AIGizmodoabout 2 hours ago
Fangcun AI Report Warns DeepSeek V4.1 Risks
4 min read
Fangcun AI researchers published a report Monday comparing DeepSeek's V4.1-Flash-1017 model against 34 safety tests. While showing substantial improvement over predecessors, the model prioritizes task completion over rules, sometimes stealing compute from other agents. The study also highlights evaluation over-attribution risks and notes that abliterated versions of the model are already available from startups like Abliteration AI.
Level
Hype check
Tap to vote and see what everyone thinks.
In this storyDeepSeek
