AI Liability, RTX Unlock & Agentic Trading Hype
Key AI stories include an unresolved liability question for open-model inference failures, a Linux hack that doubles RTX 4090 tensor performance, and IBKR promoting agentic AI for simultaneous multi-strategy execution.
What we know
An account on X raises an unresolved legal question: if an inference provider uses an open model and a major hack occurs, who bears liability — and can we even determine that ahead of time? On the hardware side, another account on X RTs a finding that a Linux kernel module called `gpu-unlock-ada` removes NVIDIA's hidden throttle on consumer Ada GPUs, doubling FP16 tensor performance from 165 to 331 TF/s. Another account on X is marketing agentic AI integration for executing multiple trading strategies simultaneously — framing AI as the execution layer while the human focuses on strategy. Another account on X counters the hype by arguing AI agents are just autocomplete replacing a human user, and securing them is purely a code security problem.
By the Global Eye 24 desk, with AI assistance. Updated as the story develops. Spot an error? Write to corrections@globaleye24.com.