Meta Platforms(US:META) 量子位·2025-04-07 04:19Core Viewpoint - Meta's latest model Llama 4 has received significant criticism shortly after its release, with users expressing disappointment primarily regarding its coding capabilities and performance in various tests [1][4][12]. Group 1: User Feedback and Performance - Users have reported that Llama 4 failed in basic tests, such as the "atmospheric programming" ball rebound test, where the ball passed through walls [5][4]. - Despite good scores in official evaluations, Llama 4's performance drastically declined in third-party benchmark tests, placing it at the bottom of the rankings [8][12]. - The disparity between official scores and third-party evaluations raises concerns about potential data overfitting or vote manipulation in the rankings [12]. Group 2: Internal Issues and Allegations - Joelle Pineau, the head of Meta AI research, announced her departure just days before Llama 4's release, indicating possible internal turmoil [14]. - An anonymous report surfaced claiming that a former employee requested not to be credited in Llama 4's technical report, suggesting dissatisfaction with the model's development [15][19]. - Previous leaks regarding data issues have been noted, with claims that data leaks have persisted since Llama 1, raising questions about the integrity of the training data used [22]. Group 3: Comparison with Competitors - Llama 4's performance has been compared unfavorably to competitors like DeepSeek V3, which has shown superior training outcomes and lower operational costs [35][37]. - The rapid advancement of competitors in the AI space has led to concerns about Meta's ability to keep pace, especially following the recent controversies surrounding Llama 4 [35][37].