Tech • AI • Robotics • Game

Midday edition

Wednesday, November 18, 2026

  • AI-generated illustration
    Alphabet’s Isomorphic Labs raised $5 billion to accelerate AI-assisted drug discovery, one of the largest financing rounds yet for computational biology. The capital signals investor confidence that models can move beyond generating text toward identifying therapeutic targets and designing molecules, despite the long clinical validation cycle. For both technology and pharmaceuticals, the round raises the stakes around proprietary biological data, lab integration and measurable pipeline results. Biology just closed a mega-round 😉
    Sage S2.1 Flash introduced a 320-billion-parameter general-purpose mixture-of-experts model built to reason over scientific evidence and act in dynamic environments. Specialized encoders for molecular graphs and biological data distinguish it from general chat systems, potentially making agentic workflows more useful in research. Its scale also underscores the compute and data demands of scientific AI, where domain representations can matter as much as fluent language. The lab coat now has 320 billion pockets 😉
    Full article
  • AI-generated illustration
    Attackers published malicious versions 1.26.1 and 1.26.2 of Microsoft’s agent-framework package on PyPI on November 17, targeting cloud credentials. The incident, flagged by multiple security researchers including Google-owned Wiz, turns a trusted dependency for AI-agent development into a supply-chain entry point. It matters because agent frameworks often hold broad permissions and secrets, magnifying one poisoned package’s reach across cloud environments. Dependency management just rolled a natural one 😉
    Java4Shell, tracked as CVE-2026-114938, enables unauthenticated remote-code execution against Jira, Jenkins and Keycloak when they run on Java 17.0.6 or later atop servers supporting AVX-512 VBMI. That unusually specific hardware-software intersection still threatens widely deployed development and identity systems, where pre-authentication compromise can hand attackers immediate control. Operators must inventory processor features as well as software versions, complicating remediation at enterprise scale. This shell arrives with boss-level privileges 😉
    Full article
  • AI-generated illustration
    Researchers demonstrated that language models can rewrite their own weights, challenging the standard deployment pattern in which a model remains frozen after one training run. Persistent self-editing could let systems absorb lessons without full retraining, potentially lowering update costs and accelerating adaptation. It also raises hard questions about auditability, rollback and whether safety checks survive internal changes. The model has discovered its own firmware menu 😉
    Separate work unified the design of self-improving agent harnesses, the runtime layer handling context, memory, multi-turn control, sandboxed tools and verification. A related WebAgents framework split agent creation, hosting and execution so independently built services can collaborate across an agent-native web. Together, these projects shift competition beyond foundation models toward orchestration software that can evolve and interoperate. Even the sidekick now wants an upgrade tree 😉
    Full article
  • AI-generated illustration
    Argentum AI signed a $630 million collaboration with Kyndryl to build and operate data centers using Nvidia GPUs, with deployment beginning in late 2026. The agreement shows how AI capacity is being assembled through partnerships linking workload providers, infrastructure operators and dominant accelerator vendors. For Nvidia, each such project extends its hardware lead into long-lived facilities and service contracts. The GPU rack has become the new party guest 😉
    Giga Energy signed a $1 billion lease for a Texas-based AI data center with a major cloud provider, adding another nine-figure commitment to America’s compute buildout. Unlike a speculative announcement, a lease ties capital to physical capacity, power demand and a specific operating geography. The deal matters because electricity access and site readiness are becoming as decisive as chip supply in scaling AI services. Texas just installed more RAM 😉
    Full article
  • AI-generated illustration
    On the gaming front, Firaxis Games laid off approximately 60 employees, one-third of its workforce, while studio head Heather Hazen departed. The cuts marked a sharp reset for the more than 30-year-old studio behind Civilization and XCOM, after a new XCOM project stalled over the summer and missed a major 2026 development milestone. The contraction raised questions about Firaxis’s production pipeline and the economics of strategy gaming. Even Gandhi’s tech tree lacks a layoff immunity upgrade 😉