Tag
This paper introduces SIA, a self-improving AI loop that combines scaffold rewriting and weight updates (via LoRA) to enhance task performance. Tested on three diverse tasks, it outperforms setups using only scaffold improvements.
Sakana AI launches RSI Lab in Tokyo, dedicated to recursive self-improvement (RSI) where AI builds AI, aiming to achieve self-improvement without unlimited computational resources.
Researchers from HKUST, ByteDance, and UCL propose SCORE, a co-evolutionary training framework that jointly trains an LLM as both a deep research report generator and an evaluator, using a meta-harness to dynamically adjust evaluation difficulty and prevent reward saturation. Experiments show consistent improvement in open-ended research report quality.
HexoAI releases SIA, an open-source self-improving AI that recursively improves its abilities to achieve any goal.
Marco Oram shares his exploration of automating LLM fine-tuning using Fireworks AI Agent, fine-tuning a small Qwen model to integrate with his PaperWiki project, demonstrating a step toward self-improving AI.
Recursive Superintelligence raised over $650 million at a $4 billion valuation to develop AI that can improve itself with minimal human involvement, backed by prominent researchers from leading AI companies.
Recursive Superintelligence, a new startup aiming to automate knowledge discovery with self-improving AI, has launched with over $650 million in funding led by GV and Greycroft.
Today we launch Recursive, an AI company focused on recursive self-improvement and knowledge discovery to advance science and technology.
Recursive, an AI startup founded by former research leaders from OpenAI, DeepMind, and others, emerged from stealth with a $650M funding round to develop recursively self-improving AI through open-ended scientific discovery, aiming for superintelligence.
SIA is a self-improving AI framework that uses a meta-agent, target agent, and feedback agent to autonomously improve performance on benchmark tasks, achieving significant gains on LawBench, GPU kernel optimization, and single-cell RNA denoising.