FLASH24 English

AI safety evaluation shifts from output checking to structural memory tests

New research published on 5 October argues that current benchmarks fail to test how AI agents organize long-term memory, a gap that exposes critical safety vulnerabilities in modern systems.

Reflection unveils Beam to counter Chinese AI efficiency

On 5 October, Reflection AI released Beam, a 501-billion-parameter open-weight model designed to match Chinese rivals like GLM-5.2 while using 3-4 times less inference compute.

Vals AI finds magnetic semiconductors with Claude agents

Vals AI reported on 5 October that Claude Opus 5.5 agents identified two room-temperature antiferromagnetic semiconductors.

UC Riverside study maps internal features driving AI confidence and accuracy

Small models move to the edge: Reika, Jeff-Code, and d1 redefine local AI

AI agents in enterprise face a knowledge gap, not a model gap

Reflection debuts Beam, a 501B open-weight model for efficient reasoning

AI governance moves from principles to deployment at October summits

Data centers face power and cooling squeeze as regulators tighten grip

EU car targets drive EV sales as electricity market reforms lag

Electric battery tech hits new milestones as EV sales surge in Europe

Apple tightens macOS privacy rules as AI agents demand full disk access

Local agents, strict budgets, and the fight for developer control

Ideogram 4.5 launches to fix image editing drift and preserve detail

Open weights dominate AI token processing as Google restricts free access

DOJ arrests Californian over $300M Nvidia chip smuggling to China

arXiv caps submissions as AI floods preprint servers with low-quality papers

US decouples from Chinese tech as Singapore builds its digital edge

AI agents stop being chatbots, start breaking databases

Origami geometry scales ocean seaweed farming for carbon removal