AI news digest — August 5, 2026
10 items, each with its source.
Google Signs AI Deals With 200 Publishers
Google has reportedly signed two-year AI training deals with more than 200 global publications. The agreements include NDAs and no-sue clauses in exchange for access to content, raising concerns about publisher independence and copyright compensation.
aiweekly.coMicrosoft Caps AI Spend and Token Usage for Engineers
Microsoft introduced division-level AI token budget targets, instructing engineers to stop 'tokenmaxxing.' The internal memo addresses the high costs of AI development and excessive token usage, highlighting the growing financial burden of scaling AI.
aiweekly.coxAI Launches Grok Voice Think Fast 2.0 Update
xAI has launched Grok Voice Think Fast 2.0 at $0.08 per minute, featuring a 0.70s first response time. The update auto-migrates users starting today, offering significant token reductions and cost efficiency for developers.
aitoolsrecap.comLiquid AI Ships New LFM2.5-2.6B Agentic Model
Liquid AI released a 2.69B-parameter hybrid convolution model tuned for on-device agentic workloads. It features a 131K context window and matches the performance of models four times its size, paving the way for more efficient local AI.
aiweekly.coSK Hynix and SanDisk Unveil New HBF Memory Standard
SK Hynix and SanDisk published the first OCP technical specification for High Bandwidth Flash. The new NAND-based memory tier for AI inference supports capacities up to 512GB and speeds of 3TB/s, addressing the critical memory bottleneck in AI hardware.
aiweekly.coSamsung and SK Hynix Test Chinese AI Chip Tools
Samsung Electronics and SK Hynix are evaluating chipmaking equipment from China's Advanced Micro-Fabrication Equipment. The move serves as a strategic hedge against future US restrictions on AI and semiconductor technology.
aiweekly.coRazer and NUS Establish Joint AI Research Lab
Razer and the National University of Singapore launched a joint AI research lab to advance gaming intelligence. The initiative focuses on digital humans and ambient intelligence, aiming to create more immersive and adaptive gaming experiences.
razer.comJD Ships Real-Time AI Video Editing on Single GPU
JD's Joy Future Academy released JoyAI-Video-Edit, a 16B-parameter autoregressive diffusion framework. The system runs end-to-end 720p editing at 30 FPS on a single Nvidia B200 GPU, significantly accelerating live production workflows.
aiweekly.coPAST-Bench Evaluates AI Agent Memory and Self-Improvement
Researchers introduced PAST-Bench, a new benchmark with 204 episodes to test whether personal AI agents recursively self-improve. The study evaluated seven base models and four agent frameworks, quantifying the actual gains from retained experience.
aiweekly.coNew Zealand Faces Growing AI Policy and Regulation Gap
A new analysis highlights the growing gap between rapid AI advancement and the lack of comprehensive AI policy in New Zealand. The report calls for urgent regulatory frameworks to address the societal impacts and infrastructure demands of artificial intelligence.
thespinoff.co.nzFeed, daily deep-dive and bytes — readable offline, with push alerts for the topics you follow.