SpaceXAI collaborates with AI coding startup on legal
Grok 4.5 has been designed to effectively handle finance and coding tasks as well
Elon Musk’s SpaceXAI has collaborated with artificial intelligence coding startup Cursor on a new AI model for legal, finance, and coding tasks, reported Bloomberg.
Grok 4.5 is the first joint AI model for SpaceXAI and Cursor following the former’s agreement to acquire the latter. Per Cursor’s blog post, safeguards have been incorporated to enhance the mixture-for-experts model’s cybersecurity capabilities.
The model has been trained on “trillions of tokens of Cursor data” encompassing various user interactions with codebases and software tools, the coding startup wrote in the post. Thus, Grok 4.5 can learn from current software and developer-agent interactions.
The model’s training drew from “high-quality STEM tasks, research papers, and other knowledge work,” Cursor added. It also used its older coding-specialist model Composer 2.5 to develop Grok 4.5.
“We used reinforcement learning on difficult problems in realistic environments spanning both software engineering and broader knowledge work. These environments teach the model to investigate problems, use tools, recover from mistakes, and verify results,” Cursor wrote in the blog post. “Many of these problems had to be designed to be difficult enough that even frontier models fail at them.”
The startup came up with a distributed agent system to build the environments at scale.
“Engineers specify a problem and how a solution is verified, and large groups of agents construct, test, and refine each environment. Some would have taken teams of hundreds of engineers months to build. This is one of the ways in which we used the previous model to accelerate progress on the next model,” Cursor wrote in the post.
Grok 4.5 has been rolled out across desktop, web, iOS, CLI, and its SDK. The model represents an attempt by Musk to catch up to rivals Anthropic and OpenAI while also courting Wall Street clients, according to Bloomberg.
Earlier this year, a lawyer used Grok to research and draft documents in a bid to vary or remove the interlocutory suspension of his licence by the Law Society of Ontario Tribunal. The resulting output was riddled with inaccuracies in citations, hyperlinks, and the application of the tribunal’s rules – which the lawyer failed to check.
Related Stories
AI News
Synchrony Says AI Has Put Every Checkout in Play
33 minutes ago
AI News
OpenAI says it detected malign activity months before Hugging Face attack
33 minutes ago
AI News
AI drives 3 in 10 business tasks, widening the productivity gap
33 minutes ago
AI News
Kazakhstan Makes 50 Government Services Available Through eGov GPT
1 hour ago
AI News
Artificial intelligence: key updates and developments (13
1 hour ago
AI News
Thailand ranks 26th for AI traffic as ChatGPT dominates
2 hours ago
AI News
Amazon factual lawsuit: Twitch streams used for AI training
2 hours ago
AI News
Some assembly required, with machine learning
3 hours ago