today's digest covers regulatory moves to speed up ai infrastructure, a potential chip market shakeup, and new research on agent privacy, space-based vision models, and more.
- data centers get fast lane to grid - ferc orders grid operators to fast-track data center connections, addressing ai infrastructure delays.
- amazon in talks to sell trainium ai chips - amazon web services is exploring selling its trainium ai chips directly to other companies, potentially challenging nvidia's dominance.
- research agents leak private data through web queries - mosaicleaks shows that deep research agents often expose private information when they search the web, and training them to be better at tasks makes the leakage worse.
- first in-orbit vision-language model demo on spacecraft - navi-orbital runs a vision-language model on a low earth orbit satellite to classify scenes, describe content, and answer operator questions in plain english.
- general intuition seeks $300m for ai agents that learn from gameplay - the new york startup uses medal's 2 billion yearly gameplay videos to train embodied ai agents in spatial-temporal reasoning, attracting big-name investors.
- llm-guided autotuning speeds up pytorch kernel tuning - pytorch's helion dsl uses llms to propose kernel configurations, matching bayesian optimization performance while exploring 10x fewer configs and cutting tuning time by 6.7x.
- ceo-bench tests ai agents as startup founders - a new benchmark evaluates language model agents on long-horizon business management by simulating 500 days of running a startup.
other notable stories include a look at fine-tuning methods beyond lora, coordination challenges in human-ai teams, and a defense-in-depth approach to securing ai agents.