To The Moon

$Meta Platforms, Inc.(META)$ announced this week that it will partner with major cloud computing providers including Microsoft to bring its large AI language model Llama 2 into commercial applications. This is indeed a major strategic move, signaling that Meta is actively transforming its business for the future.

1.Llama 2 Competes with ChatGPT

According to Meta's official data, Llama 2 has 40% more training data compared to the previous version, and includes 70 billion, 130 billion and 700 billion parameter versions. Llama 2 pre-trained models were trained on 200 trillion tokens, with context length twice as long as Llama 1. Its fine-tuned models have already been trained on over 1 million human annotations.

Similar to ChatGPT, Llama 2 also went through three stages of pre-training (PT), instruction fine-tuning (SFT), and human feedback reinforcement learning (RLHF). In external benchmark tests including reasoning, coding, proficiency and knowledge, Llama 2 significantly outperformed other open source language models like MMLU and TriviaQA in many aspects.

$Meta Platforms, Inc.(META)$ said Llama 2 can be used for free for research and commercial purposes. Developers can directly download the model from the official Llama 2 website. To avoid the emergence of deepfakes, inappropriate chatbots, and other harmful content previously seen after leaks of LLama, Llama 2 has undergone red team testing to prevent the model from generating bad and harmful content. Usage guidelines and code of conduct have also been formulated for developers.

2.Does Meta Rely on AI for Its Transformation?

In the past, Meta mainly made money through social media platforms like Facebook and Instagram by selling ads, which is Meta's most core and largest revenue source, accounting for about 98% of total revenue. But now growth in that business is slowing down, while other businesses are still in early growth stages and generate relatively little revenue. In the short term, Meta still needs to rely on social media ads to sustain revenue growth, so it must consider transforming to find new revenue sources.

Opening up Llama allows Meta more new attempts and possibilities in commercializing AI. Meta can leverage Llama to provide more AI services like voice assistants, article generation, etc. to more customers and earn revenue from it. This opens up new profit growth areas for Meta.

Some other factors to consider:

1. Under current tech development trends, whichever company has leading AI technology can seize first-mover advantage. Llama's development can greatly enhance Meta's capabilities in cutting-edge AI fields like speech recognition and natural language processing.

2. Opening up Llama can also attract more developers to build innovative applications based on the model. This can nurture an ecosystem of apps and services built on Meta's AI capabilities, which is also crucial for Meta's AI business development.

3. Meta can also apply Llama to its own social and virtual reality products to provide more personalized and intelligent user experiences through AI. For example, virtual assistants that can understand and respond to user questions. Additionally, advanced AI is an indispensable component in Meta's envisioned metaverse.

$Meta Platforms, Inc.(META)$ hopes to commercialize AI models like Llama 2, but running these large models requires massive computing power. By partnering with Microsoft Azure, Meta can leverage Microsoft's cloud computing capabilities to run its AI services. The partnership between Meta and Microsoft also makes perfect sense.

3.Why Didn't the Stock Price React?

The stock price remained relatively flat after the news broke, suggesting the market is not very optimistic about this opening. The main reason is that it's unclear how much money Meta can make from this. Compared to Microsoft, Meta is far behind in terms of commercializing AI capabilities. Aside from doubts about monetization abilities, the fierce competition in the AI field should also be considered. Facing aggressive moves by giants like OpenAI and Google, it is completely unknown whether Meta can do better than them.

The increased uncertainties and future risks make investing in Meta more difficult, especially since Meta's heavy investments in the metaverse have yet to produce satisfactory results.

In general, opening up Llama 2 has more of an experimental nature for now. Whether it will ultimately succeed still requires time to verify. If the main reason to buy into Meta is the opening up of large models, wouldn't it be better to wait and see?