These genetic libraries plan for worse-case scenarios. An employee at the Leibniz Institute of Plant Genetics and Crop Plant Research in Germany shows off a specimen of frozen plant seeds from the institute's genebank. Breakthroughs, discoveries, and DIY tips sent every weekday. Amid the 872-day siege of Leningrad in the early 1940s, nine people died protecting a library. This library was not for books, but for seeds collected from around the globe.

A drone attack on the town of Kalogi, in Sudan's South Kordofan region, is said to have hit a kindergarten and killed at least 50 people, including 33 children. The Rapid Support Forces (RSF), the paramilitary group battling the army in Sudan's civil war, was accused of Thursday's attack by a medical organisation, the Sudan Doctors' Network, and the army. There was no immediate comment from the RSF. The RSF in turn accused the army of hitting a market on Friday in a drone attack in the Darfur region, on a fuel depot at the Adre border crossing with Chad. Sudan has been ravaged by war since April 2023 when a power struggle broke out between the RSF and the army, who were formerly allies .

Explosions have been heard near the Sudanese capital of Khartoum, a day after the paramilitary Rapid Support Forces (RSF) said it would agree to a humanitarian ceasefire. Residents in Khartoum, which is controlled by the army, told the AFP news agency that they were woken overnight by the sound of drones and explosions. The blasts appeared to take place near a military base and a power station in the early hours of Friday morning, the residents said. The RSF has not addressed these accounts, but Sudan's military-led government said it would be wary of agreeing to a truce as the group did not respect ceasefires. The two sides have been embroiled in a civil war that has killed at least 150,000 people and forced 12 million others from their homes since it erupted in April 2023.

It was a Monday night in Myanmar's Chang U township in the central Sagaing region, where nearly 100 people had gathered to mark Thadingyut, the festival of the full moon. Some held candles at the event, which doubled as both a celebration and a protest against the military, which seized power in 2021, plunging the country into a bloody civil war. But the celebration soon turned into horror as a motorised paraglider - known locally as a paramotor - flew overhead and dropped bombs onto the crowd. The attack lasted just seven minutes, but at least 26 people died as a result and dozens more were injured. Initially, I thought the lower part of my body had been severed, one 30-year-old who was at the gathering told news agency Reuters.

The RSF is trying to show that they don't need to reach Port Sudan by land in order to be able to have an impact there,

Personalized alignments for individual users have been a long-standing goal in large language models (LLMs). We introduce Drift, a novel framework that personalizes LLMs at decoding time with implicit user preferences. Traditional Reinforcement Learning from Human Feedback (RLHF) requires thousands of annotated examples and expensive gradient updates. In contrast, Drift personalizes LLMs in a training-free manner, using only a few dozen examples to steer a frozen model through efficient preference modeling. Our approach models user preferences as a composition of predefined, interpretable attributes and aligns them at decoding time to enable personalized generation. Experiments on both a synthetic persona dataset (Perspective) and a real human-annotated dataset (PRISM) demonstrate that Drift significantly outperforms RLHF baselines while using only 50-100 examples. Our results and analysis show that Drift is both computationally efficient and interpretable.

There's a moment in the 1967 film The Graduate that has become renowned. At a party thrown by his parents to celebrate his graduation, Benjamin (Dustin Hoffman) is approached by Mr McGuire, an elderly bore who wants to say "just one word" to him: "plastics". "Exactly how do you mean?", asks the hapless Ben. "There's a great future in plastics," says McGuire. "Think about it." Listening last week to the spending plans of the techlords who run Microsoft, Alphabet, Amazon and Meta leads one to wonder if something analogous might have happened to them on their graduation nights. Except that in their cases, the magic word would have been "AI".

Current LLM structured pruning methods typically involve two steps: (1) compression with calibration data and (2) costly continued pretraining on billions of tokens to recover lost performance. This second step is necessary as the first significantly impacts model accuracy. Prior research suggests pretrained Transformer weights aren't inherently low-rank, unlike their activations, which may explain this drop. Based on this observation, we propose Lillama, a compression method that locally distills activations with low-rank weights. Using SVD for initialization and a joint loss combining teacher and student activations, we accelerate convergence and reduce memory use with local gradient updates. Lillama compresses Mixtral-8x7B within minutes on a single A100 GPU, removing 10 billion parameters while retaining over 95% of its original performance. Phi-2 3B can be compressed by 40% with just 13 million calibration tokens, resulting in a small model that competes with recent models of similar size. The method generalizes well to non-transformer architectures, compressing Mamba-3B by 20% while maintaining 99% performance.

The contest was scheduled for two days, although rain meant that it ran into three. Five thousand people attended the St George's Club for the first day's play, and over the course of the match, an estimated 100,000 (approximately 4.2m in today's money) was gambled on the outcome. The Canadians, somewhat exhausted and bedraggled by a journey that had brought them up the St Lawrence River and across Lake Ontario by boat before catching trains to New York, batted first and made 82 against an US attack that consisted solely of two men born in Yorkshire: Sam Wright and Harry Groom. Canada's star was David Winckworth, who made 12 with the bat before sending down a few round-arm thunderbolts of his own, taking four wickets as the Americans were dismissed for 64. Winckworth again top-scored with 14 as Canada stacked up another 63, setting the USA 82 to win.

A Voice of America journalist Wenhao, who specialises in Chinese online disinformation, posted on X that the "biggest US related news on China's internet for the past few days is Texas governor declaring war with the federal government, which did not happen in reality".