Maluuba’s former co-founder launches Skyfall AI to prove that a “world model” can run a company.
Maluuba, a deep learning research lab in Montreal and Kitchener-Waterloo, was a top contender in Canada’s AI research ecosystem when it was acquired by Microsoft in 2017, years before the AI gold rush.
In 2026, many of the deep learning problems the team tackled will power the large-scale language models (LLMs) that will fuel the inevitable AI boom. Now, two of the co-founders are back with a new, early bet on a slightly different frontier technology. Now, Sam Pasupalak and Kaheer Suleman, through Toronto and San Francisco-based Skyfall AI, are developing an AI world model aimed at simulating the physical world. They want to prove that this technology can perform all of a company’s operational tasks with little or perhaps no need for humans.
“Our vision is to democratize technology and enable all businesses to operate as autonomously as possible.”
This week, Skyfall announced that its first act would be to acquire an existing business for up to $1 million and install an AI CEO. With backing from investors such as Inovia Capital and Garage Capital, Skyfall wants to use its AI world model to take over companies and run them fully autonomously, thereby demonstrating “enterprise super intelligence.” The company is currently acquiring “micro B2B-SaaS” businesses, and its goal is to double the acquired companies’ revenue in six months.
“Our vision is to democratize technology and enable every business to operate as autonomously as possible,” Pasparak said in an interview. beta kit.
This is a significant departure from Maluuba’s original mission of helping machines understand human language. Maluuba was founded in 2011 after the co-founders attended the University of Waterloo and worked at Velocity Garage. Maluuba was advised by Canadian AI pioneers and Turing Award winners Richard Sutton and Joshua Bengio. The company developed conversational AI technology to try to teach machines how to understand natural language. We partnered with consumer electronics giants Samsung, LG, and BlackBerry to build a demo that queries the entire Harry Potter book in the same way we currently query the entire Internet via ChatGPT.
Related: Maluuba team explains why language is the key to making machines intelligent
Part of the bet is precisely the dishonest attitude towards the dominance of AI models. Paspalak hopes to prove that the world model is better than the LLM in several characteristics important to human business leaders, such as long-term planning and high-stakes decision-making with uncertainty.
“Sam’s focus is on real-world applicability, rather than static benchmarks that reward a brute-force approach,” Karamdeep Nijjar, a partner at Inovia Capital, wrote in a LinkedIn post.
Skyfall brands itself as Frontier Neolabs, which Palpasak compared to the early days of OpenAI, but it is a for-profit company. “We believe that technology breakthroughs are absolutely necessary to unlock this vision, which is why we need a research-first organization for the first few years until we have a product that can be deployed at scale,” he said. beta kit. As part of the research process, the company will publicly document its progress in taking over corporate operations with AI.
In contrast to LLMs, which are trained on vast amounts of text data, world models are trained on physical, spatial, and motion data about the environment. This is like building an internal 3D representation of the world, which proponents argue allows the model to integrate cause and effect rather than just pattern matching. These models have been attracting a lot of attention lately. General Intuition, a global modeling startup in Silicon Valley, was recently valued at US$2.3 billion (CAD $3.2 billion) for a dataset of video gameplay clips it uses for robots and other physical AI.
The founder said he was partially inspired by a paper his team co-authored last year. The paper investigated the limitations of LLM when playing 90’s business simulator games. roller coaster tycoon. If the technology works as hoped, AI agents trained on the world’s models will be in charge of operational aspects of companies, from finance to product to marketing. But Pasparak acknowledges that there are human skills that even global models can’t replicate, such as managing relationships inherent in business.
RELATED: What Maluva’s defeat of Ms. Pac-Man means for AI research
Similar to other acquisition models such as private equity acquisitions, this structure will impact human employment. Mr. Paspalak said the first acquisitions would be owner-managed companies or companies with one to five contractors or employees who would be offered “generous severance packages” and support in finding new work within the network. He added that this also applies to large companies that Skyfall plans to acquire in the future.
The Skyfall experiment, in which machines take over human roles, appears to be a direct response to concerns expressed by labor market experts and the wider public about job losses resulting from the integration of AI. Big tech leaders like Anthropic’s Dario Amodei and Meta’s Mark Zuckerberg warned last year about mass white-collar job losses to AI, but have since softened their stance on the impending apocalypse. at the same time, quartz Economists are reportedly increasingly worried about job losses due to technology. More than 200 economists and researchers have signed an open letter warning of “massive job losses” and economic restructuring.
In Pasparak’s view, relying on AI to handle operational tasks will ultimately free up people to do more artistic and creative work. “People should be working on things they’re passionate about, and we don’t want them doing boring, monotonous operational tasks,” he said. The CEO likened AI to historic innovations like the printing press that eliminated the need for scribes and moved their jobs elsewhere (although some research suggests that AI may actually enhance jobs rather than reduce them).
If Skyfall (a James Bond-inspired name and a play on words for a secret AI agent) pays off, the total addressable market will be much larger than the current AI economy, which is primarily focused on automating software engineering, Paspalak argues. The opportunity, he claimed, is “in the trillions of dollars” as his company seeks to automate every part of every business on the planet.
Feature image courtesy of Skyfall AI.
