4 articles
A new arXiv paper introduces DeepSeek Elastic Compute, or DSec, described as a sandbox infrastructure for effective agentic training at scale. The paper lists a large DeepSeek-led...
A new paper finds that real ternary LLMs do not use {-1, 0, +1} weights evenly, with zeros making up as much as 51.5% of weights across 29 tested models. The authors introduce BITC...
A new paper demonstrates that Ken Thompson’s trusting-trust attack can extend beyond compilers to ordinary build utilities. The researchers compromised GNU strip in NixOS’s binary...
An arXiv paper systematically tests transformer attention variants that share or collapse query, key, and value projections across synthetic, vision, and language tasks. The strong...