TECH NEWS – Twitter may be worth less than half of what Musk paid for it last autumn, but development of the long-rumoured artificial intelligence-based texting model hasn’t stopped.

Grōk is obviously going head-to-head with OpenAI’s ChatGPT, but Musk has plans to go further: in time, he wants the technology to be integrated into the many millions of AI-enabled Tesla electric vehicles. That’s why he launched xAI in July with his eclectic team (DeepMind, OpenAI, Google Research, Microsoft Research, Tesla, University of Toronto were the team’s former workplaces), and to train the large language model (LLM) on the ChatGPT model, they used 10,000 GPUs bought from Nvidia in the first half of the year.

So far it seems that the size of a SuperPrompt context window with Grōk is 25 thousand characters (ChatGPT is limited to only 4096!). His response time is at screen refresh speed, and his personality is “spicy” and humorous. Its knowledge is based on a variation of the 886.03 GB “The Pile” and the entire Twitter platform of exabytes for fine tuning. The “Live” search engine will initially focus on Twitter for context. An API is planned, the outputs and prompts are “voice ready”. Image generation, as well as image and audio recognition are planned, and the current model has hints of this. A smaller quantized Grõk will run natively in the Tesla using local processing power.

Musk says the system will be available to all Premium+ subscribers once it comes out of early beta. What could be powerful about the system is that it can detect news better than other AI platforms by training on exabytes of Twitter data, so it can distinguish a recent, just-happened event from bias, early testers say.

But it’s interesting because back in March Musk called for a pause in all AI-related activity to assess the risks to society, but it seems that was just bluster from him to catch up on his ChatGPT backlog… another Musk move.

Source: WCCFTech