Musk’s ChatGPT Rival Isn’t as Smart as It Seems; Creators of a Key Open-Source Model Get Offer From Index Ventures

Elon Musk finally unveiled his alternative to OpenAI’s ChatGPT, which he developed because he felt it was too woke, as my colleagues first revealed eight months ago. Why do it on a Sunday? Perhaps because OpenAI’s Dev Day event in San Francisco is this morning, and Musk’s xAI team was hoping to upstage it. (Musk also doesn’t distinguish between weekdays and weekends, much to the dismay of our social media reporter: He’s a famous workaholic, after all).

At first look, I was slightly underwhelmed by xAI’s Grok, which feels like the millionth LLM-powered chatbot following the release of Bard and others. Musk previously described the mission of xAI as understanding the “true nature of the universe,” so I expected more than a snarky chatbot that answers “spicy questions that are rejected by most other AI systems,” per xAI’s own description. It’s somewhat ironic that xAI sees itself that way, as Musk was an outspoken supporter of pausing work on the other AI systems over safety concerns.

Taking Grok’s Temperature 

Here’s the interesting part. According to results shared by the xAI team, the LLM behind Grok outperforms OpenAI’s GPT-3.5 and Meta Platforms’ Llama 2 models on a number of benchmarks, though it doesn’t reach GPT-4 levels of performance.