This paper presents Dream-RSI, a framework that lets AI agents improve their exploration strategies by learning from past discovery attempts without constantly running expensive new experiments. The system uses historical search data as a simulator to test and refine exploration policies cheaply before deploying them back into the real search space.

• Solves the exploration bottleneck in recursive self-improvement by creating a replay simulator from accumulated discovery history, enabling off-policy feedback without costly online re-evaluation
• Keeps the exploration layer separate from the underlying agent, making strategies explicit and programmable while maintaining compatibility with existing systems
• Demonstrates competitive or better results across algorithm engineering, mathematical optimization, and GPU kernel engineering while substantially cutting discovery costs

Higgsfield launched Supercomputer 2.0, an enterprise marketing automation agent built on NVIDIA’s Agent Toolkit that handles ideation, creative production and posting in one interface. It orchestrates over 35 image, audio and video models with policy guardrails and permission controls, claiming adoption by 78% of Fortune 500 firms and 12,000 businesses worldwide. To prove its speed, the startup used the platform to produce a 95-minute AI-generated film in just 14 days.

• Higgsfield's new NVIDIA-built agent claims 78% of Fortune 500 companies as users, but that figure is self-reported with no third-party audit.
• A 15-person team used the platform to make a 95-minute AI film (Hell Grind) in 14 days for under $500,000, which premiered at Cannes.
• Despite McKinsey estimating agentic AI could handle two-thirds of marketing work, fewer than 10% of CMOs have actually deployed full end-to-end AI workflows, highlighting a gap between hype and adoption.
• Higgsfield faces competition from well-funded rivals and especially Meta's Advantage+, which already reaches 8 million advertisers.