Kitchener-Waterloo-based AI research startup Transformer Lab has launched what co-founder and CEO Ali Asaria said is \u201cthe most ambitious thing I\u2019ve ever been a part of.\u201d<\/p>\n
For the past two years, Transformer Lab\u2019s four-person team has been quietly working to fully automate the research process.<\/p>\n
\u201cWe see this as the next revolution, the one that actually replaces what we currently call the AI revolution.\u201d<\/p>\n
Today, Transformer Lab announced it has opened access to the fruit of that labour, Primus, an autonomous research tool Asaria claims can connect users with AI agents capable of producing \u201cPhD-level\u201d research in a small fraction of the time as their human counterparts.<\/p>\n
A global shortage of machine learning (ML) engineers has constrained AI progress; Asaria hopes Primus can fix that, and more.\u00a0<\/p>\n
\u201cIf you work in a neolab, but you can\u2019t hire the best [ML] engineers because there\u2019s only so many of them in the world, and most of the best ones are going to these few companies, well, now you have access to as many of them as you want at the touch of a button,\u201d Asaria told BetaKit in an interview.<\/p>\n
Asaria teamed up with co-founder Tony Salomone, who worked at Asaria\u2019s previous retail tech startups Tulip and Well.ca, to build Transformer Lab in 2024. The AI research lab officially launched earlier this year with funding from a group that includes Toronto\u2019s Ripple Ventures and Kitchener-Waterloo\u2019s Garage Capital.<\/p>\n
Since forming, Transformer Lab has been spending time with researchers to understand their processes step-by-step and automate them.<\/p>\n
Primus projects run dozens or hundreds of different agents, which each use different AI models appropriate to their respective tasks, Asaria said. Some are tiny, while others are frontier models from big-name providers. The company is also training models of its own.<\/p>\n
Transformer Lab claims Primus can turn a single prompt-based research question from a hypothesis to a completed research paper in \u201chours to days\u201d without additional human direction.<\/p>\n
Using Primus, Transformer Lab claims that it has produced and published 30 \u201cMasters to PhD-level\u201d research papers in 30 consecutive days across domains ranging from materials science to physics, the interoperability of large language models, seismology, protein design, audio generation, and 3D vision. While Transformer Lab did not provide more detail on how these papers were assessed to be at that level, one of them has already received a citation in published research by the Google DeepMind team.<\/p>\n
With Primus, Asaria hopes to play a part in ensuring more than just a few tech giants own the AI future, and accelerate the pace of research across all sorts of scientific domains\u2014beyond just ML. \u201cWe see this as the next revolution, the one that actually replaces what we currently call the AI revolution,\u201d he said.<\/p>\n
RELATED: Ali Asaria\u2019s Transformer Lab launches worldwide<\/p>\n
Transformer Lab has made Primus free to start, and is rolling it out to the public on a waitlisted basis, admitting researchers and developers as AI compute capacity permits.<\/p>\n
\u201cWe\u2019re a small team, and we have plenty of ideas that get shelved because we lack the resources to execute them,\u201d Rubric Labs co-founder and CEO Sarim Malik said in a news release. \u201cWith Primus, it feels like adding a team of world-class, senior ML engineers to your roster overnight who can chip away at your ideas.\u201d\u00a0<\/p>\n