Adaptation Labs will focus on new approaches to effective AI.
Former executives at Toronto-based Cohere and its research division are launching a startup aimed at creating powerful artificial intelligence (AI) that can “adapt and learn continuously.”
“This will change the way everyday users and companies around the world can create and own AI.”
Sarah Hooker
Former Cohere Labs CEO Sarah Hooker will lead the new venture along with her former Cohere colleague Sudip Roy, who was the company's senior director of insights.
According to one of the researchers, Adaptation Labs will work on “adaptable intelligence.” publish on X from Hooker. In a separate Post on LinkedInShe said the company will lead an “extremely effective adaptation” that will allow AI to evolve in real time. “This will change the way everyday users and companies around the world can create and own AI,” she wrote.
In his post, Roy wrote that he was taking on a new challenge “at the intersection of efficiency and adaptability.” He said the next breakthrough in AI “will not just involve larger models – it will involve systems that can adapt efficiently without the computational cost of traditional gradient-based methods.”
Gradient methods such as gradient descentare machine learning methods widely used to train large language models (LLMs). They involve optimizing the parameters of a mathematical model to obtain the smallest number of possible errors. Other approaches to AI training do not use gradients. One example is evolutionary algorithmsa computational method inspired by natural selection, where solutions to a mathematical problem are represented as individuals in a population.
Industry standard approaches to training AI models, such as those used by OpenAI and Google, require a huge amount of effort. data and energy. Adaption Labs appears to be looking to improve the efficiency of AI development to reduce energy consumption and costs. Some job postings mention “serious computing budgets” that the company says will force it to innovate and collaborate across its technology stack.
CONNECTED: Marzi Fadai appointed head of Cohere Labs
Although headquartered in California, Adaption Labs is hiring for several hybrid positions in Canada and the UK, including Brand Designer, Design Engineer and Performance Research Engineer.
After spending five years as a research scientist at Google DeepMind, Hooker joined Cohere in 2022 to lead Cohere Labs. Earlier this summer she announced she was leaving but staying until September to oversee applications for the Cohere Fellows program. Following Hooker's departure, Cohere Labs appointed researcher Marzi Fadai to lead the division.
Hooker led research at Cohere Labs on several fronts, including multilingual capabilities, inference efficiency, and LLM consistency and security. Cohere's Aya series of models has provided LLM support for 55 new languages, she said.
The researcher also co-authored paper this spring pointing out the supposed unreliability of the popular chatbot leaderboard LM Arena's rankings, which drew much applause but some backlash from the artificial intelligence research community.
For his part, Roy joined Cohere in 2021 after spending seven years at Google as a research scientist. Before becoming the Senior Director of Insights for a Canadian startup, he was a CTO. According to his LinkedIn, Roy will be the CTO of Adaption Labs.
Cohere, founded in 2019 by former Google researchers, creates LLM programs that are used in chatbots and applications for businesses and governments. It is one of Canada's best-funded and most valuable private technology companies, valued at US$7 billion.
Artistic image courtesy of Sudip Roy via LinkedIn and Sarah Hooker via LinkedIn.
