Ignoring the anti-GPL nature of Rust rewrites, it seems that it has been foolish to roll out rust-coreutils wholesale instead of perhaps an option to be selected while installing from the live image, which would almost certainly have lead to enthusiasts testing these utilities; and after enough data had been collected could then have been put in the main branch.

This bug report is old so it should have been fixed by now, but I wanted to commentate on rust-coreutils.

cross-posted from: https://lemmygrad.ml/post/11141812

A report by official Chinese media on Wednesday unveiled two new types of anti-drone laser weapons. A developer said such laser weapon systems, supported by intelligent targeting and multi-sensor integration, demonstrate strong responsiveness and adaptability, while enabling coordinated operations within an integrated combat network.

Footage released in the CCTV News report showed how the systems countered low-altitude, slow-flying small drones simulating covert penetration on the battlefield, flying at an altitude of 50 to 80 meters, which is within the blind spot of conventional radar systems.

More advanced threats such as fiber-optic drones, capable of transmitting real-time data and video through physical links and thus resistant to electronic interference, pose additional challenges and often require physical destruction, according to the CCTV News report.

In response to such threats, multiple counter-drone systems were featured. A container-like platform identified as the “Guangjian-11E,” or Light Arrow-11E, multi-mode terminal jamming system can rapidly lock onto targets and disrupt their onboard sensing and targeting functions, the report said.

The “Guangjian-21A”, or Light Arrow-21A, system, meanwhile, operates in a “shoot-on-the-move” mode, allowing it to maintain sustained firepower while reducing exposure to counterattacks, reported CCTV News.

In terms of engagement methods, the “Guangjian-21A” specializes in “hard-kill” capabilities, generating high-density energy beams over several kilometers to penetrate drone structures and destroy internal circuits or propulsion systems within seconds. By contrast, the “Guangjian-11E” employs a soft-kill approach, using pulsed laser energy to precisely disable key components, effectively blinding drones and disrupting their reconnaissance and data transmission functions, reported CCTV News.

These complementary “soft” and “hard” kill methods address targets at different ranges and threat levels, filling gaps in each other’s operational coverage, CCTV News said.

Both systems are equipped with phased-array radar and infrared detection systems, and can interconnect via both wired and wireless links to enable real-time data sharing and precise target identification and engagement, according to the CCTV News report.

The laser systems, as next-generation equipment, feature multi-source detection and intelligent identification, enabling rapid response and adaptive targeting. Integrated into a broader combat system, the platforms can coordinate “soft” and “hard” kill methods and link with radar and electro-optical sensors to form a rapid “detect-to-destroy” chain, Zhou Shuiliang from the Aviation Industry Corporation of China was quoted by the CCTV as saying.

Previously, multiple types of anti-drone equipment were reviewed in formations at China's V-Day military parade on September 3, 2025 to mark the 80th anniversary of the victory in the Chinese People's War of Resistance Against Japanese Aggression and the World Anti-Fascist War, addressing the importance of anti-drone warfare on the modern battlefield.

Anti-drone missile and artillery systems, high-energy laser weapons, and high-power microwave weapons reviewed at the parade are a powerful "iron triangle" that can both "soft kill" and "hard destroy" unmanned aerial vehicles.

Given the widespread deployment of drones on the battlefield, countries are now prioritizing the development of efficient counter-drone systems and platforms, Wang Yunfei, a Chinese military affairs expert, told the Global Times.

The inclusion of high-energy laser and high-power microwave counter-drone systems in the parade indicates that both have demonstrated strong performance in realistic training, and suggests that China’s laser- and microwave-based counter-drone technologies are at the forefront globally.

Zhang Xuefeng, another Chinese military affairs expert, told the Global Times that high-energy laser weapons can quickly and accurately detect, aim, and track targets. Using high-energy lasers, it only takes a few seconds to shoot a drone down.

"This genre of anti-drone systems has high accuracy, minimal collateral damage, the advantage of unlimited ammunition, and very low operating costs, making it the lowest cost anti-drone equipment in a single deployment," Zhang said.

Hopefully a sign the bubble is bursting

cross-posted from: https://hexbear.net/post/8058924

https://www.reddit.com/r/NewsWithJingjing/comments/1s3ffye/china_unveils_atlas_drone_swarm_operations_system/

TurboQuant looks like a pretty massive deal for running local models efficiently. The core issue they are tackling is the memory bottleneck caused by the key value cache during generation. When you are doing long context inference storing all those high dimensional vectors eats up VRAM extremely fast. Traditional vector quantization helps but usually introduces memory overhead because you have to store scaling factors or constants in full precision for every small block of data. That overhead can easily add an extra bit or two per parameter which ruins the compression targets people are aiming for.

TurboQuant solves the problem by combining two clever mathematical tricks to eliminate that overhead entirely and get the cache down to 3 bits without losing accuracy. The first part is an algorithm called PolarQuant. Instead of looking at the vectors in standard cartesian coordinates it converts them into polar coordinates. This basically separates the magnitude from the direction. Because the angles map onto a fixed predictable circular grid the model no longer needs to store those dynamic bounding boxes or normalization constants that traditional methods require. That step handles the bulk of the compression to capture the main signal of the vector.

The second piece of the puzzle is where they use something called Quantized Johnson Lindenstrauss or QJL to clean up the residual error left over from the first step. QJL uses a mathematical transform to shrink that leftover error down to just a single sign bit of positive or negative one while preserving the relative distances between the data points. This acts as a mathematical error checker that fixes any bias in the attention scores. Because it only uses one bit and preserves the geometry of the space the attention mechanism can still calculate accurate logits without needing full precision data.

They tested this on open weights models like Gemma and Mistral across heavy needle in a haystack and LongBench tasks. They managed to compress the KV cache down to 3 bits with literally zero drop in accuracy and they did not even need to do any fine tuning or calibration. On top of saving a massive amount of VRAM the 4 bit version actually speeds up attention logit computation by up to 8x on H100 GPUs compared to standard 32 bit floats. This seems like a massive leap forward for anyone trying to run long context models on constrained hardware or scale up huge vector search databases.

Naturally I expect it to still be used as propaganda generation since it will likely be government-only tech now.

But also, lol, lmao even. Get fucked. Hope you like the

cross-posted from: https://lemmy.ml/post/44890650

Does the old man think we can't tell this is AI-juiced? Whatever gets the Amazon book bux Powell 😭 Nonetheless, this is extremely cool. Cutting edge iridium semiconductors and other newfangled shit keeps falling off my radar because I mainly care about the energy+obligation web binding humanity together.

In the second week of February 2026, a quiet revolution landed in the pages of Science Advances. A team from Peking University’s School of Electronics, led by researcher Qiu Chenguang in collaboration with academician Peng Lianmao of the Chinese Academy of Sciences, unveiled the world’s smallest ferroelectric field-effect transistor (FeFET). Its physical gate length: just 1 nanometer. Its operating voltage: a mere 0.6 volts. Its memory performance: an on/off current ratio of up to 2 × 10⁶, programming speeds as fast as 1.6 nanoseconds, and switching energy around 0.45 fJ/μm - roughly one-tenth the best previously reported figures.

For the first time, ferroelectric memory has achieved voltage compatibility with state-of-the-art logic transistors (typically ~0.7 V). Data can now flow between memory and compute units at the same low voltage, without charge pumps, voltage converters, or the energy penalties that have plagued attempts to integrate non-volatile memory with logic at scale. The device uses metallic single-walled carbon nanotubes (m-SWCNTs) as the gate electrode, a 2D molybdenum disulphide (MoS₂) channel, a ferroelectric layer of CuInP₂S₆ (CIPS), multilayer graphene, and hexagonal boron nitride (h-BN) in a van der Waals heterostructure. The magic lies in the “nanogate” effect: the ultra-sharp 1 nm tip concentrates the electric field, creating localised intensities high enough to switch the ferroelectric polarisation well below its nominal coercive voltage, while dramatically improving capacitance coupling. Short-channel effects - the bane of conventional scaling - are rendered irrelevant.

This is not another headline-grabbing lab curiosity. It is a material foundation for the next era of computing: ultra-low-power, in-memory, edge-native intelligence that can operate at the scale of angstrom nodes without the voracious energy appetite of today’s hyperscale silicon. And when placed within the broader Chinese technology stack - 2D semiconductors, graphene-enhanced structures, triboelectric and piezoelectric nanogenerators, flexible energy storage - it becomes something far more consequential. It becomes the energetic substrate for what I have called Digital Westphalia: a global digital order grounded in nation-state sovereignty, energetic realism, and systemic abundance rather than rentier extraction and entropic financialisation. Understanding the Breakthrough in Plain Terms

Ferroelectric transistors have long promised non-volatile memory that retains data without power, switches at high speed, and consumes almost no standby energy. Their polarisation states act like tiny, permanent magnets for electrons. The problem has always been scaling and voltage. Traditional FeFETs required gate voltages above 1.5 V to flip the ferroelectric layer reliably, while logic transistors had dropped below 0.7 V. Integrating them meant wasteful voltage step-up circuits and heat. Shrinking the gate below 5 nm triggered short-channel leakage and loss of control.

The Peking team solved both with a radical redesign. By replacing a planar gate with a 1 nm-diameter carbon nanotube, they turned the gate into a nanotip lightning rod for electric fields. Simulations and measurements show field strengths inside the CIPS layer reaching 2.7 × 10⁶ V/cm at just 0.6 V applied - more than five times the material’s coercive field in a conventional geometry. Capacitance coupling efficiency soars because the tiny gate concentrates voltage drop across the ferroelectric rather than wasting it elsewhere. The result is a device that not only scales but thrives at the 1 nm limit, immune to the physics that doom conventional designs.

Reviewers noted the achievement’s elegance: voltage efficiency exceeding 125 % (operating voltage below coercive voltage), retention and endurance suitable for real-world use, and compatibility with existing CMOS processes via 3D heterogeneous integration. The implications for large-model inference, edge intelligence, wearables, and IoT are immediate. A single edge node can now perform sophisticated inference with energy budgets measured in femtojoules rather than picojoules or worse, opening the door to truly battery-free or energy-harvesting devices.

Situating the Advance in a Thermoeconomic Framework

To grasp why this matters beyond the semiconductor roadmap, we need a different lens: Systemic Exchange Value (SEV), or what I shorthand as thermoeconomics. Traditional economics treats value as subjective preference revealed in prices, with GDP as the aggregate scorecard. It abstracts away the energetic and entropic realities that actually sustain life and production. SEV starts from the opposite premise: economies are first and foremost systems of energy transformation. All use-value is ultimately embodied energy - direct and indirect - embedded in material configurations that deliver services over time. Exchange-value is the monetary claim on that embedded energy, modulated by the Energy Return on Energy Invested (EROEI) of production and use.

Three interlocking circuits define the system:

The thermodynamic circuit: real transformation of energy into ordered structures (infrastructure, devices, knowledge), inevitably producing entropy (waste heat, disorder);

The exchange-value circuit: endogenous money and financial claims that allocate claims on future available energy in potential (AEP - available energy in potential); and

The information circuit: data, algorithms, and coordination mechanisms that reduce uncertainty and thereby improve EROEI by minimising wasteful friction.

Productive systems expand systemic abundance when they increase net AEP faster than entropy degrades it. Maladaptive systems - those that channel liquidity into low-EROEI activities (speculative finance, planned obsolescence and rent extraction, for instance) - accelerate entropy, erode adaptive capacity, and eventually face collapse or forced reorganisation.

In information technologies, the dominant metric has been “performance per watt,” but SEV demands a deeper accounting: the total energetic cost of the entire lifecycle, including the embodied energy of fabrication, the operational exergy destruction (waste heat), the systemic coordination overhead, and the long-term adaptive value created. Hyperscale cloud architectures score poorly here. They concentrate enormous computational capacity in a few geographic nodes, requiring massive cooling, redundant power generation, and transcontinental data transmission. Every query to a large language model can consume energy equivalent to a household’s daily use. The EROEIu (use-phase return) looks impressive in narrow benchmarks but collapses when externalities - grid strain, water consumption, geopolitical chokepoints on undersea cables - are internalised.

The nanogate FeFET flips this script. By slashing operating voltage and enabling seamless memory-logic integration, it dramatically reduces exergy destruction at the device level. When scaled into arrays for in-memory computing, it collapses the von Neumann bottleneck, cutting data movement energy by orders of magnitude. Paired with 2D materials that can be fabricated at lower thermal budgets and with graphene or MXene-enhanced nanogenerators that harvest ambient mechanical or thermal energy, entire nodes become energetically autonomous. The information circuit now operates with far higher informational EROEI: more useful computation per joule invested, less entropy exported as heat, and greater resilience because intelligence is distributed rather than centralised.

China’s systematic investment in the full stack - from wafer-scale 2D growth to open-source AI frameworks like DeepSeek, to self-powered IoT ecosystems - is building precisely the high-EROEI infrastructure that SEV identifies as adaptive. It is creating durable use-value that compounds over decades rather than depreciating in quarters. In thermoeconomic terms, it is expanding the envelope of available energy in potential for the entire digital sphere. Digital Westphalia as the Political Expression of Thermoeconomic Realism

Digital Westphalia names the possible emerging global digital order that aligns political-information sovereignty with these energetic realities. Just as the 1648 Peace of Westphalia ended the Thirty Years’ War by enshrining territorial sovereignty and non-interference, today’s digital analogue reasserts nation-state primacy over data regimes, technical standards and infrastructure governance - while preserving interoperability through open protocols.

The old model was a de facto American imperium: hardware, software, standards and data flows routed through U.S.-controlled chokepoints (Northern Virginia data centers handling ~70 % of global traffic, SWIFT, undersea cables). This delivered rents to a handful of platforms and intelligence leverage to one state, but at the cost of universal vulnerability and entropic inefficiency. Sanctions, extraterritorial export controls, and deplatforming demonstrated the fragility. Check out Newman and Farrell’s Underground Empire for a detailed discussion of these realities.

Digital Westphalia offers an alternative: sovereign digital territories that can choose their own data localisation, governance, and ecosystem providers, yet interconnect via open-source standards (RISC-V, Linux contributions from Huawei, HarmonyOS adaptability). The nanogate breakthrough, embedded in a Chinese stack that emphasises modularity, open architectures, and energy autonomy, supplies the material base. Nations or regions can now deploy federated networks of edge intelligence without building hyperscale data centers or begging for foreign chips under export-control threat. A developing country can equip rural health posts with self-powered wearable monitors and localised diagnostic models that run inference on-device. A mid-sized power can maintain sovereign AI capabilities for agriculture, disaster response, or industrial optimisation without ceding data sovereignty or energy security.

The savings are thermoeconomic as much as fiscal. Less need for continent-spanning transmission infrastructure. Lower grid pressure. Reduced geopolitical risk premiums on energy imports for compute. Higher systemic EROEI because intelligence is co-located with the phenomena it observes and acts upon. In SEV terms, this is liquidity allocated to high-adaptive-capacity uses rather than siphoned into fictitious capital or low-EROEI consumption.

Strategic Implications in a Multipolar World

For much of the world, the choice is sharpening. The world can double down on a U.S.-centric stack that promises cutting-edge performance but delivers vendor lock-in, energy intensity, and exposure to export-control volatility. Or we can engage the emerging open, sovereign-capable ecosystem that lowers the barrier for genuine digital autonomy. The latter does not require “choosing sides”; it requires recognising that energetic and informational realism now favours distributed, interoperable sovereignty over centralised techno-feudalism.

The Peking University nanogate is one device. But it exemplifies a broader pattern: China’s willingness to invest in the thermodynamic foundations of the information age while others financialise their way toward entropy. The full stack - 2D materials scaling, nanogenerators, open-source models, distributed ledger coordination - is creating the possibility of Digital Westphalia at planetary scale. Nations that seize it will expand their adaptive capacity; those that cling to the old imperium risk locking themselves into maladaptive rigidity.

This is the real significance of the 1 nm breakthrough. It is not merely smaller and lower-power. It is a material refutation of the assumption that computational abundance must come at the price of energetic profligacy and political subordination. In thermoeconomic terms, it augments available energy in potential. In geopolitical terms, it makes Digital Westphalia not utopian but all but inevitable.

The question for policymakers, strategists and citizens is no longer whether a new digital order is coming. It is whether we will shape its emergence in alignment with energetic realism and sovereign dignity or allow entropic forces to dictate the terms.

https://x.com/CNSpaceflight/status/1993158707056984359

Posted this in a comment but it's a must read for all.

I thought the military used LLMs for war but I was wrong. This substack talks about a piece of tech called Maven, used since Iraq, and goes into the ethics and validity of "shortening the kill chain"