Top researchers leave tech giants to launch their own start-ups
Top researchers are leaving major tech companies such as Google, Meta and OpenAI to launch their own AI ventures. These start-ups are attracting significant investment from venture capitalists who believe in the potential of new approaches to AI model architecture.
Intense competition between leading AI labs has created opportunities for smaller, more agile companies. Investors see value in these start-ups because founders have often gained unique insights by working on the front line of AI development. They understand what works at scale and can pinpoint areas where innovation is lacking within established organisations.
The focus on commercial goals within large AI labs can limit researchers’ freedom to explore new ideas. Smaller start-ups offer room for more exploratory research, especially outside the dominant paradigm of large language models (LLMs).
Recursive Intelligence, founded by former employees of Anthropic and Google DeepMind, is developing AI tools for chip design, for example. The founders realised that chip manufacturers would be more likely to trust a neutral partner than a competitor such as Google. They assembled a team with expertise from various companies, including Google, Anthropic, Nvidia, Apple and xAI.
Periodic Labs, founded by former OpenAI and DeepMind employees, focuses on the development of autonomous labs. This reflects growing doubts within the AI research community about whether simply scaling up existing LLM approaches will be enough for future breakthroughs.
AMI Labs, founded by Yann LeCun, former AI chief at Meta, believes that current AI models excel at generating content but struggle with real-world applications that require grounding, causality and reliable behaviour. As AI moves off the screen and into sectors such as robotics and healthcare, tackling these limitations is becoming crucial.
Ineffable Intelligence, which recently raised a record $1.1 billion (around €1 billion) in seed funding, focuses on reinforcement learning, where AI models learn through experience rather than relying solely on human-provided data.
This approach contrasts with many leading AI models that have been trained on vast amounts of internet text. Humans&, another San Francisco-based start-up founded by former Anthropic and xAI employees, is also exploring reinforcement learning.
Business AM
Subscribers 0
Fans 0
Followers 0
Followers