Meta Exec denies artificially Llama 4 sweat

On Monday, Meta Exec denied that the company has trained new artificial intelligence models to provide specific criteria while hiding weaknesses in the models.

Executive Director, Ahmed Al -Dahli, Vice President of Amnesty International in Meta, He said in a post on x It is “simply incorrect” who trained Meta Llama 4 MAVERICK and Llama 4 Scout Models On “test sets”. In artificial intelligence standards, test sets are sets of data used to evaluate the performance of the model after its training. Testing group training may lead to exaggeration of the standard measurement scores of the model, making the model look more capable than it is already.

During the weekend, Rumors are baseless This artificial definition has strengthened the results of the new models that started spreading on X and Reddit. Rumors appear to have arisen from a post on a Chinese social media site from a user claiming to be resigned from Meta in protest against the company’s standards.

Reports that Mavrick and Scout Lead Bad on Some tasks Rumors were nurtured, as did a dead decision using An unpublished experimental version of Maverick To achieve better degrees on the standard LM Arena. The X researchers have Stark is observed Differences in behavior From the downloadable MAVERICK compared to the model hosted on LM Arena.

The signifier acknowledged that some users see “mixed quality” from Mavreke and Scouts through the various clouds’ wardrobe who host the models.

Al -Dal said: “Since we dropped the models as soon as they are ready, we expect it to take several days until all public applications are contacted.” “We will continue to work through our error repairs and movement partners.”