After days and weeks of gloomy skies and relentless rain for some, there has finally been a change to our weather in the United Kingdom. Arctic air across the UK means the weekend starts cold and frosty with some snow and ice, especially in northern parts. But, there will be lots of sunshine for most throughout Saturday. However, it will be temporary as rain with more snow and ice spreads overnight into Sunday. Further Met Office yellow warnings for ice and snow have been issued across Scotland and northern England from 21:00 GMT to 10:00 on Sunday.

You would be forgiven for thinking the rain this year has been relentless - because in some parts of the UK, it actually has been. Here at BBC Weather we have been watching computer models closely for signs of when that pattern will change. These computer-generated forecasts go out about two weeks into the future - and models have often been hinting at a change to colder and drier weather on that timescale. However, they have then reverted to the familiar wet pattern as we have got closer to the time. Now though, there are stronger signals of a change for some of us - albeit perhaps only a temporary one.

Pots, quadrants, confederation constraints, group position grids... the 2026 World Cup finals draw on Friday is not going to be a straightforward affair. There's a lot to unpack so we're going to explain it as simply as we can. Luckily, Fifa will have a computer to do most of the heavy lifting and make sure everything runs smoothly. Though as Uefa found out in 2021, sometimes technology does go wrong. Let's hope there will be no gremlins in Washington once the draw ceremony kicks off.

Team GB Olympic and Paralympic athletes are being offered a new form of artificial intelligence-based protection from online abuse. UK Sport, the body that funds Olympic and Paralympic sports, has signed a contract worth more than £300,000 to give thousands of athletes access to an app that detects and hides abusive posts sent by other users on social media. Athletes are able to sign up for free and can protect their accounts throughout the Games cycle up to Los Angeles 2028. The level of abuse our athletes are facing online is unacceptable - to do nothing about this is not an option, UK Sport director of performance Kate Baker said about a deal that is the first of its kind in British sport. The app, called Social Protect, uses AI to try to ensure athletes see as few abusive messages sent their way as possible.

How to create three easy Halloween makeup looks with GlowUp's Axel Axel A D Brown was one of the final three close MUA An abbreviation of makeup artist on series 5 of Glow Up. Their background is in drag, sci-fi and nerd culture. This means Axel's take on using make up to create glamorous creatures and creepy monsters is perfect for this time of year. There's something special about Halloween that allows people to let their guard down and fully express themselves, Axel says. Nobody really cares what you're doing because everyone's crazy and weird looking." BBC Bitesize asked Axel to show us how to create three seriously spooky glow-ups. How to paint a frightening Frankenstein's monster head Start with a clean, moisturised face. Green, white, red and black makeup is all you need to complete this look. Slide 1 of 4, Man with a clean face dressed in a black shirt, Step 1 Start with a clean, moisturised face. Go full-on with the orange face paint and get ready to do some more line work. Use a fine paint brush and if you're worried about wobbly lines, try holding your elbow while you draw as this can steady a shaky hand Don't worry if you don't have any - you can achieve the same look with a green water-based paint, just add clear hairspray to help it stay on all night Axel loves to do makeup looks based on creatures and monsters. "Whenever I'm sketching out ideas, the first thing that comes to me is usually a colour combination that intrigues me, Axel says.

The rapid success of Vision Large Language Models (VLLMs) often depends on the high-resolution images with abundant visual tokens, which hinders training and deployment efficiency. Current training-free visual token compression methods exhibit serious performance degradation in tasks involving high-resolution, text-oriented image understanding and reasoning. In this paper, we propose an efficient visual token compression framework for text-oriented VLLMs in high-resolution scenarios. In particular, we employ a light-weight self-distillation pre-training stage to compress the visual tokens, requiring a limited numbers of image-text pairs and minimal learnable parameters. Afterwards, to mitigate potential performance degradation of token-compressed models, we construct a high-quality post-train stage. To validate the effectiveness of our method, we apply it to an advanced VLLMs, InternVL2. Experimental results show that our approach significantly reduces computational overhead while outperforming the baselines across a range of text-oriented benchmarks. We will release the models and code soon.

Large Language Models (LLMs) demonstrate remarkable capabilities in question answering (QA), but metrics for assessing their reliance on memorization versus retrieval remain underdeveloped. Moreover, while finetuned models are state-of-the-art on closed-domain tasks, general-purpose models like GPT-4o exhibit strong zero-shot performance. This raises questions about the trade-offs between memorization, generalization, and retrieval. In this work, we analyze the extent to which multimodal retrieval-augmented VLMs memorize training data compared to baseline VLMs. Using the WebQA benchmark, we contrast finetuned models with baseline VLMs on multihop retrieval and question answering, examining the impact of finetuning on data memorization. To quantify memorization in end-to-end retrieval and QA systems, we propose several proxy metrics by investigating instances where QA succeeds despite retrieval failing. Our results reveal the extent to which finetuned models rely on memorization. In contrast, retrieval-augmented VLMs have lower memorization scores, at the cost of accuracy (72% vs 52% on WebQA test set). As such, our measures pose a challenge for future work to reconcile memorization and generalization in both Open-Domain QA and joint Retrieval-QA tasks.

The robustness of large language models (LLMs) against knowledge conflicts in unimodal question answering systems has been well studied. However, the effect of conflicts in information sources on vision language models (VLMs) in multimodal settings has not yet been explored. In this work, we propose \segsub, a framework that applies targeted perturbations to image sources to study and improve the robustness of VLMs against three different types of knowledge conflicts, namely parametric, source, and counterfactual conflicts. Contrary to prior findings that showed that LLMs are sensitive to parametric conflicts arising from textual perturbations, we find VLMs are largely robust to image perturbation. On the other hand, VLMs perform poorly on counterfactual examples (<30% accuracy) and fail to reason over source conflicts (<1% accuracy). We also find a link between hallucinations and image context, with GPT-4o prone to hallucination when presented with highly contextualized counterfactual examples. While challenges persist with source conflicts, finetuning models significantly improves reasoning over counterfactual samples. Our findings highlight the need for VLM training methodologies that enhance their reasoning capabilities, particularly in addressing complex knowledge conflicts between multimodal sources.

Lara Croft has been named the most iconic video game character of all time to mark the 20th Bafta Games Awards. It's been 28 years since Tomb Raider introduced gamers to Lara and she's changed quite a bit in that time. "She's gone from being very pointy and childlike in drawing to very filmic," says Shelley Blond, the actress who voiced the original Lara. The character beat the likes of Mario and Sonic for the title in a poll of gamers. Shelley tells BBC Newsbeat she's not surprised, even though she admits she's never played the game herself.

Recently, Large Language Models (LLMs) have been serving as general-purpose interfaces, posing a significant demand for comprehensive visual knowledge. However, it remains unclear how well current LLMs and their visually augmented counterparts (VaLMs) can master visual commonsense knowledge. To investigate this, we propose ImageNetVC, a human-annotated dataset specifically designed for zero- and few-shot visual commonsense evaluation across 1,000 ImageNet categories. Utilizing ImageNetVC, we benchmark the fundamental visual commonsense knowledge of both unimodal LLMs and VaLMs. Furthermore, we analyze the factors affecting the visual commonsense knowledge of large-scale models, providing insights into the development of language models enriched with visual commonsense knowledge. Our code and dataset are available at https://github.com/hemingkx/ImageNetVC.