Cognition adds persistent Memory and daily “Dreaming” to Devin, enabling the coding agent to retain each user’s preferences, corrections and project lessons in a Git-backed Markdown repository across sessions, while open-sourcing its agent-memory standard.
Former Groq engineers Joshua Rubin and Benjamin Serebrin sue to challenge Nvidia’s $20 billion transaction, alleging shareholders are denied a required vote as $17 billion funds a technology license and $3 billion goes to restricted stock for employees joining Nvidia.
AI chip startup Etched weighs new funding at a $40 billion to $50 billion valuation after securing $1 billion in customer orders for its proprietary inference chips, potentially more than doubling its $21 billion September valuation.
Gatewai launches the beta of Framefields, an Apache-2.0 Node.js package that renders code-first video directly via WebGPU, Dawn, Metal or Vulkan instead of headless Chromium, while offering on-device ONNX vision tools for segmentation, pose tracking and person mattes.
OpenAI begins testing visual ads in ChatGPT image generation later in October with an initial U.S. advertiser group, while adding measurement and attribution partners including Hightouch, LiveRamp, AppsFlyer and Branch.
Living Models says its Gemma 4–powered BOTANIC-1 system can pinpoint causal crop mutations in hours, ranking the validated CmEIN3 melon flower-sex mutation first among 2,494 candidates while running locally on one NVIDIA L4 GPU.
Cohere launches North 2, an enterprise agentic AI platform that lets organizations build multi-step agents and deploy them in self-hosted, VPC, hybrid, on-premises or air-gapped environments, with North Admin enforcing access controls and token-spend limits.
OpenAI seeks a $30 billion funding round at a $1.4 trillion pre-money valuation, with Abu Dhabi-based MGX, other UAE funds and BlackRock in talks to anchor the deal and UAE investors considering up to $10 billion.
DeepSeek releases DeepGEMM Ascend, using DeepJIT runtime kernel compilation to avoid CUDA builds at installation and a fused Mega MoE kernel that overlaps NVLink transfers with tensor-core computation.
StayLameBro releases the pre-release Backburner, which uses an iPhone 17 Pro Max over 10 Gb/s USB-C as memory for an Apple Silicon Mac, pushing local Qwen3.8-27B AI context beyond 64,000 tokens to an estimated 196,000-229,000 tokens.