From the American Enterprise Institute, March 19:
When I wrote about recursive artificial intelligence last September, I described an innovation still in its early stages, powerful in concept, but not fully realized in practice. Six months on, AI has made meaningful strides toward true recursivity. The acceleration has, well, accelerated.
A quick refresher: recursive AI refers to systems that can improve themselves, testing their own outputs, identifying weaknesses, and generating better versions without waiting for human researchers to do the work. The concern was twofold: The systems would be powerful and so fast that communities, individuals, and workers would have little time to adapt.
Here is what has changed since then.
A brief digression into the history of theoretical mathematics helps illustrate how recursive AI is developing in real time. In 1969, Volker Strassen discovered a faster algorithm for matrix multiplication, the core computational operation underlying modern AI systems. His insight reduced the arithmetic steps required: Seven scalar multiplications instead of eight at each recursive step, a seemingly small gain that compounds across billions of calculations. The practical record set by that work went essentially unchallenged for 56 years.
Google DeepMind’s AlphaEvolve, announced in May 2025, finally surpassed it. Rather than relying on flashes of insight from human researchers, AlphaEvolve uses automated evaluators that generate and test thousands of candidate solutions, selecting the best and iteratively improving them. The result: a new algorithm for multiplying 4×4 complex-valued matrices using 48 scalar multiplications, one fewer than Strassen’s long-standing benchmark of 49. That single-digit improvement had eluded mathematicians for more than half a century.
The mechanism AlphaEvolve uses is a more powerful version of something now available to anyone working with AI-assisted code. I recently started using a similar approach in my own workflow automation, and the only word for it is “magical.” You provide the AI with existing code and instruct it to fix and verify the result. The system iterates until it has a working product. This is AlphaEvolve in miniature.
So what happens when the frontier AI labs start applying these self-guided recursive systems to their own research and engineering work? METR, which monitors progress in advanced AI models, has been tracking the answer. Its original March 2025 study found that the duration of tasks AI agents can complete autonomously has roughly doubled every seven months over the prior 6 years.
More recent data suggests that the pace has accelerated: in 2024 and 2025, the doubling time shortened to approximately four months. By late 2025, the most capable models were reliably completing tasks complex enough to require five hours of skilled professional work. This isn’t a speed question but a measure of the difficulty of challenges AI is able to take on. As a consequence, the time between major improvements in model performance is shrinking....
....MORE
And May 13 we see, via India's Office Chai:
Recursive Raises $650 Million At $4.65 Billion Valuation To Create Self-Improving AI
Brand-new startups are continuing to raise massive rounds to work on new niches of AI.
Recursive, a stealth-mode AI lab, has emerged from the shadows with one an audacious pitch: build an AI that improves itself — endlessly, and without human intervention. To back that bet, the company has raised $650 million at a $4.65 billion valuation in a round led by GV (Google Ventures) and Greycroft, with participation from AMD Ventures and NVIDIA.
The round is part of a broader surge in mega AI funding that has defined the past two years, as investors pour capital into teams chasing the next frontier of intelligence — beyond scaling laws, beyond human-in-the-loop development.
The Founding Team
Recursive was founded by former team leaders from OpenAI, Google DeepMind, Meta AI, Salesforce AI, and Uber AI. Richard Socher, one of the most-cited researchers in AI history, anchors the founding team. He previously served as Chief Scientist and EVP at Salesforce, where he built and led the company’s entire AI research and product stack. Before Salesforce, he founded MetaMind, which Salesforce acquired. He is widely credited with bringing deep learning into natural language processing and pioneering word vector representations that underpinned today’s large language models.
Tim Rocktäschel, co-founder and research lead, is a professor of AI at University College London and former Director and Principal Scientist at Google DeepMind, where he ran the Open-Endedness research group. His team won the ICML 2024 Best Paper Award for Genie, an interactive world model capable of generating playable environments from a single image.
The co-founders are joined by former OpenAI researchers Jeff Clune, Josh Tobin, and Tim Shi. Clune is a pioneer in evolutionary algorithms and open-ended AI systems; his Darwin Gödel Machine work at Sakana AI demonstrated that AI agents could autonomously rewrite their own code to improve benchmark performance. Tobin built OpenAI’s robotics capabilities and later co-founded Gantry, an ML monitoring startup.
In total, Recursive has over 25 people and is growing fast. The team includes researchers in open-ended algorithms, quality diversity algorithms, AI-generating algorithms, self-improving coding agents, automated red teaming, prompt engineering, world models, vision transformers, and retrieval-augmented generation.
What Makes Recursive Different
Most AI labs today are racing to build bigger, better foundation models — and that race requires armies of human researchers to design experiments, curate data, run evaluations, and decide what to work on next. Recursive’s thesis is that this human bottleneck is the real speed limit on AI progress.
The company’s approach draws a direct parallel to evolution: just as Darwinian processes produced intelligence through an open-ended archive of interestingly different discoveries — building from replicating molecules to sight, language, and science — Recursive wants to replicate this dynamic in software. The system would grow its own archive of innovations, with each discovery enabling the next, in a loop that has no ceiling.
Socher has called this the “third and perhaps final stage of neural networks” — a system that automates evaluation, data selection, training, post-training, and even research direction itself. Where existing labs like OpenAI, Anthropic, and Google DeepMind have explored aspects of automated research, none has organized an entire company around recursive self-improvement as its core commercial thesis....
....MUCH MORE
If interested see also:
January 28 - So it Begins: "Silicon Valley Wants to Build A.I. That Can Improve A.I. on Its Own"
The headline at TechCrunch was "AI chip startup Ricursive hits $4B valuation two months after launch"
Serious money believes these women are on to something.
May 12 - "AI Is Starting to Build Better AI"
Not there yet but some very smart people think it's close.
