic from syntactic features in a self-supervised manner. Across multiple datasets and models, T-SAEs recover smoother, more coherent semantic concepts without sacrificing reconstruction quality. Strikingly, they exhibit clear semantic structure despite being trained without explicit semantic signal, offering a new pathway for unsupervised interpretability in language models.

In this paper we propose a sequential minimax optimization (SMO) method for solving a class of constrained bilevel optimization problems in which the lower-level part is a possibly nonsmooth convex optimization problem, while the upper-level part is a possibly nonconvex optimization problem. Specifically, SMO applies a first-order method to solve a sequence of minimax subproblems, which are obtained by employing a hybrid of modified augmented Lagrangian and penalty schemes on the bilevel optimization problems. Under suitable assumptions, we establish an operation complexity of $O(\varepsilon^{-7}\log\varepsilon^{-1})$ and $O(\varepsilon^{-6}\log\varepsilon^{-1})$, measured in terms of fundamental operations, for SMO in finding an $\varepsilon$-KKT solution of the bilevel optimization problems with merely convex and strongly convex lower-level objective functions, respectively. The latter result improves the previous best-known operation complexity by a factor of $\varepsilon^{-1}$. Preliminary numerical results demonstrate significantly superior computational performance compared to the recently developed first-order penalty method.

The SpaceX founder's seven-figure grant supports digital archaeology projects documenting Roman heritage, with Musk suggesting AI could help fill historical knowledge gaps.

A Swedish man discovered the 12th century buried treasure near his summer home. Breakthroughs, discoveries, and DIY tips sent every weekday. It only costs a few dollars to buy a tub of bait worms for fishing, but many people are fine with sourcing them straight from the ground. There's always a chance you may find more in the dirt than wriggling invertebrates. Take a recent example near Stockholm, Sweden: According to county officials last month, an unnamed fisherman scrounging for worms at his summer house discovered a corroded copper cauldron containing around 13 pounds of treasure from the Middle Ages.

HMP Wandsworth gets green light to use AI after team sent in to find'quick fixes' after spate of mistakes Artificial intelligence chatbots could be used to stop prisoners from being mistakenly released from jail, a justice minister told the House of Lords on Monday. James Timpson said HMP Wandsworth had been given the green light to use AI after a specialised team was sent in to find "some quick fixes". A double manhunt was launched last week after the incorrect release of a sex offender and a fraudster from the prison in south-west London. Release errors over the past fortnight have been seized upon by opposition MPs as evidence of the helplessness of ministers in the face of chaos within the criminal justice system. David Lammy, the justice secretary, is expected to address parliament about the number of missing prisoners when MPs return on Tuesday. It is understood that AI could be used to read and process paper documents; help staff cross-reference names to ensure that inmates are no longer hiding their past crimes behind aliases; merge different datasets; and calculate release dates and sentences.

Are we closer to a Gaza international peace force? How Israel is using'no war, no peace' model in Gaza How is Israel using PR firms to frame its war? At least two people including a child have been killed in an Israeli drone strike east of Khan Younis in southern Gaza, according to Al Jazeera reporters in the besieged Palestinian territory. Hamas condemned Israel's "daily and continuous violations" since a truce came into effect last month, accusing it of maintaining a campaign of bombardments and demolitions across the besieged enclave. The Israeli military said the Palestinians killed on Monday posed "an immediate threat" to its forces. Israeli forces have also been systematically destroying homes inside the so-called "yellow line", a temporary withdrawal boundary agreed in the ceasefire.

The cephalopod's disappearing act could help your next sunscreen blend in. Breakthroughs, discoveries, and DIY tips sent every weekday. Cephalopods like octopuses, squid, and cuttlefish have the mesmerizing ability to change the color of their skin to camouflage into the surrounding environment. Multiple biological processes involving a natural pigment called xanthommatin drives this unique ability. As such, various industries are interested in using xanthommatin in products such as paint and natural sunscreen, but the pigment has been hard to research.

If the US Has to Build Data Centers, Here's Where They Should Go A new analysis tries to calculate the coming environmental footprint of AI in the US and finds that the ideal sites for data centers aren't where they're being built. A data center for cryptocurrency mining, cloud services, and AI computing in Stutsman County, North Dakota.Video: halbergman/Getty Images Tech companies have invested so much money in building data centers in recent months, it's actively driving the US economy--and the AI race is showing no signs of slowing down. Meta chief Mark Zuckerberg told President Donald Trump last week that the company would spend $600 billion on US infrastructure--including data centers--by 2028, while OpenAI has committed already to spending $1.4 trillion. An extensive new analysis looks at the environmental footprint of data centers in the US to get a handle on what, exactly, the country might be facing as this buildout continues over the next few years--and where the US should be building data centers to avoid the most harmful environmental impacts. The study, published in the journal Nature Communications on Monday, uses a variety of data, including demand for AI chips and information on state electricity and water scarcity, to project the potential environmental impacts of future data centers through the end of the decade. The study models a number of different possible scenarios on how data centers could affect the US and the planet--and cautions that tech companies' net zero promises aren't likely to hold up against the energy and water needs of the massive facilities they're building.

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 LSEG .