onlyTrustedInfo.comonlyTrustedInfo.comonlyTrustedInfo.com
Font ResizerAa
  • News
  • Finance
  • Sports
  • Life
  • Entertainment
  • Tech
Reading: Meta exec denies the company artificially boosted Llama 4’s benchmark scores
Share
onlyTrustedInfo.comonlyTrustedInfo.com
Font ResizerAa
  • News
  • Finance
  • Sports
  • Life
  • Entertainment
  • Tech
Search
  • News
  • Finance
  • Sports
  • Life
  • Entertainment
  • Tech
  • Advertise
  • Advertise
© 2025 OnlyTrustedInfo.com . All Rights Reserved.
Tech

Meta exec denies the company artificially boosted Llama 4’s benchmark scores

Last updated: April 7, 2025 2:45 pm
OnlyTrustedInfo.com
Share
2 Min Read
Meta exec denies the company artificially boosted Llama 4’s benchmark scores
SHARE

A Meta exec on Monday denied a rumor that the company trained its new AI models to present well on specific benchmarks while concealing the models’ weaknesses.

The executive, Ahmad Al-Dahle, VP of generative AI at Meta, said in a post on X that it’s “simply not true” that Meta trained its Llama 4 Maverick and Llama 4 Scout models on “test sets.” In AI benchmarks, test sets are collections of data used to evaluate the performance of a model after it’s been trained. Training on a test set could misleadingly inflate a model’s benchmark scores, making the model appear more capable than it actually is.

Over the weekend, an unsubstantiated rumor that Meta artificially boosted its new models’ benchmark results began circulating on X and Reddit. The rumor appears to have originated from a post on a Chinese social media site from a user claiming to have resigned from Meta in protest over the company’s benchmarking practices.

Reports that Maverick and Scout perform poorly on certain tasks fueled the rumor, as did Meta’s decision to use an experimental, unreleased version of Maverick to achieve better scores on the benchmark LM Arena. Researchers on X have observed stark differences in the behavior of the publicly downloadable Maverick compared with the model hosted on LM Arena. 

Al-Dahle acknowledged that some users are seeing “mixed quality” from Maverick and Scout across the different cloud providers hosting the models.

“Since we dropped the models as soon as they were ready, we expect it’ll take several days for all the public implementations to get dialed in,” Al-Dahle said. “We’ll keep working through our bug fixes and onboarding partners.”

You Might Also Like

Great white sharks are going north. Here’s what the numbers say

Decoding the Future of Flight: NASA’s X-59 Quiet Supersonic Jet Completes Historic First Flight

New Zealand Landslide Survival Search: Why Campers Were Caught Off-Guard by Record Rain

U.S. Cuts Antarctica’s Only Research Icebreaker Ship under Trump Budget Squeeze

Russia’s FSB targets foreign embassies in Moscow in cyber espionage campaign, Microsoft says

Share This Article
Facebook X Copy Link Print
Share
Previous Article U.S. crude oil losses deepen as Trump tariffs fuel recession fears U.S. crude oil losses deepen as Trump tariffs fuel recession fears
Next Article “Feel for Hardik Pandya. He is incredible but sadly team is losing”- Fans react as MI captain’s all-round brilliance goes in vain in IPL 2025 “Feel for Hardik Pandya. He is incredible but sadly team is losing”- Fans react as MI captain’s all-round brilliance goes in vain in IPL 2025

Latest News

Prince Andrew’s Legal Peril Deepens: Transatlantic Probe Targets Giuffre Family
Entertainment July 11, 2026
Sofia Vergara’s Etro Dress: The Keyhole Cutout That’s Turning Heads on Italian Streets
Entertainment July 11, 2026
Rick Springfield at 76: How the ‘Jessie’s Girl’ Icon Redefined Aging in Rock with His Viral Physique
Entertainment July 11, 2026
Prince Harry and Meghan’s Children Reunite with King Charles: A Royal Family Milestone After Years of Tension
Entertainment July 11, 2026
//
  • About Us
  • Contact US
  • Privacy Policy
onlyTrustedInfo.comonlyTrustedInfo.com
© 2026 OnlyTrustedInfo.com . All Rights Reserved.