Summary

  • Less than two years after it was launched, xAI’s AI model Grok 3 appears to be the most advanced yet, beating all competitors in benchmarks and the user-evaluated Chatbot Arena
  • The implications of this are that competition in the labs is increasing, which means that models will soon be released faster and more frequently
  • This is good for users but may destabilise developers who are used to relying on consistent models
  • Grok 3 was trained quickly thanks to xAI’s Collosus supercluster in Memphis, which vindicates the huge investments made in AI accelerators
  • There is also a move towards open-sourcing large language models, exemplified by Grok 1 being open-sourced by xAI, and speculation that OpenAI will follow this lead
  • However, the reaction to Grok 3 has been mixed, with some criticism and other praise regarding its capabilities

By Ben Dickson

Original Article