Did you know Court Filings Expose Meta’s Determined Drive to Surpass OpenAI’s Latest GPT-4 Model

 

It looks like the competition in the world of AI is heating up after new court filings shared Meta’s intention of overpowering the greatness of OpenAI’s GPT model.

The news has to do with the timeframe when Facebook’s parent firm was busy working on its Llama 2 model as internal messages shared with the court dated to that period. This happens to be a part of the firm’s current cases linked to AI copyrights called Kadrey vs Meta.

The company’s VP for Generative AI stated in October of 2023 how the goal is to beat GPT-4 and reach its level of competition if it wants to win. For that, he shared how 64k GPUs were on their way. Therefore, the only task remaining is designing a frontier that would help in winning the race.

What is shocking is how the company does release its own fair share of AI models but its main goal right now is limited to AI leaders getting more focused on putting out competition like OpenAI and Anthropic. They even wished to put them behind an API so they could worry about them less.

The top executives and researchers at Meta wanted to keep two leading models as the gold standard to beat and that was GPT-4 from OpenAI and Claude from Anthropic. Interestingly, there was also discussion about another leading AI startup company from France called Mistral but again, Meta brushed them off as newbies that were easy to beat out and how they needed to do better.

These reports are clear proof of how tech firms are working hard to race and upstage all others in the industry having state-of-the-art cutting-edge AI tech. The court filings shared how competition was at its peak and leaders at Meta were obsessed with being the best and that meant behaving aggressively to get exclusive training rights for Llama.

Now, prosecutors linked to the case shared how the tech giant was willing and did end up cutting corners to ship out AI models and perform training on books that were guarded by copyrights. There was a mix of datasets used for the Llama 2 model and therefore that really ended up screwing up the system.

The news is not too surprising for some, especially after CEO Mark Zuckerberg shared with the world his desire to be the best in AI and shut any performance gap its models had with others from Google and OpenAI.