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:2607.08839v1 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:2607.08867v1 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:2607.08896v1 Announce Type: new Abstract: We describe our entry to the ICIP 2026 Grand Challenge on Extreme In-the-Wild License Plate Super-Resolution (XLPSR), which scored 9.73 wECR on the public validation leaderboard. The system pairs a Hybrid Attention Transformer super-resolution (HAT) front-end with an ensemble of two scene-text recognisers (PARSeq-S and CLIP4STR-B) and a confidence-weighted character-voting scheme that abstains on uncertain positions. We treat XLPSR as a recognition task gated by image legibility: the SR step exists to lift characters out of sub-pixel territory, and the asymmetric scoring rule (+2 / -1 / 0) is exploited explicitly through abstention. Our pipeline runs in 1.7 s per sequence on RTX 3090 (max 2.7 s, p99 2.4 s), well under the 60 s/sequence Docker budget.

arXiv:2607.08945v1 Announce Type: new Abstract: Accurate cocoa mapping is increasingly important for deforestation monitoring, supply-chain transparency, and regulatory applications. Spatial aggregation in conventional medium-resolution Earth observation (EO) imagery may limit cocoa detection in heterogeneous smallholder landscapes. In Cote d'Ivoire, we therefore evaluated how mapping performance varies across landscape conditions, whether very high resolution (VHR) imagery provides a meaningful advantage, and whether foundation-model embeddings improve decametric cocoa mapping. We developed models using 0.5 m Pleiades VHR imagery, a 10 m Sentinel-2 annual composite, and embeddings from TESSERA and AlphaEarth Foundations (AEF), and additionally assessed four publicly available cocoa mapping products. Performance was evaluated through a landscape-stratified accuracy assessment using 2,821 independently interpreted reference points distributed across gradients of tree cover density and landscape fragmentation. The VHR model achieved the highest performance (F1 = 0.92) and maintained F1-scores above 0.90 across all strata. Among the decametric inputs, TESSERA performed best (F1 = 0.86), followed by AEF (F1 = 0.82) and Sentinel-2 (F1 = 0.76). Of the existing cocoa products, the Kalischek product performed best (F1 = 0.83), comparable to the internally trained AEF model. Performance differences between VHR and decametric approaches increased with fragmentation and under low and high tree cover density conditions. Targeted VHR acquisition may therefore be particularly beneficial in complex cocoa landscapes, while foundation-model embeddings offer a scalable alternative for large-area mapping.

arXiv:2607.08970v1 Announce Type: new Abstract: Recent benchmarks for VLMs largely assess single- or limited-view perception, leaving untested the core cognitive ability to integrate observations across viewpoints into a coherent, world-centric (allocentric) 3D mental model. We introduce MultiView-Bench, a diagnostic benchmark expressly designed to evaluate multi-view integration for holistic 3D scene comprehension. Unlike existing datasets that focus on pixel-level mapping or camera-relative navigation, MultiView-Bench requires models to decouple object positioning from transient perspectives and ground them in a fixed global coordinate system. This capability serves as a prerequisite for VLMs before being deployed for downstream tasks such as mechanical part assembly. Our systematic evaluation of frontier VLMs reveals consistent failure modes: strong performance on 2D planar relations from a single image, but marked difficulty with 3D spatial relations and with aggregating information across views. We further identify biases in VLMs, such as struggles with unconventional axis directions and sensitivity to object colorways and texture variations. Acknowledging these limitations, we propose ViewNavigator, a multi-agent framework that actively selects informative viewpoints, perceives, and fuses multi-view evidence, improving diverse base models on MultiView-Bench even under a strict budget-matched comparison (and by 3-5x for the full agent).

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.

Naomi Rea / Financial Times: How museums are using AI chatbots to reach new audiences, as some museum staff worry AI-generated inaccuracies and bias could undermine them as trusted sources  —  Touted as a way to engage visitors and boost funding, new tools are triggering concerns around trust and ethics

James Norton as Ormund Hightower in "House of the Dragon" season three. Theo Whiteman/HBO Lord Ormund Hightower becomes a key player in "House of the Dragon" season three. Portrayed by James Norton, the character is Alicent's cousin and commander of the Hightower army. Here's a rundown of what happens to Ormund in the book, "Fire & Blood." Warning: Spoilers ahead for "House of the Dragon" season three, episode four, and the book "Fire & Blood." There are plenty of Targaryens vying for power in "House of the Dragon," but the silver-haired dragonlords are not the only threats in Westeros. HBO's tentpole "Game of Thrones" prequel depicts the Dance of the Dragons, which finds Rhaenyra Targaryen (Emma D'Arcy) pitted against her half-brother, Aegon II Targaryen (Tom Glynn-Carney), in a brutal war of succession. Aegon II — along with his three younger siblings, Aemond (Ewan Mitchell), Helaena (Phia Saban), and Daeron (Benjamin Evan Ainsworth) — was born to the late King Viserys and Alicent Hightower. Naturally, his claim to the Iron Throne is supported by House Hightower, a noble family from the Reach with a strong connection to the Faith of the Seven. "House of the Dragon" has already featured several prominent Hightowers, including Alicent (Olivia Cooke), her father Otto (Rhys Ifans), and her brother Gwayne (Freddie Fox). In season three, Alicent's cousin Ormund (James Norton) is introduced as another key player in the ongoing schism. Ormund is the head of House Hightower and commands their impressive army. In season three, episode three, "Rhaenyra Triumphant," Ormund is ordered — under threat of three massive dragons — to surrender and bend the knee to the realm's new queen. He does, albeit deceptively. Ormund and the Hightower army — strengthened by his ward, Daeron Targaryen, and the prince's dragon, Tessarion — go on to seize control of Tumbleton, a market town near King's Landing. From his unassuming perch, Ormund continues devising schemes to overthrow Rhaenyra and cement his family's influence. Ormund commands the Hightower army in "House of the Dragon." Theo Whiteman/HBO In the fourth episode of season three, viewers learn even more about Ormund as Rhaenyra scrounges for dirt on her enemy. According to Alicent, Ormund "sees himself as a scholar" who "despises the ignorant and uncouth." He studies histories, collects tapestries, and has a strong "sensitivity to odors." According to Gwayne, he's a "stiff-necked windbag" who always has a "canny plot" up his sleeve. During his hostile stay at Tumbleton, Ormund is revealed to have a devious military mind, a vicious temper, and, indeed, a secret plot: to crown the teenage Daeron as the next king of Westeros, leap-frogging both of his older brothers. Ormund raised Daeron after the prince was sent to live in Oldtown as an infant. He believes Daeron is more of a Hightower than a Targaryen — and that he's more reliable than any of his siblings, who were all raised in King's Landing. "You are a good boy. You speak kindly, and you say your prayers. I have raised you in the light of the Seven, and the Father smiles upon you," Ormund tells Daeron. "But there is a taint in your blood. The Targaryens are a savage race, poor in intellect, but rich in cunning. With dark spells, they created abominations to subdue what was rightfully ours. We are the superior men." "The gods have put you to divine purpose, my boy," he adds. "You will restore our ancient order." Ormund is poised to throw a wrench in Rhaenyra's reign, especially if his story is faithful to the source material, George R. R. Martin's "Fire & Blood." Keep reading to find out Ormund's fate in the book. Ormund's takeover of Tumbleton proves consequential and deadly James Norton as Ormund Hightower in "House of the Dragon." Kevin Baker/HBO Even as Rhaenyra claims the Iron Throne in the book, Aegon II and Aemond (with his massive dragon Vhagar) remain at large, while the Hightowers continue to resist the queen in the Reach. As Martin writes, "No war can be counted as won whilst foes remain unconquered." Prince Daeron the Daring, as Ormund styles him, becomes known as the greatest threat to Rhaenyra's reign, advancing on King's Landing with an army of 20,000 men. Lord Corlys Velaryon, the Hand of the Queen, urges Rhaenyra to pardon the great houses that still oppose her — including Hightower, Lannister, and Baratheon — if they agree to swear fealty to her instead. He also encourages her to send her half-brothers to the Wall to live out their days as Sworn Brothers of the Night's Watch. Rhaenyra's husband, Prince Daemon, argues the opposite. He says that pardoning traitors and oathbreakers will only encourage future rebellions. Rhaenyra tends to agree with Daemon, though she decides to "steer a middle course." She agrees to offer pardons to Houses Baratheon and Lannister, but only after Aegon's line has been vanquished. Rhaenyra sends Daemon to find and kill Aemond, and two of her newly recruited dragonriders, Ulf White and Hugh Hammer, to kill Daeron. In what came to be known as the Battle of Tumbleton, about 6,000 of Rhaenyra's loyalists — including the Northern soldiers known as the Winter Wolves — sack the town. Although Rhaenyra's men are greatly outnumbered by the Hightower's host, they assume that Ulf and Hugh are arriving soon from King's Landing to assist with their dragons, Silverwing and Vermithor. Ormund is killed in the chaos that ensues. Legend says he was slain by Lord Roderick Dustin, aka Roddy the Ruin, after the Northerner cut through hundreds of foes to find him. Under normal circumstances, Ormund's death may have turned the tide in Rhaenyra's favor. Without their commander, the Hightower's host is scattered, their banners toppled. Instead, Ulf and Hugh ignore Rhaenyra's orders and wreak havoc with dragonflame. The book's fictional historians disagree as to why Ulf and Hugh suddenly decide to burn civilians and soldiers alike — whether it's fear of facing Daeron (unlikely, as both Silverwing and Vermithor are older and bigger than Tessarion), fear of facing Ormund's large army (unlikely for similar reasons), disenchantment with Rhaenyra (fairly likely), or straight-up greed (most likely). It seems Ulf and Hugh realized they could seize power for themselves, and chose to prove it with fire and blood. "As neither man could read nor write, we shall never know what drove the Two Betrayers (as history has named them) to do what they did," the book says, though whatever the reason for Ulf and Hugh's treason, the result is the same: "Tumbleton, that prosperous market town, was reduced to ash and embers." Read the original article on Business Insider

Financial Times: A profile of Asha Sharma, an outsider elevated by Satya Nadella to lead Xbox, where she made sweeping job cuts and shed game studios within her first few months  —  Elevated to fix Microsoft's struggling gaming division, she is making tough decisions about its future

In this article, we cover an introduction to the latest NVIDIA Nemotron 3 Nano Omni model and create a simple chat application using the NVIDIA API. The post Introduction to NVIDIA Nemotron 3 Nano Omni appeared first on DebuggerCafe.

Bloomberg: Memo: Tang Jie, founder of Z.ai, the Chinese lab behind the GLM models, argues frontier AI capabilities should stay “as open and widely accessible as possible”  —  The founder of Chinese AI lab Zhipu argued that frontier artificial intelligence should remain broadly accessible rather …

A toy dynamical model of whether the AI workforce that builds future AI ends up cooperative or uncooperative: where the basin boundary lies, what current evidence says about which side we are on, and what would tell us we are on the good path.

<p><strong><a href="https://handbook.gitlab.com/handbook/people-group/directly-responsible-individuals/">Directly Responsible Individuals (DRI)</a></strong></p> I went looking for a definition of "Directly Responsible Individuals" and the best I found was in the GitLab handbook. Apparently the term originated at Apple, where it's used to describe the person who is "ultimately accountable for the success or failure of a specific project, initiative, or activity".</p> <p>I've been thinking about this term recently in the context of LLM-powered agents and how they fit into human organizations. I don't think an agent should <em>ever</em> be considered the DRI for a project - that's something that feels uniquely human to me, because humans can take accountability for their actions where machines cannot.</p> <p>(See also <a href="https://simonwillison.net/2025/Feb/3/a-computer-can-never-be-held-accountable/">IBM's legendary 1979 training slide</a> that states "A computer can never be held accountable, therefore a computer must never make a management decision.") <p>Tags: <a href="https://simonwillison.net/tags/apple">apple</a>, <a href="https://simonwillison.net/tags/management">management</a>, <a href="https://simonwillison.net/tags/ai">ai</a>, <a href="https://simonwillison.net/tags/gitlab">gitlab</a>, <a href="https://simonwillison.net/tags/generative-ai">generative-ai</a>, <a href="https://simonwillison.net/tags/llms">llms</a>, <a href="https://simonwillison.net/tags/ai-ethics">ai-ethics</a>, <a href="https://simonwillison.net/tags/coding-agents">coding-agents</a></p>

<p><strong>Release:</strong> <a href="https://github.com/simonw/shot-scraper/releases/tag/1.11">shot-scraper 1.11</a></p> <p>Some minor improvements, mainly around command option consistency and making the <code>server:</code> mechanism <a href="https://shot-scraper.datasette.io/en/stable/multi.html#running-a-server-for-the-duration-of-the-session">used by</a> both <code>shot-scraper video</code> and <code>shot-scraper multi</code> work if the server takes longer than a second to start serving traffic.</p> <blockquote> <ul> <li><code>server:</code> processes used by <code>shot-scraper multi</code> and <code>shot-scraper video</code> now wait up to 30 seconds for the target URL to accept connections, polling for port availability and replacing the previous fixed one-second delay. <a href="https://github.com/simonw/shot-scraper/issues/197">#197</a></li> <li>The <code>shot-scraper</code>, <code>pdf</code>, <code>html</code>, <code>accessibility</code> and <code>har</code> commands now have a <code>--js-file</code> option for loading JavaScript from a local file, standard input or <code>gh:username/script</code>, as an alternative to <code>--javascript</code>which accepts the string of JavaScript directly as an argument. <a href="https://github.com/simonw/shot-scraper/issues/192">#192</a></li> <li><code>shot-scraper multi</code> supports the equivalent <code>js_file:</code> YAML key. </li> <li>The <code>shot-scraper javascript</code> and <code>shot-scraper html</code> commands now have a <code>--timeout</code> option for consistency with other commands. <a href="https://github.com/simonw/shot-scraper/issues/118">#118</a></li> </ul> </blockquote> <p>Tags: <a href="https://simonwillison.net/tags/shot-scraper">shot-scraper</a></p>

Koyo Murata left his small hometown for Tokyo before returning a few years later to transform an abandoned community center into a guest house. Provided by Koyo Murata Koyo Murata, now 25, grew up in a small countryside town in Japan. He moved to Tokyo to study electrical engineering and spent three years working in the city. Last year, he returned to his hometown to transform an abandoned community center into a modern guest house. This as-told-to essay is based on a conversation with Koyo Murata, 25, the founder and operator of the Villa ASO guest house in Taiki, Mie Prefecture, Japan. His words have been edited for length and clarity. I grew up in a quintessential countryside town in Japan with about 8,000 people. My elementary and middle schools had only about 100 students combined. There are even fewer children there now. After school, we'd play in the river or organize soccer matches in the park. I felt lucky to be surrounded by rivers, mountains, and the sea. Looking back, I realize my hometown, Taiki, was already part of a broader trend in Japan. Rural towns like it are disappearing as people move to Tokyo. The capital is where you find the best jobs. Most kids who grow up in small towns like mine dream of making it to the big city. I got to live that dream I moved away at 19 to study electrical engineering at Mie University; after graduating, I spent three years in Tokyo working in construction management and later in independent sales. My life was full, but I didn't like how crowded the city could be. The greater Tokyo metropolitan area consists of nearly 37 million people. The trains were always packed, and people seemed more uptight than the friendly folks back home. Over time, I began to realize how valuable life in the countryside really was. Murata received a call from his dad, telling him the community center in his hometown was going to be torn down. Provided by Koyo Murata That's why, in August 2025, I decided to move back home. In Japan, we call it a "U-Turn" — when someone leaves for the city and later returns to their hometown. My parents and older brother now live in Tokyo and run their own businesses. I wanted to challenge myself by building a career of my own. At the same time, I wanted to find a way to share the beauty of the Mie countryside with the rest of the world. The opportunity came when my parents called me one day to tell me that an abandoned community center in our town was about to be knocked down. Murata hired a construction company to complete 70% of the work, but finished the rest with some friends. Provided by Koyo Murata My father said, "This could be your chance to do something with it." That's when I made my U-Turn and moved back to Taiki to start my own business and find a new use for the old building. I began a project to transform the former community center into Villa ASO, a guest house for visitors to the area. Reviving the old The former community center was a gathering place for local residents and also served as an evacuation assembly point. But it had been sitting unused for 10 years. When I started my company, NexusTown, my idea was to renovate the building and turn it into a guest house that could accommodate groups of up to 8 people. I hope to bring more tourism to our shrinking town. I borrowed about 10 million yen, or $62,000, from my brother's company and received another 10 million through a government-backed startup loan from the Japan Finance Corporation. Villa Aso, a private villa, opened in December 2025. Provided by Koyo Murata Over three months, we hired a construction company to complete 70% of the work, but the remainder was DIY — just me and some friends. It was hard work; none of us had any experience in renovation. I also had to teach myself how to market the guest house, how to put it on Airbnb and Booking.com, and how to set the right prices. In December, when it officially opened, and I saw our first guests enjoy their stay, I was glad I had done it. The house can accommodate groups of up to 8 people. Provided by Koyo Murata Making it a success In January, we were operating at 30%, in February 60%, and by March we'd reached 100%. Our guests have included college students on spring break, multigenerational families, and working adults from all over the country. So far, we've had only one group of inbound tourists from abroad, but I hope we can host more overseas visitors in the future. The house is close to the Kumano Kodo, an ancient pilgrimage route that attracts hikers and travelers from around the world. The villa has a private sauna, and there's a natural hot spring 3 minutes away on foot. Provided by Koyo Murata People who come to Villa ASO get to enjoy the same mountains, river, and sea that I did when I was growing up. They can slow down and enjoy the stars at night. They can meet the warm people of Mie. In addition to the villa's private sauna, there's also a natural onsen, or hot spring, three minutes away by foot that I always recommend and which guests can visit for free. The villa costs on average 28,900 to 33,700 Japanese yen, or $179 to $208, for two guests, with 5,000 yen added for additional party members. The living room and kitchen are one big open space. Provided by Koyo Murata In the future, I'm hoping to connect the villa experience with local restaurants and activities like seated meditation and harbor fishing. My goal is to revitalize the town's economy, and my work is never done. Along with managing the guest house, I also work for a talent-matching agency. I'm always thinking of new ways to renovate and improve the villa. I hope to recoup the investment in about three to four years. Murata is teaching himself how to market the guest house and how to set the right prices. Provided by Koyo Murata The future of small-town Japan The declining population of Taiki worries me. I don't think the town will ever vanish entirely, but it may eventually be merged with another township. It makes me sad to think that could happen to the place where I was born. If this project can help even one or two more people fall in love with my hometown, I'll be happy. Read the original article on Business Insider