News › Topic
6 stories in the last 60 days, newest first.
A US senator-elect's claim that Chinese AI merely copies American breakthroughs has triggered pushback from AI researchers. Scientists point out Chinese labs now produce world-leading open models and efficient architectures that global researchers rely on daily.
New evaluation details from METR show Anthropic's unreleased Model 4 scored above 50% across 15 separate capability benchmarks. The accelerated progress, driven by AI helping build future AI, has outpaced forecasts and raised questions about current benchmark limits.
Apple has cut internal AI spending budgets for staff even while encouraging employees to adopt tools, Bloomberg reports. Further reductions are reportedly planned as AI compute costs mount.
Microsoft has filed a patent application for AI tech that can edit what people say live on air, Computerworld reports. The tool can swap words, cut profanity or filler, and automatically adjust lip movements in video feeds in real time to match the altered audio.
Evaluators at METR found Anthropic is progressing nearly twice as fast as historical baselines, largely driven by using AI internally for research and development rather than massive compute increases. Tests on unreleased 'Model 4' also cleared 50% across every tested benchmark.
OpenAI revealed an upcoming cybersecurity model developed 948 distinct privilege-escalation exploits in the Linux kernel during automated tests. Researchers warned that low-privileged accounts on unpatched operating systems should now be treated as functionally equivalent to root access.
