META EXEC DENIES THE COMPANY ARTIFICIALLY BOOSTED LLAMA 4'S BENCHMARK SCORES

Summary from the TLDR newsletter

An unsubstantiated rumor that Meta artificially boosted its Llama 4 benchmark results began circulating on X and Reddit over the weekend. Meta's VP of generative AI Ahman Al-Dahle has denied the rumors, saying that Meta did not train its models on test sets. The rumor appears to have originated from a post on Chinese social media from a user claiming to have resigned from Meta in protest over the company's benchmarking practices. Researchers have observed stark differences in the behavior of the publicly downloadable models compared with the models hosted on LM Arena.