Elon Musk’s AI chatbot, Grok, is leveling up as it plays catch-up with its major competitors in the large language model (LLM) arena.
xAI, the startup founded by billionaire entrepreneur Musk last July in response to his qualms with OpenAI’s profit-generating model, released key features of the upcoming version of Grok on Thursday: Grok-1.5.
The latest version of the LLM (or the system that understands and generates human language text) has some pretty significant upgrades, according to the company. The announcement focuses on Grok’s improved performance metrics — a decided contrast from the CEO’s branding of Grok as a snarky, sarcastic alternative to existing chatbots.
The company now touts Grok’s more sophisticated ability to carry out coding and math-related tasks, bringing it closer to results seen with Google’s Gemini model and OpenAI’s ChatGPT. In the MATH benchmark — one of two mathematical measures that includes grade school to high school problem-solving — Grok-1.5 scores 50.6% accuracy, the company said. That’s up from Grok-1’s 23.9% score. Meanwhile, Gemini Pro 1.5 scored 58.5%, ChatGPT scored 52.9%, and Anthropic’s Claude 3 Opus, which it released early this month, hit 61%, the highest of all other AI models.
Read more: Amazon’s $4 billion investment in AI startup Anthropic is its biggest ever
In addition to math, Grok-1.5 also marks an improvement on the model’s language capabilities. The chatbot will be better able to understand long context and advanced reasoning, with an increased memory capacity of as much as 16 times the previous context length, xAI said.
To be clear: Grok still lags behind just about every other major AI model in nearly all criteria. Despite the odds, Musk is still boasting that an upcoming Grok-2 will “exceed current AI on all metrics,” adding that the next model is in training.
Early testers and existing Grok users on X, which Musk purchased as Twitter in October 2022, will be able to try out 1.5 by next week, Musk said in the same post on the site.
When xAI launched Grok in November, Musk billed the model as a mouthier and more personable addition to the existing AI landscape — with access to the plethora of knowledge and personalities on X.
xAI has been scrambling to catch up to its well-adopted competitors. Perhaps to attract more users, Musk made Grok open-source earlier this month. The move comes as Musk wages an ongoing spat-turned-lawsuit with OpenAI — the company Musk helped found in 2015 and left 3 years later over ideological differences.