Meta postponing the release of his most recent AI model star-news.press/wp

Target (Target-2.36%) Performed the edition of its latest model of artificial intelligence, internally known as “behemoth”, raising concerns among employees about the efficiency and direction of massive AI investments.

Engineers have fought to improve the performance of a large language model compared to previous versions in the LLAMA model, Wall Street magazine (NWSA-0.90%) Reports. This encouraged an internal debate on whether the upgrade is significant enough to deserve a public edition, sources said.

Shares in the target were more than 2% by Thursday close.

It was initially set for April debut during the Metaine and the conference and programmer AI, launching was pushed in June and is potentially delayed until at least fall.

When the target published two smaller llama models in aprilThe company is praised due to the speed with which RIVALs entered when it comes to developing its own generational AI tools.

The company is now facing increasing pressure to show meaningful progress with behemoth and signal to Wall Street, can monitor return investment. The target spent billions on AI development, with Plans for up to $ 65 billion In the capital expenditures of this year. His share price on the ratio of earnings is 25.05, from 15. May, which means that investors pay $ 25.05 for every $ 1 earnings.

Although the target has publicly claimed that behemoth surpasses competing models such as Openai, Google (Googl-1.57%), and anthropic in some benchmarks, internal sources have said in the journal that the model faces significant challenges of training, and its performance in the real world may not meet in the company’s public claims.

The target Ai Putor started with his basic research team AI, which was in early 2023. developed original models of Llam. However, 11 out of 14 authors of initial lagging paper have left the company sinceAnd newer versions build a different team.

Controversy has also created in April when It was discovered If the Metel Meto sent to a wide used leader and Chatbot was not the same version that is released to the public. Meta later admitted that it was optimized by a sent model specifically for a scale.

The target is not alone in dealing with delays. The following main model of Openay, GPT-5, also delayed, and the company instead of overturning the intermediate version called GPT-4.5. Anthropic is the release of Claude 3.5 opus, larger and powerful AI models at the same time, although the edition is direct.