Author: Li Dan
Source: Hard AI
In recent months, Meta has made efforts to launch the new generation of artificial intelligence (AI) model Llama 3 and its large version Llama 3.1 with over 400 billion parameters, aiming to create the most powerful open-source large language model (LLM). Recent news indicates that besides focusing on research and development, in order to achieve the ambition of becoming a leader in the AI field, Meta's CEO Mark Zuckerberg must acquire a new skill: selling their own AI software to large enterprise clients.
On Wednesday, August 21st, Eastern Time, it was reported that Llama has been struggling to gain attention on AWS, the cloud business platform under Amazon, which is the world's largest cloud computing service provider. AWS offers various LLM models to enterprise clients, with Anthropic's Claude being the most popular model on the platform. According to a Microsoft employee, Microsoft's salespeople usually only promote Llama to clients with data expertise, such as companies with engineers and data scientists.
For Zuckerberg, who aims to make Llama 3 an industry benchmark, the above news adds more pressure as Meta initially had to rely on other tech giants to promote their software to clients. The treatment Llama receives on Amazon and Microsoft may force Meta to establish its own sales and marketing team to directly promote the product to enterprises.
However, some companies have shown a more positive response to Llama. Previously, Amazon mentioned that Nomura, Japan's largest investment bank, and the U.S. outsourcing digital services company TaskUS are using Llama through the AWS platform. It was reported that an insider at Goldman Sachs revealed that the company will invest over 100 million USD this year to run Llama on Microsoft's Azure cloud to achieve code automation.
Furthermore, Meta is exploring ways to get more companies to use Llama. According to sources cited by the media on Wednesday, Meta's executives have been in discussions with Snowflake, a provider of cloud computing data storage warehouses, in recent weeks to expand the existing Llama sales agreements, allowing Snowflake's customers to use a customized version of Llama trained on their own data on the platform.
So far, Meta's various efforts have not yet helped Llama surpass its strong competitors. A project evaluating AI models conducted jointly by Lmsys and the University of California called Chatbot Arena shows that Llama still lags behind models introduced by competitors such as OpenAI, Google, and Anthropic.
A spokesperson for Meta stated on Wednesday that with the participation of numerous partners, Meta launched Llama 3.1 in July this year, listing some of the partners including AWS, Google Cloud, Azure, NVIDIA, Databricks, Snowflake, Groq, and Dell
Wall Street View once mentioned that Meta released a total of three Llama 3.1 models last month, with Llama 3.1 405B containing 405 billion parameters, making it the largest model Meta has ever produced. Zuckerberg referred to Llama 3.1 as the &#34;starting point of art&#34; and positioned it against large models from OpenAI and Google.
At the same time, NVIDIA's AI Foundry will provide customized services for the Llama 3.1 model to global enterprises, including NVIDIA, AWS, Azure, Databricks, and Dell. Meta has a total of 25 Llama-related enterprise partners. Accenture is the first enterprise to use the new service to build customized Llama 3.1 models for clients, with industry leaders such as Saudi Aramco, AT&amp;T, Uber, and others being the first users to access the new Llama NVIDIA NIM microservices.
On the day of the Llama 3.1 release, on social media platform X, most netizens congratulated Meta, considering Llama 3.1 a great masterpiece and a victory for the open-source community. Some netizens also questioned how much energy such a massive model would consume and what impact it would have on the environment.
Some industry insiders hold an optimistic view of Meta's prospects. &#34;Now, people are excited about Lamborghinis. That's OpenAI,&#34; said Laurent Gil, co-founder and chief product officer of Cast AI, a company that helps reduce cloud costs for enterprises, comparing OpenAI to Lamborghinis and stating that everyone wants to buy a Lamborghini, but Llama 3 is building Toyotas. There are far more Toyotas on the market than Lamborghinis.
Gil believes that most enterprises are unsure about which model to use and are caught up in the hype surrounding OpenAI. He thinks that in this competition of AI models, Meta will emerge victorious because enterprise customers will realize that Llama is a high-quality and lower-cost proprietary model alternative

Meta's most powerful model Llama 3 marketing encounters setbacks: exposed to be "looked down upon" by Amazon and Microsoft