Facebook supremo Mark Zuckerberg is redirecting Meta-wide efforts to build artificial general intelligence and wants to secure a whopping 350,000 or more Nvidia H100 GPUs by the end of the year to make that happen.
This comes as OpenAI reportedly signaled it wants to raise billions of dollars in funding to build a network of chip fabs to supply enough AI accelerators to meet its needs.
Zuck's boffins are right now training Llama 3, their latest content-generating large language model, and plan on releasing it openly, the billionaire confirmed Thursday via Instagram.
Zuckerberg is the latest CEO to throw his hat in the ring in the race to build AGI, a theoretical technology that can outperform humans in many cognitive tasks.
Training increasingly powerful models needed for AGI to become more than a pipe dream requires huge amounts of accelerators, and Zuckerberg said Meta intends to amass 350,000 H100s for the task.
Instead, he opined AGI will help build and fill out the project's interconnected worlds of 3D virtual reality, as the corporation has previously indicated when asked whether the CEO has given up on it in favor of artificial intelligence.
The original article contains 489 words, the summary contains 191 words. Saved 61%. I'm a bot and I'm open source!
Current neural networks are basically a set of highly sophisticated statistical rules. So unless the Facebook engineers have come up with some new breakthrough, they probably won't be able to build something much better than what's already around, as statistics can only take you so far.
“As of January 2022, there were about 2.3 billion class A shares in Meta, and 412.86 million class B shares. But although class B shares represent just 15 per cent of total stock, they represent 64 per cent of the votes. And it means Zuckerberg alone controls more than 57 per cent of votes – meaning the only way he can be removed as chief executive is if he votes himself out.”