(Credit: SOPA Images / Contributor / LightRocket via Getty Images)
Elon Musk's xAI startup has released an upgraded version of the Grok chatbot. It's available now on X with a Premium ($8/month) or Premium+ ($16/month) X subscription.
The model, dubbed Grok-2, is still in beta, but xAI says it's a "significant step forward" from the previous model, Grok-1.5. It also offers image generation capabilities, bringing it closer to competitors like OpenAI, which has Dall-E.
"Compared to its predecessor, Grok-2 is more intuitive, steerable, and versatile across a wide range of tasks, whether you're seeking answers, collaborating on writing, or solving coding tasks," says xAI.
The company also debuted a smaller, faster version called Grok-2 mini. Both are available to X users through a newly designed interface, shown below. Grok-2 aims to be the "state-of-the-art AI assistant" experience, with real-time information gathered from X posts. The mini model is still "capable," but it "offers a balance between speed and answer quality."
Grok-2 has already outperformed models from other big names in AI, including OpenAI, Anthropic, Google, and Meta. That's according to the Large Model Systems Organization (LMSYS) chatbot arena, a rating system run by the University of California, Berkeley. The LMSYS site crowdsources blind tests from thousands of people who rank chatbots before the site reveals which one they were chatting with.
Grok-2 is number three on the LMSYS leaderboard under the code 'sus-column-r.' It's still below OpenAI's GPT-4o but above GPT-4o mini. It's also above Claude 3.5 from Amazon-backed Anthropic, Google's Gemini Advanced, and Meta's Llama 3.
How is it possible that a relatively new chatbot is already competing with the likes of OpenAI, Google, and Meta? Though the LMSYS ranking system is popular with AI insiders, it may not be a comprehensive assessment. It also does not explicitly rate the ability to report the news, which is a big focus for Grok.
But Grok has been caught spinning up inaccurate headlines and false information on X, particularly around elections. One error prompted five Secretaries of State to urge Musk to redirect US election-related inquiries to CanIVote.org to protect against chatbot hallucinations.
Musk did not respond to that, but when the National Association of Secretaries of State reached out to X about inaccuracies last month, X simply noted that Grok was scheduled for an update to Grok 2 in August. So perhaps Grok-2 is Musk's response.
Today, Musk retweeted a post from someone who said Grok is "truthful and well informed." The Grok-2 announcement also mentioned the model's improved performance with "providing accurate, factual information." It aims to do this by "correctly identifying missing information, reasoning through sequences of events, and discarding irrelevant posts."
Another consideration: Most chatbots across companies could have access to similar data and appear similar to the average user. In a blog post on the future of OpenAI, Edward Zitron argues that "transformer-based models have become heavily commoditized" and are "all trained on the same massive datasets." The differences between large models could be slight and may continue to narrow as they run out of readily available data to gobble up.
Since Grok's November 2023 release, xAI has rapidly released new versions. Grok-1.5 arrived in April 2024, and now we have Grok-2. That's three upgrades in less than a year, and this latest iteration won't be with us for long. Musk says Grok-3 is coming later this year.