MITB Banner

The Biggest Winner from Threads: LLaMA

Meta is in the AI race, not the social media race

Share

The Biggest Winner From Threads: LLaMA
Listen to this story

Threads is all the hype on the internet right now, arguably, for all the wrong reasons. People have been migrating from Twitter to Threads and comparing its features, but the real reason to launch the text-based social media platform is possibly to acquire data and build another large language model (LLM), which would be called LLaMA, or ‘Super LLaMA’. 

A month ago, on the Lex Fridman Podcast, Meta founder Mark Zuckerberg said that the company is working on another version of LLaMa, its open source LLM. On the same podcast, Zuckerberg also spoke about how the company is working on an alternative to Twitter, codenamed P-92. Within a month, it was launched as Threads, a microblogging platform clearly inspired by Elon Musk’s Twitter. Now, Musk is suing Meta for copying its platform.

Interestingly, Zuckerberg has tweeted after 11 years, and that too with just a meme.

Zuckerberg has also been praising the approach of Mastodon and Bluesky for building a decentralised platform for users. He further adds how a platform like Wikipedia can be driven using the community. Interestingly, he also said that the next version of LLaMa would be possibly trained on all the services that Meta offers, which now includes the text-based social media platform Threads. 

Decentralised for the win

Twitter boss Elon Musk has long been wanting to build an alternative to OpenAI’s ChatGPT. For that, it is speculated that he might leverage Twitter data to make it more (or less) aligned. But there has been no news about something like this even being made ever since Musk said that he was roping in a DeepMind researcher to build TruthGPT.

Now, Zuckerberg knows there is something powerful behind the idea of training a chatbot based on a social network’s data. The goal of building an alternative to Twitter has possibly now turned into building an alternative to OpenAI. The best thing that Meta can do here is keep the technology open source, which Zuckerberg and even Yann LeCun, the Meta AI chief, has been advocating for all this while

Now that Zuckerberg has Threads, he can collect and hoard data from users and use it for building generative AI models. Interestingly, the bid to open source AI models is continued, and this time ‘Super LLaMa’, would be decentralised. This does sound like Zuckerberg is giving the control of its data back to the users, but he could have a few tricks up his sleeve.

To make a case for OpenAI’s GPT-3.5 and GPT-4, the model is built by scraping the internet, a lot of which is Wikipedia pages, which are open to the public. Now, if Meta decides to make Threads decentralised, it can be one of two things — the data would either be completely publicly accessible or accessible only to Meta. 

In either cases, Meta would be able to avoid all legal issues of training on public data, or in this social media platforms, by simply saying that the platform is decentralised. The current version of LLaMa is only trained on web scraped data, and not on Meta’s services. However, it seems like Meta might actually do it. 

Meta is in the AI race, not the social media race

Instagram, WhatsApp, or Facebook, have a huge amount of data available in text, video, or images format. This could be beneficial for building a multimodal AI in the future. But when it comes to the legality of using data, the questions would always remain.

While OpenAI is trying to solve the so-called alignment problems, Meta is catching up. By betting big on open source, Zuckerberg is also putting away its responsibility of aligning AI to everyone. Clearly, the plan is wild but might work in Meta’s favour. 

On the other hand, Zuckerberg knows that even with the current capabilities of LLaMa, it is still “an order of magnitude smaller than what OpenAI and Google are doing”. Possibly, Musk has now given the idea to Zuckerberg to build something on a text-based social media platform. So, even when Meta might be four years behind OpenAI in terms of AI, Threads might make it possible for it to catch up, and build something to kill ChatGPT, before Musk does. Musk needs to catch up. 

Interestingly, Zuckerberg has already started restricting accounts on Threads, something that Musk said is one of the main reasons to acquire Twitter, and also build a rival to OpenAI’s woke chatbot. So even if Meta builds a chatbot, it would not be similar to what Musk wants to build. 

Moreover, Meta is already spearheaded ahead of Twitter and even ChatGPT when it comes to users. Threads crossed 10 million users within seven hours of its launch whereas ChatGPT took seven days to cross 1 million users. The fishy part is only the privacy policies around Threads. Users currently cannot even delete their accounts without deleting their Instagram accounts. It seems like a forced way to keep users on the platform, instead of offering them something that Twitter does not.

Share
Picture of Mohit Pandey

Mohit Pandey

Mohit dives deep into the AI world to bring out information in simple, explainable, and sometimes funny words. He also holds a keen interest in photography, filmmaking, and the gaming industry.
Related Posts

CORPORATE TRAINING PROGRAMS ON GENERATIVE AI

Generative AI Skilling for Enterprises

Our customized corporate training program on Generative AI provides a unique opportunity to empower, retain, and advance your talent.

Upcoming Large format Conference

May 30 and 31, 2024 | 📍 Bangalore, India

Download the easiest way to
stay informed

Subscribe to The Belamy: Our Weekly Newsletter

Biggest AI stories, delivered to your inbox every week.

AI Forum for India

Our Discord Community for AI Ecosystem, In collaboration with NVIDIA. 

Flagship Events

Rising 2024 | DE&I in Tech Summit

April 4 and 5, 2024 | 📍 Hilton Convention Center, Manyata Tech Park, Bangalore

MachineCon GCC Summit 2024

June 28 2024 | 📍Bangalore, India

MachineCon USA 2024

26 July 2024 | 583 Park Avenue, New York

Cypher India 2024

September 25-27, 2024 | 📍Bangalore, India

Cypher USA 2024

Nov 21-22 2024 | 📍Santa Clara Convention Center, California, USA

Data Engineering Summit 2024

May 30 and 31, 2024 | 📍 Bangalore, India