钛媒体:引领未来商业与生活新知 05月16日 14:31
Meta Reported to Push Back Launch of Flagship AI Model
index_new5.html
../../../zaker_core/zaker_tpl_static/wap/tpl_guoji1.html

 

据报道,Meta公司可能推迟其旗舰人工智能模型Behemoth的发布。该模型属于Llama 4系列,但由于对其能力存在担忧,发布计划已被推迟。Meta工程师在显著提升模型能力方面面临挑战,引发了对其改进是否足以公开声明的质疑。尽管Meta增加了AI投资,但Behemoth的两次推迟引发了对其AI投资有效性的担忧。Meta此前宣称Behemoth超越了GPT-4.5等竞争对手,并将其用于提炼Llama 4的其他模型。公司增加了资本支出,以支持AI努力和基础设施硬件成本的增加。

🤔Meta的LLM模型Behemoth的发布可能被推迟,原因是Wall Street Journal援引知情人士的说法,称Meta内部对其能力存在担忧,该模型属于Llama 4系列。

📈Meta增加了AI支出,今年的资本支出预计在640亿至720亿美元之间,高于此前预计的600亿至650亿美元,这反映了对数据中心投资的增加,以支持其人工智能工作,以及基础设施硬件成本的增加。

💪Meta此前宣称Behemoth是一个拥有2880亿活跃参数的模型,包含16个专家,性能超越GPT-4.5、Claude Sonnet 3.7和Gemini 2.0 Pro等竞争对手,并用它来提炼Llama 4系列中的Scout和Marverick模型。

TMTPOST -- Meta Platforms, Inc. shares tumbled as much as 3.2% following a report about possible delaying launch of the company’s flagship artificial intelligence (AI) model, and settled around 2.4% lower on Thursday.


Credit:Meta

Meta is pushing back rollout of its large language model (LLM) dubbed “Behemoth” under its Llama 4 family due to concerns about its capabilities, the Wall Street Journal cited people familiar with the matter on Thursday.it was reported that Meta engineers are struggling to substantially improve capabilities of the model, stirring questions among the staff about if improvements over its predecessor versions are significant enough to justify public statements about its capabilities.

Meta didn’t respond to the report. If it is accurate, the social media titan has so far postponed its planned release of the Behemoth model twice. Early in the development of the model, the company was said to internally arrange the release in April to coincide with its inaugural AI conference for developers, but later delayed the internal target of rollout to June, and the current target became fall or even later.

Meta unveiled the first models in the Llama herd on April 5. it said the two smaller Llama 4 models, Scout and Marverick, “are our best yet thanks to distillation from Llama 4 Behemoth” and touted Behemoth is “a 288 billion active parameter model with 16 experts that our most powerful yet and among the world’s smartest LLMs.” It said Llama 4 Behemoth outperforms GPT-4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro on several STEM benchmarks.

The Thursday report aroused new concerns over the effectiveness and direction of Meta’s mounting AI investments. Two weeks ago, Meta announced updated guidance and showed its determination to ramp up AI spending this year. It now sees capital expenditure (Capex) for the year be in the range of $64billion to $72 billion, increased from its prior outlook of $60billion to $65 billion. That represented a lift of more than 7%.

“This updated outlook reflects additional data center investments to support our artificial intelligence efforts as well as an increase in the expected cost of infrastructure hardware,” Meta said in the earnings release.

Meta Chief Finance Officer Susan Li said the hike in expected infrastructure hardware is the result of “suppliers who source from countries around the world.” The CFO told analysts  “there’s just a lot of uncertainty around this, given the ongoing trade discussions”, adding that the company is "working on our end on mitigations by optimizing our supply chain."

更多精彩内容,关注钛媒体微信号(ID:taimeiti),或者下载钛媒体App

Fish AI Reader

Fish AI Reader

AI辅助创作,多种专业模板,深度分析,高质量内容生成。从观点提取到深度思考,FishAI为您提供全方位的创作支持。新版本引入自定义参数,让您的创作更加个性化和精准。

FishAI

FishAI

鱼阅,AI 时代的下一个智能信息助手,助你摆脱信息焦虑

联系邮箱 441953276@qq.com

相关标签

Meta AI模型 Behemoth
相关文章