self-improving-ai
(5)
Deep Cogito
Deep Cogito is a San Francisco AI research company building open-source large language models with hybrid reasoning - the same model can answer instantly or stop to reason before responding. Founded in 2024 by former Google engineer Drishan Arora, the company trains its Cogito models with a method called Iterated Distillation and Amplification (IDA), which lets a model improve by learning from its own reasoning rather than from human labels. Its stated goal is general superintelligence, and its flagship 671B open model competes with the strongest open systems from DeepSeek and Meta.
Axiom
Axiom (Axiom Math) is a Palo Alto-based AI lab building a self-improving, superintelligent reasoner - starting with an AI mathematician. Its AxiomProver system invents new theorems, proves them formally in Lean, and verifies them before they reach a human, aiming to give AI the one thing language models lack: provable correctness. Founded in 2025 by mathematician Carina Hong, the company has raised $264M and claims machine-generated proofs accepted by peer-reviewed journals.
Carina Hong
Carina Hong is the founder and CEO of Axiom Math, an AI startup building an autonomous "AI mathematician" that can generate and formally verify its own proofs. A Guangzhou-born first-generation college student, she finished dual MIT degrees in math and physics in three years, won the Morgan Prize, became a Rhodes Scholar at Oxford, then left a Stanford PhD-and-JD track to start Axiom in March 2025. Her thesis is blunt: "Math is AGI." Within roughly nine months the company raised about $264M total, reached a reported $1.6B valuation, and pulled researchers from Meta FAIR, Meta GenAI, and Google Brain/DeepMind.
Tim Shi
Tim Shi is the co-founder and former CTO of Cresta, the AI platform for contact centers that grew to over $100M ARR and raised $401M from Sequoia, a16z, and Greylock. A Tsinghua CS graduate who did early AI research at OpenAI alongside Andrej Karpathy — including the 'World of Bits' paper on web-based reinforcement learning agents — he co-founded Cresta in 2017 with Zayd Enam after both dropped out of Stanford's AI PhD program. In 2025, he co-founded Recursive Superintelligence, which emerged from stealth with $650M at a $4.65B valuation to build self-improving AI systems.
ZeroEval
ZeroEval is a New York-based AI startup from Y Combinator's Summer 2025 batch building an auto-optimizer for AI agents. Founded by Jonathan Chavez and Sebastian Crossa - two friends who met in college in Mexico - the platform captures every interaction your AI agent makes, scores quality with custom LLM judges, and automatically turns real production data into better prompts. The result: agents that get smarter after launch without manual intervention. Trusted by DoorDash, Datadog, Hugging Face, and Harvard Medical School, ZeroEval closes what the founders call 'the last mile reliability gap' in agentic AI.