Google’s Dream-RSI recursively improves an AI discovery agent’s exploration policy—the rules for branching, parallel searches and stopping—not the coding model’s weights. In the reported experiments, a policy-development agent tests revised exploration strategies against outcomes already recorded in earlier searches, then deploys a selected policy for another online run. It is a research demonstration of recursive improvement at the orchestration layer, not evidence that an AI system autonomously rewrote its own model or solved recursive self-improvement generally.
What does Dream-RSI change?
The paper, Dream-RSI: Recursive Self-Improvement through Evolving Worlds, describes a discovery system with two distinct roles. A coding agent proposes candidate solutions; an evaluator measures their outcomes. An exploration policy directs how the discovery process searches: which branches to pursue, how to group work in parallel and when to stop.
As an Amazon Associate I earn from qualifying purchases.
Dream-RSI changes that policy over successive rounds. The authors say the underlying coding agent remains unchanged, and the project repository’s “zero gradient steps” description refers to the absence of coding-agent weight updates in the reported method. That distinction matters: the system recursively improves how it uses the agent, rather than training the agent itself. The paper and official repository describe the work and its experimental framing.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →How the recursive improvement loop works
- Explore online. The current policy directs the discovery agent’s search. The system records the decisions, candidate proposals and evaluated results as a tree.
- Replay recorded search history. The stored tree becomes a simulator of the branches the discovery process actually reached. A candidate policy can select and order recorded branches, alter parallel groupings and choose when to stop.
- Test policy changes against that history. A policy-development agent edits exploration-policy code and scores candidate policies using outcomes already in the tree. For those replayed branches, this avoids calling the expensive discovery agent and evaluator again.
- Deploy and extend the record. A selected policy directs a new online run. Its results add new history to the pool that later policy-development rounds can replay.
The project page describes the discovery tree as “an exact simulator of the search space it reached.” The qualification is important: it is exact for the reached, recorded search space, not a complete model of every possible solution or branch in the underlying problem. The project page presents the authors’ explanation and links to the paper and project materials.
#1 Best Overall
- EVOLUTION AMD RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
- AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
- AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 128GB pool, which is perfect for running LLMs such as Deepseek 70B Q8, which runs comfortably on this machine.
- EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 12% better performance in digital content workloads.
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
What the reported experiments show
The paper evaluates eight scientific-discovery tasks across algorithm engineering, mathematical optimization and GPU kernel engineering. Its reported results are comparisons on specified tasks, with particular agents, baselines and budgets; they do not establish a universal advantage for every discovery problem. The authors’ paper reports the experiments and their task-specific results.
Algorithm engineering
In a Lasso path solver experiment, the authors report lower downstream runtime while using fewer discovery-agent calls than recursive fixed exploration. Against SimpleTES, they report up to 162 times fewer calls. The repository also highlights a particular comparison reporting 1.22 times faster downstream runtime and 1.74 times less discovery compute. These headline values belong to the stated experiment and baseline; they are not general performance guarantees.
Rank #2
- Built for Local AI Development: AMD Ryzen AI Halo is designed for local AI development and inference, featuring 128GB unified memory and support for up to 200B parameter models to build and run intensive AI workloads locally.
- 128GB Unified Memory: Features 128GB LPDDR5x unified memory at 8000 MT/s with 256 GB/s memory bandwidth, providing a shared memory pool across the CPU, GPU, and NPU to support larger AI models.
- AMD Ryzen AI Max+ 395 Processor: Features 16 cores, 32 threads, and Zen 5 architecture, paired with AMD Radeon 8060S integrated graphics featuring 40 RDNA 3.5 compute units and an AMD XDNA 2 NPU with up to 50 TOPS.
- Linux AI Developer Platform: Purpose-built for Linux-based AI development with full AMD ROCm software support and preloaded tools, models, and workflows optimized for local AI development.
- Compact, Connected Design: Includes a 2TB M.2 SSD, 10GbE LAN, Wi-Fi 7, Bluetooth 5.4, USB-C connectivity, and HDMI 2.1b.
Mathematical optimization
The evaluated problems include Sum-Difference, Autocorrelation and Circle Packing. The paper reports that Dream-RSI matches or exceeds selected strong baselines on two tasks and remains competitive on Autocorrelation. In a cited comparison, it uses fewer than 1,000 generations where SimpleTES uses 51,200. This is a task- and baseline-specific result, not a ranking across mathematical optimization as a whole.
Recommended Free Tools
GPU kernel engineering
On four KernelBench tasks, the authors report reaching comparable performance with fewer generations on VGG16 and LayerNorm, and higher performance under comparable budgets on ConvDiv and ConvMax. In selected comparisons, they report up to 2.09 times higher performance and up to 2.43 times fewer generations to reach comparable performance. The figures apply to those kernel tasks and experimental budgets, not GPU workloads generally.
Rank #3
- EVOLUTION RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
- AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
- AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 128GB pool, which is perfect for running LLMs such as Deepseek 70B Q8, which runs comfortably on this machine.
- EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 12% better performance in digital content workloads.
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
What replay can—and cannot—tell the system
Replay gives a policy-development process inexpensive feedback about different ways of navigating known branches. It cannot supply the actual outcome of a genuinely novel branch that was never explored and therefore has no recorded result. The usefulness of the simulator is consequently bounded by the breadth and quality of the search history available to it.
The paper’s guarantee that a retained policy scores no worse than the incumbent applies to the replay data used for selection. It does not guarantee that the next online run will improve, that replay scores reliably predict future performance, or that gains will transfer to tasks beyond those tested. The authors report empirical online results on their test tasks, but do not establish a general theorem connecting replay score to future online gains.
How to interpret the paper’s significance
Dream-RSI is evidence that a discovery system can use its own recorded search history to iteratively improve its search strategy, then test that strategy in further online work. That is a meaningful form of recursive improvement—but at the meta-exploration layer. It is not, based on the reported method, a coding model that modifies its own weights, a system with a complete simulator of its problem domain, or proof that recursive self-improvement is solved.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →When judging its comparisons with fixed exploration or another discovery system, check whether the methods use the same task, agent, evaluator, initialization and budget. Also distinguish the outcome being measured: solution quality, online calls or generations, compute, downstream runtime, or kernel performance. A result can be favorable on one axis without establishing an advantage on all the others.
Paper and code availability
The preprint is titled Dream-RSI: Recursive Self-Improvement through Evolving Worlds, by Tong Zheng and coauthors, with affiliations listed for Google, Google DeepMind, the University of Maryland, College Park, and the University of Virginia. The arXiv record dates it September 14, 2026. The official repository says the full codebase, discovered programs and reproduction scripts are being prepared for release; its current materials should not be described as a complete independent reproduction package. Read the arXiv record or visit the project repository.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
