Meta snaps up AI training chips and builds data centers to create more powerful chatbots

According to the Wall Street Journal, Meta has been snapping up training and building data centers to create a more powerful new chat that it hopes will be as sophisticated as OpenAI's GPT-4. The company reportedly plans to start training new large-scale language models in early 2024, and CEO Mark Zuckerberg plans to continue releasing new models in open source mode, so other companies can use these types of AI tools for free.

Meta has been buying more H100 AI training chips and is beefing up its infrastructure so that this time around, it doesn't need to rely on Microsoft's Azure cloud platform to train new chatbots, the Journal wrote. The company assembled a group to build the model earlier this year, with the goal of speeding up the creation of AI tools that can mimic human expressions.

Meta has reportedly split a lot of its computing resources between multiple LLM projects this year, leading to an exodus of AI researchers. It also faces stiff competition in the generative AI space. OpenAI said in April that it wasn’t training GPT-5 and “won’t for some time,” but Apple is reportedly pouring millions of dollars into its own “Ajax” AI model every day, which it apparently believes is even more powerful than GPT-4. Google and Microsoft are both expanding the use of AI in their productivity tools, with Google looking to use generative AI in Google Assistant. It also has generative AI initiatives throughout its organization, which could result in a chatbot-driven Alexa.

Reviewing Editor: Peng Jing