Welcome to Tech Beat, your daily briefing on the stories shaping our digital world.
A small AI model is making a big claim today. Researchers say VibeThinker, a three billion parameter model, outperforms Anthropic's Claude Opus four point five on reasoning benchmarks. The method combines supervised fine-tuning with a reinforcement technique called GRPO, suggesting that clever training may matter more than raw scale.
Meanwhile, a tool called deletedby dot com is drawing attention for a quieter reason. It lets anyone browse a Reddit user's full post history, including content the user thought they'd deleted. It's a reminder that deletion online rarely means erasure, and that the gap between what users expect and what platforms actually store can be significant.
And in an embarrassing turn for Meta, the company has paused an internal employee tracking program after a data exposure let staff across the entire organization view sensitive colleague information. This happened while Meta is already navigating controversy over its AI training practices, and it raises familiar questions about how well large tech companies protect their own people's data.
Keep surfing. Tech Beat out.
