This article analyzes the information leaked by Meta's internal files, revealing the intensity of internal competition in the process of developing the LLAMA3 model and the strong desire to surpass the OpenAI GPT-4. This information comes from a lawsuit on an artificial intelligence copyright. The document unsealed by the court shows the internal discussion and decision -making process of META executives in terms of data acquisition, model performance improvement, and comparison with competitors. The pressure and challenges of technical leadership.
Recently, Meta's internal emails on LLAMA3 were made public, which revealed the efforts and dilemma faced by the company executives to surpass the GPT-4 model of OPENAI. Meta executives expressed their strong desire to surpass competitors in the email, and actively sought to improve model performance, and even risked the risk of using copyright protection data. However, this eagerness of success has also triggered copyright litigation. Although LLAMA3 finally released and achieved good results, there was huge pressure and risks behind it. Through the disclosure of these internal documents, we can learn more deeply about the fierceness of large technology companies in the field of artificial intelligence, as well as their strategies and challenges they have adopted in pursuit of technical leadership.
Although Meta released an open AI model, the company's AI leadership obviously pays more attention to those competitors who do not disclose the weight of the model, such as Anthropic and Openai, and regard their Claude and GPT-4 as working standards. Although the French AI startup Mistral has been mentioned many times, META executives seem to be quite impressive. Al-Dah said in the information: "Mistral is a piece of cake for us, and we should be able to make better."
In the field of AI, major companies are scrambling to launch advanced AI models, and these court documents show the high tension atmosphere of Meta in this competition. Among multiple information, Meta's AI leaders mentioned that they are "very positive" in data needed for training LLAMA. A executive even said: "Llama3 is the only thing I care about." They discussed how to improve the data set to increase the performance of LLAMA3.
However, the prosecutor of the case accused Meta executives in the process of being eager to launch the AI model. It may be omissions in data use, involving some copyright -protected books. Touvron mentioned that LLAMA2's data set combination is "poor effect" and explores how to use better data sources to improve LLAMA3. Al-Dahh asked: "Do we have a suitable data set? Is there anything that cannot be used for the reason for stupid reasons?"
Marka Zuckerberg, CEO of Meta, previously stated that the performance gap between the LLAMA model and the closed source models of the LLAMA model and the company such as OpenAI and Google, Google, and other companies. These internal messages show that Meta is under tremendous pressure on the pursuit of this goal. In a letter in July 2024, Zuckerberg mentioned: "This year, LLAMA3 is competitive in the most advanced model and leads in some areas."
In April 2024, Meta finally released LLAMA3. This open AI model performed well in the competition and surpassed the open options from Mistral, but the data used by the training model -these data are said to have been obtained by Zuckerberg’s’s’s’s. Approval is facing a review of multiple lawsuits.
Points:
Meta executives focused on the GPT-4 model that surpassed OpenAI during the development of LLAMA3.
The company performs positive performance in obtaining data, but is facing allegations to use limited data.
Zuckerberg looks forward to the future LLAMA model can become the most advanced choice in the industry.
All in all, during the development of LLAMA3, Meta reflects the fierce trend of competition in the field of artificial intelligence, and also exposes the legal and ethical risks that may face when pursuing speed and leading position. In the future, how to balance technology development and compliance will become an important issue in the artificial intelligence industry.