<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0">
  <channel>
    <title>The Whole Day</title>
    <link>https://thewhole.day/</link>
    <language>en</language>
    <description>One page of science going right. One paragraph per story. Nothing grim, ever.</description>
    <item>
      <title>Solana Schedules Alpenglow Upgrade for 150-Millisecond Finality</title>
      <link>https://newscord.org/article/solana-sets-oct-21-for-alpenglow-upgrade-targeting-150-millisecond-finality--Story_20261014_SolanasetsOct21forAl6b3100ee</link>
      <guid isPermaLink="false">twd-s00</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>Solana developers have scheduled the network's landmark Alpenglow consensus upgrade for October 21 at epoch 1064, targeting an irreversible transaction finality time of approximately 150 milliseconds. Replacing the legacy TowerBFT voting protocol, Alpenglow moves validator voting off-chain, allowing nodes to exchange votes directly and compile compressed authorization certificates into block headers. The architecture slashes confirmation times from nearly 13 seconds down to sub-quarter-second speed, providing point-of-sale merchants and cross-chain liquid bridging protocols with instantaneous transaction finality while dramatically reducing on-chain message bloat.</description>
    </item>
    <item>
      <title>Anthropic Tests Autonomous AI Replication Across Everyday Hardware</title>
      <link>https://www.anthropic.com/research/self-replicating-ai</link>
      <guid isPermaLink="false">twd-s01</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>Researchers at Anthropic demonstrated that lightweight AI models with as few as 9 billion parameters can independently set up inference servers across diverse commercial hardware, including single-board computers and edge accelerators. Testing automated replication routines without datacenter connectivity, models like Qwen3.5-9B configured working runtime environments across networked hardware nodes, while larger reasoning models achieved faster token generation speeds during propagation. The experiment provides empirical benchmarks showing that decentralized, hardware-agnostic AI replication is approaching feasibility on consumer-grade devices, prompting new security recommendations for edge hardware manufacturers.</description>
    </item>
    <item>
      <title>Ethereum Foundation Launches zkAPI Mainnet for Private AI Interactions</title>
      <link>https://decrypt.co/379991/ethereum-now-pay-for-ai-without-revealing-identity</link>
      <guid isPermaLink="false">twd-s02</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>The Ethereum Foundation and the Open Anonymity Project have deployed zkAPI on Ethereum mainnet, allowing users to pay for artificial intelligence queries without exposing their identity or search content. Users prepay in USDC or Ether to mint short-lived, zero-knowledge cryptographic keys that prove payment validity without linking the user's wallet address to their prompt history. Designed by Ethereum researchers Davide Crapis and Vitalik Buterin, zkAPI separates the payment settlement layer from API access, preventing AI hosts and payment processors from building behavioral profiles on individual users.</description>
    </item>
    <item>
      <title>Artificial Neural Networks Discover Self-Organizing Learning Rules</title>
      <link>https://news.mit.edu/2026/how-artificial-neural-networks-build-their-own-learning-machinery-1013</link>
      <guid isPermaLink="false">twd-s03</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>Researchers at MIT's McGovern Institute revealed that artificial neural networks trained without prescribed backpropagation rules can spontaneously synthesize their own local learning mechanisms. Reporting in the Proceedings of the National Academy of Sciences, scientists observed that when networks adjust connection weights using internal feedback signals, they organize into modular loops strikingly similar to biological synaptic plasticity. The discovery demonstrates that functional learning architectures emerge naturally from general computational constraints, providing computer scientists and neuroscientists with a new model for understanding how biological brains reorganize during learning.</description>
    </item>
    <item>
      <title>Singapore and Google DeepMind Build On-Device Scam Detection Model</title>
      <link>https://deepmind.google/blog/on-device-ai-scam-detection-mddi/</link>
      <guid isPermaLink="false">twd-s04</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>Google DeepMind and Singapore's Government Technology Agency co-developed a fine-tuned, on-device AI model designed to detect complex social engineering and impersonation scams in messaging applications. Trained in an isolated sovereign sandbox using anonymized reporting data, the model runs directly on user hardware to identify manipulative conversation patterns without compromising end-to-end encryption or sending private messages to cloud servers. Presented at Singapore International Cyber Week, the initiative provides a blueprint for privacy-preserving digital defense against financial fraud that has cost Singaporeans over S$4 billion since 2019.</description>
    </item>
    <item>
      <title>Ethereum's Sepolia Testnet Activates Glamsterdam Upgrade with 200M Gas Limit</title>
      <link>https://decrypt.co/380538/ethereum-test-raises-block-limit-3x-current-capacity</link>
      <guid isPermaLink="false">twd-s05</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>Ethereum's primary test network, Sepolia, activated the Glamsterdam upgrade at epoch 353,024, expanding its block gas limit to nearly 200 million—more than three times the 60 million gas cap enforced on Ethereum mainnet. The upgrade allows developers and node operators to evaluate client behavior under high computational load, with testnet blocks processing up to 85.9 million gas per block without performance degradation. Glamsterdam lays the infrastructure for upcoming scaling and zero-knowledge verification improvements on Ethereum, giving developers a live environment to test high-throughput smart contract execution before mainnet deployment.</description>
    </item>
    <item>
      <title>Google DeepMind Launches Open Multimodal Model EmbeddingGemma 2</title>
      <link>https://developers.googleblog.com/google-ai-edge-with-embeddinggemma-2/</link>
      <guid isPermaLink="false">twd-s06</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>Google DeepMind introduced EmbeddingGemma 2, an open-weight 740-million-parameter model that unifies text, code, images, video frames, and audio into a single 768-dimensional vector space. Built on the Gemma 4 architecture under an Apache 2.0 license, the model enables direct cross-modal retrieval—such as locating a specific scene in a video using a typed text query or audio snippet—without relying on separate intermediate transcription or captioning models. With modular encoders designed to operate on consumer hardware in as little as 191 megabytes of memory, the system brings private, zero-shot semantic search directly to mobile devices.</description>
    </item>
    <item>
      <title>Aptos Deploys Encrypted Mempool on Mainnet to Eliminate Frontrunning</title>
      <link>https://ct.com/news/aptos-rolls-out-encrypted-mempool-to-curb-frontrunning-sandwich-attacks</link>
      <guid isPermaLink="false">twd-s07</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>Layer-1 blockchain Aptos has launched an optional Encrypted Mempool feature on its mainnet to protect users against frontrunning and sandwich attacks. Built for institutional traders, AI agents, and DeFi applications, the system encrypts transaction details—such as asset amounts and target smart contracts—directly on the user's device. While the sender address and gas fee remain visible, the transaction contents stay hidden inside the mempool queue while validators reach consensus on block ordering. Once order is locked, validators collectively decrypt and execute the transaction without any single party having viewed its contents in advance, adding only 20 milliseconds of latency.</description>
    </item>
    <item>
      <title>Anthropic Releases Free Vulnerability Scanner for Open-Source Software</title>
      <link>https://www.anthropic.com/research/launching-opt-in-vuln-finding-service-for-open-source</link>
      <guid isPermaLink="false">twd-s08</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>Anthropic launched OSS Scanner, a free service that deploys frontier language models to perform continuous vulnerability scans across open-source codebases. Informed by internal tests that identified more than 29,000 potential software flaws—including thousands of direct patch requests from maintainers—the service provides participating developers with automated analysis and suggested fixes before security flaws can be exploited. Academic benchmarks show that modern AI systems now identify over 85% of known software vulnerabilities on standardized code-auditing suites, transforming AI-assisted defensive engineering for critical software.</description>
    </item>
    <item>
      <title>Starknet Weighs Layer-1 Transition to Build First Quantum-Safe Network</title>
      <link>https://decrypt.co/380585/ethereum-l2-starknet-jumps-20-after-saying-it-wants-to-become-an-l1</link>
      <guid isPermaLink="false">twd-s09</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>Zero-knowledge scaling network Starknet announced it is actively evaluating an architectural transition from an Ethereum Layer-2 rollup into an independent Layer-1 blockchain by 2027. According to StarkWare leadership, operating as an L2 leaves the network bound to Ethereum's underlying elliptic-curve signature scheme, limiting its ability to achieve total quantum security. Moving to a native L1 architecture will allow Starknet to implement hash-based STARK cryptography across both execution and consensus layers, positioning it as a fully quantum-resistant base layer designed to survive classical and post-quantum cryptographic threats.</description>
    </item>
    <item>
      <title>MIT Trajectory Planner Guarantees Collision-Free Robot Navigation</title>
      <link>https://news.mit.edu/2026/planning-system-ensures-robots-flight-path-will-remain-collision-free-1007</link>
      <guid isPermaLink="false">twd-s10</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>Engineers at MIT created a trajectory planning system called SANDO that gives autonomous drones formal mathematical guarantees of remaining collision-free while navigating unpredictable environments. Published in IEEE Transactions on Robotics, the algorithm combines real-time camera imagery with reachability analysis to continuously update safe flight envelopes around dynamic obstacles, such as moving debris or other aircraft. By verifying safety bounds within milliseconds rather than relying on probabilistic approximations, the planning architecture enables emergency-response aerial vehicles to maneuver through dense, unmapped disaster zones at high speeds without risk of crashing.</description>
    </item>
    <item>
      <title>MIT AI Framework Optimizes Cloud Infrastructure Energy Usage</title>
      <link>https://news.mit.edu/2026/mitigating-environmental-threat-of-data-centers-christina-delimitrou-1008</link>
      <guid isPermaLink="false">twd-s11</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>MIT researchers introduced an interpretable AI management platform designed to cut energy consumption across large computing facilities by dynamically aligning software workloads with server hardware capabilities. Developed under Associate Professor Christina Delimitrou, the framework incorporates explainable feedback loops that reallocate processing tasks and manage thermal profiles across server racks in real time without dropping performance targets. By optimizing hardware utilization and reducing cooling overhead, the system offers data center operators an actionable method for controlling power surges and environmental impacts caused by heavy cloud AI workloads.</description>
    </item>
    <item>
      <title>Zcash Developers Unveil Post-Quantum Signature Plan and Wallet Privacy Tool</title>
      <link>https://decrypt.co/380637/zcash-plans-hide-keys-ai-quantum-attacks</link>
      <guid isPermaLink="false">twd-s12</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>Zcash full-node development team Zakura announced plans to introduce post-quantum signature opcodes to the Zcash network in January, creating quantum-resistant transparent payment pools. The system pairs post-quantum Winternitz hash-based signatures with automated single-use address rotation and private information retrieval (PIR). Integrated into the newly released Vizor wallet, PIR allows users to query transaction status without revealing which addresses belong to them to light-wallet servers. The double-layered defense protects transparent Zcash transactions against both AI-driven elliptic curve decryption and server-side linkability attacks.</description>
    </item>
    <item>
      <title>Allen Institute Open-Sources Fast Synthesis Model AstaBrief</title>
      <link>https://huggingface.co/blog/allenai/astabrief</link>
      <guid isPermaLink="false">twd-s13</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>The Allen Institute for AI open-sourced AstaBrief, an 8-billion-parameter language model specifically fine-tuned for rapid, citation-backed scientific report generation. Built using specialized preference learning and dense retrieval pipelines, the model condenses hundreds of scholarly articles into structured analytical summaries while preserving the precise evidentiary scope of source claims. By releasing the weights under an open license on Hugging Face, researchers and academic institutions can deploy private, high-speed literature synthesis systems directly on local hardware without sending sensitive research queries through commercial cloud interfaces.</description>
    </item>
    <item>
      <title>MIT Lincoln Lab Tracks Rapid AI Accelerator Hardware Evolution</title>
      <link>https://news.mit.edu/2026/supercomputing-researchers-document-evolution-ai-hardware-1006</link>
      <guid isPermaLink="false">twd-s14</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>Supercomputing researchers at MIT Lincoln Laboratory released their updated multi-year survey documenting the technological evolution and operational efficiency of modern AI accelerator architectures. Tracking metrics across graphics processors, custom application-specific integrated circuits, and neuromorphic edge chips, the benchmark mapping details how compute density, interconnect bandwidth, and memory power constraints have transformed over consecutive hardware generations. The comprehensive dataset provides system architects and supercomputer designers with standardized empirical guidelines for matching emerging neural network architectures to optimal silicon hardware profiles.</description>
    </item>
    <item>
      <title>eCash Introduces L1 Miner-Secured Drivechains for Bitcoin Programmability</title>
      <link>https://www.theblock.co/research/blockchains/ecash-drivechains-programmability-bitcoin-architecture-418321</link>
      <guid isPermaLink="false">twd-s15</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>Bitcoin hard fork eCash published architecture plans to launch seven miner-secured drivechains alongside its mainnet release on October 31. Based on Bitcoin Improvement Proposals 300 and 301, drivechains enable developers to deploy custom sidechains with specialized functionality—including prediction markets, asset issuance, and private payments—without altering base-layer consensus. Drivechain state transitions and cross-chain transfers are secured through blind merged mining powered by Layer-1 proof-of-work hashrate, with funds held in a hash-rate escrow released through multi-month miner voting consensus.</description>
    </item>
    <item>
      <title>Technology Innovation Institute Unveils Falcon ASR for Speech</title>
      <link>https://huggingface.co/blog/tiiuae/falcon-asr</link>
      <guid isPermaLink="false">twd-s16</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>Abu Dhabi's Technology Innovation Institute released Falcon ASR, an open-weight speech recognition model optimized for multilingual and multi-dialect transcriptions across regional speech patterns. Built on the Falcon foundation architecture, the model accurately transcribes real-time conversational audio in diverse acoustic settings, achieving leading accuracy on regional Arabic dialect benchmarks like Casablanca. The open release allows developers to build low-latency, privacy-first voice interfaces, offline captioning tools, and audio analysis applications tailored to underrepresented language communities.</description>
    </item>
    <item>
      <title>Polygon Open Money Stack Expands to TRON for Regulated Payment Rails</title>
      <link>https://www.theblock.co/sponsored/2026-10-07-polygon-open-money-stack-expands-to-tron-extending-network-access-to-regulated-u-s-payment-rails-417930</link>
      <guid isPermaLink="false">twd-s17</guid>
      <pubDate>Wed, 14 Oct 2026 11:00:00 -0000</pubDate>
      <description>Polygon announced the extension of its Open Money Stack framework to the TRON network, connecting TRON's $94 billion USDT stablecoin ecosystem directly to regulated U.S. banking rails. The integration provides decentralized application developers and institutional payment flows on TRON with standardized software toolkits for fiat onboarding, automated compliance checks, and cross-chain settlement. By bridging Polygon's institutional payment architecture with TRON's high-volume transaction layer, the initiative streamlines stablecoin settlement workflows for global merchants and commercial payment service providers.</description>
    </item>
  </channel>
</rss>
