Recursive Builds AI That Builds AI With Flint Model — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Recursive developed an open-weight AI coding model named Flint using its recursive self-improvement system, the Eureka Machine, which automated most of the model’s development process.
- Richard Socher, CEO and cofounder of Recursive, stated that the Eureka Machine devised new training methods, model architectures, and chip-specific code optimizations, enabling a small team to produce a top-tier model in months.
- Flint, a 40-billion-parameter model, outperformed similar-sized models on coding tasks in internal benchmarks and ran efficiently on a consumer Mac with 48 GB of RAM.
- Recursive raised $650 million at a $4.65 billion valuation from Google Ventures, Greycroft, Nvidia, and AMD Ventures to advance its recursive self-improvement approach.
- Jeff Clune, a cofounder of Recursive and professor at the University of British Columbia, brings expertise in indefinite AI learning, reinforcing the company’s focus on self-improving systems.
- Recursive argues that fears of uncontrolled AI via recursive self-improvement are overblown, emphasizing human oversight of the Eureka Machine and rejecting calls for legislative bans on the technique.
Why it matters: Recursive’s demonstration of a small team achieving elite AI results through AI-automated development challenges the need for massive labs and compute budgets, making advanced model creation more accessible while reigniting debate over the risks of self-improving AI systems that Congress may seek to regulate.
Ask SkimNews
