Feedback is New Scientist's popular sideways look at the latest science and technology news. You can submit items you believe may amuse readers to Feedback by emailing feedback@newscientist.com Recently, Feedback was delighted to peruse the raciest conference invitation we have ever received. We get a lot of conference invites from organisers labouring under the delusion we are doing something akin to science journalism, and they are mostly a little prosaic: what's new in G-protein signalling, more findings about the biology of molluscs, that kind of thing. Here is the opening line: "From its groundbreaking inception in London to its spectacular evolution in the vibrant heart of China, the Love and Sex with Robots Conference is gearing up for its most thrilling chapters yet: its landmark 12th International edition, scheduled for June 2026."

Summarization refinement faces challenges when extending to multi-dimension. In this paper, we introduce ReFeed, a powerful summarization refinement pipeline that enhances multiple dimensions through reflective reasoning on feedback. To achieve this, we release SumFeed-CoT, a large-scale Long-CoT-based dataset optimized for training a lightweight model with reflective reasoning. Our experiments reveal how the number of dimensions, feedback exposure, and reasoning policy influence refinement performance, highlighting reflective reasoning and simultaneously addressing multiple feedback is crucial to mitigate trade-off between dimensions. Furthermore, ReFeed is robust to noisy feedback and feedback order. Lastly, our finding emphasizes that creating data with a proper goal and guideline constitutes a fundamental pillar of effective reasoning. The dataset and model will be released.

Satisfiability (SAT) solvers based on techniques such as conflict driven clause learning (CDCL) have produced excellent performance on both synthetic and real world industrial problems. While these CDCL solvers only operate on a per-problem basis, graph neural network (GNN) based solvers bring new benefits to the field by allowing practitioners to exploit knowledge gained from solved problems to expedite solving of new SAT problems. However, one specific area that is often studied in the context of CDCL solvers, but largely overlooked in GNN solvers, is the relationship between graph theoretic measure of structure in SAT problems and the generalisation ability of GNN solvers. To bridge the gap between structural graph properties (e.g., modularity, self-similarity) and the generalisability (or lack thereof) of GNN based SAT solvers, we present StructureSAT: a curated dataset, along with code to further generate novel examples, containing a diverse set of SAT problems from well known problem domains. Furthermore, we utilise a novel splitting method that focuses on deconstructing the families into more detailed hierarchies based on their structural properties. With the new dataset, we aim to help explain problematic generalisation in existing GNN SAT solvers by exploiting knowledge of structural graph properties. We conclude with multiple future directions that can help researchers in GNN based SAT solving develop more effective and generalisable SAT solvers.

Recent cognitive modeling studies have reported that larger language models (LMs) exhibit a poorer fit to human reading behavior, leading to claims of their cognitive implausibility. In this paper, we revisit this argument through the lens of mechanistic interpretability and argue that prior conclusions were skewed by an exclusive focus on the final layers of LMs. Our analysis reveals that next-word probabilities derived from internal layers of larger LMs align with human sentence processing data as well as, or better than, those from smaller LMs. This alignment holds consistently across behavioral (self-paced reading times, gaze durations, MAZE task processing times) and neurophysiological (N400 brain potentials) measures, challenging earlier mixed results and suggesting that the cognitive plausibility of larger LMs has been underestimated. Furthermore, we first identify an intriguing relationship between LM layers and human measures: earlier layers correspond more closely with fast gaze durations, while later layers better align with relatively slower signals such as N400 potentials and MAZE processing times. Our work opens new avenues for interdisciplinary research at the intersection of mechanistic interpretability and cognitive modeling.

Kristina Cooper, vice president at Travelmation in Florida, is sharing her smart tips and tricks to avoid frustrating flight delays and cancellations when traveling this holiday season. As Americans brace for holiday travel, flyers might be thinking about whether the uptick in drone sightings will affect take-off and landing. Drone sightings across the northeast skies have been reported with some spotted near or over airport spaces. The Transportation Security Administration (TSA) expects nearly 40 million people to fly over the holidays, according to the agency's website. Sightings have been reported near airports in Connecticut, Massachusetts, New Jersey and New York, according to several airports in those states that confirmed the sightings to Fox News Digital.

An Iranian-American citizen and a Swiss Iranian have been arrested and charged by United States authorities with allegedly exporting sensitive technology to Iran that was used in a deadly drone attack on American forces based in Jordan. Islamic Resistance in Iraq, an umbrella group of Iran-backed fighters, was alleged to have carried out the drone attack that killed three US soldiers and wounded 47 others at a US military outpost in Jordan, near the Syrian border, in January. Federal prosecutors in Boston on Monday charged 38-year-old Mohammad Abedininajafabadi, who is known as Mohammad Abedini, the co-founder of an Iranian-based company, and Mahdi Sadeghi, 42, an employee of Massachusetts-based semiconductor manufacturer Analog Devices, with conspiring to violate US export laws. Abedini, a dual citizen of Switzerland and Iran, was arrested in Milan, Italy, at the request of the US government, which will seek his extradition. Sadeghi, an Iranian-born naturalised US citizen, who lives in Natick, Massachusetts, was also arrested.

Former OpenAI executive Mira Murati says it could take decades, but AI systems eventually will perform a wide range of cognitive tasks as well as humans do--a prospective technological milestone widely known as artificial general intelligence, or AGI. "Right now, it feels quite achievable," Murati said at WIRED's The Big Interview event in San Francisco on Tuesday. In her first interview since resigning as OpenAI's chief technology officer in September, Murati told WIRED's Steven Levy that she's not overly concerned about recent chatter in the AI industry that developing more powerful generative AI models is proving challenging. "Current evidence shows that progress will likely continue," Murati said. Whether we need new ideas to get to AGI-level systems, that's uncertain.

Figma cofounder Dylan Field is seemingly a big Enron fan--or rather, of the crypto-fueled semi-parodic relaunch of the company that hit the web earlier this week. Sporting an oversized Enron hoodie during his conversation with WIRED editor at large Steven Levy during The Big Interview event in San Francisco on Tuesday, Field said he's always been a fan of the Enron logo, which was the last one crafted by legendary American graphic designer Paul Rand, of ABC, IBM, UPS, and Westinghouse logo fame. But he said he also "got a real kick" out of the potential Enron relaunch, which has been tied to "Birds Aren't Real" creator Connor Gaydos. As someone who was just 9 years old when Enron imploded in 2001, Field says he wonders (optimistically, it seems) if it's possible to build a new company on the back of the tainted brand, given that his generation might not carry the kind of baggage related to the company's stumbles that others do. Either way, it seems, it's a question of the power of design, something Field and Levy focused on more broadly as their chat went on, talking not just about the creation and evolution of the Figma platform, but also where the cofounder sees the company going in the immediate future.

Surprisal theory posits that the cognitive effort required to comprehend a word is determined by its contextual predictability, quantified as surprisal. Traditionally, surprisal theory treats words as distinct entities, overlooking any potential similarity between them. Giulianelli et al. (2023) address this limitation by introducing information value, a measure of predictability designed to account for similarities between communicative units. Our work leverages Ricotta and Szeidl's (2006) diversity index to extend surprisal into a metric that we term similarity-adjusted surprisal, exposing a mathematical relationship between surprisal and information value. Similarity-adjusted surprisal aligns with information value when considering graded similarities and reduces to standard surprisal when words are treated as distinct. Experimental results with reading time data indicate that similarity-adjusted surprisal adds predictive power beyond standard surprisal for certain datasets, suggesting it serves as a complementary measure of comprehension effort.

Four days after the attempt on his life, the voice of Donald Trump booms from the speakers in Montego Bay, Jamaica: "If they needed an assassin, they should have sent for Bodyguard … about to commit a quadruple murder at Sumfest in Montego Bay." The audience are taken by surprise, having been primed for a reggae riddim to drop, and laugh. The Bodyguard crew have just taken to the stage at Sumfest Global Sound Clash, a musical gladiatorial contest where sound systems battle against one another with creative mixing, hyped-up MCs and exclusive – often incendiary – recordings featuring star guests and in-jokes. AI vocalists such as this fake Trump, however, are sending shockwaves through a decades-old musical tradition in which authenticity and originality are paramount, and sound systems pay premium rates to artists to get vocals for clashes. "AI is going to mash up the industry," says Fabian Anderson, a dub agent who liaises between artists and sound systems to secure those exclusive tracks.