Dream-RSI: Recursive Self-Improvement through Evolving Worlds • Hacker News Board

Re: Dream-RSI: Recursive Self-Improvement through Evolving Worlds

#31

Post by computably » Wed, Sep 16, 2026, 5:21 PM UTC

ctolsen wrote: post #23

EthanHeilman wrote: post #15

Earlier quoted context omitted.

Does RSI actually mean anything specific anymore? RSI, AGI, at this point seem like buzzwords. Sure AGI has definition that are measurable, say "better than 95% of humans on 95% of intellectual tasks" but if we used that definition we already have AGI and almost no one thinks we have achieved AGI. We use AIs to train AIs which we use to train AIs, why is that not RSI? How much human intervention means that is not RSI…

So few terms in AI are well defined. We will get ASI via AGI because of RSI but neither of those three things have any definition except pure vibes. I struggle with the argument that RSI doesn't already exist like you say, it's existed since before the term LLM (hey, one that can be defined!) was common parlance. Though the biggest use for those is not superintelligence, it's to serve you ads and get your kids addict…

Narrow pre-LLM models that spit out content and ad recommendations have never been capable of also suggesting, let alone implementing, self-improvements.

Re: Dream-RSI: Recursive Self-Improvement through Evolving Worlds

#32

Post by ex-aws-dude » Wed, Sep 16, 2026, 5:22 PM UTC

eggbrain wrote: post #28

Perhaps I'm not understanding it correctly, but here's my take on what the paper is doing. Imagine you have a problem you want to solve (let's say, identify an OCR'd handwritten character, e.g. the MNIST Dataset). You tell 3 agents "Hey, each of you take a stab at getting really good at recognizing characters from this dataset. You can take 10 refinement steps to continue to improve ". You can't give each agent unlim…

Isn't that a challenge with RL anyway that for a lot of problems its hard to even know accuracy continuously for each step

Re: Dream-RSI: Recursive Self-Improvement through Evolving Worlds

#33

Post by Tycho » Wed, Sep 16, 2026, 5:24 PM UTC

I don’t really understand this. Some real examples would help. Making a “simulator” out of a bunch of historic states does not tell me enough.

Re: Dream-RSI: Recursive Self-Improvement through Evolving Worlds

#34

Post by computably » Wed, Sep 16, 2026, 5:24 PM UTC

EthanHeilman wrote: post #15

rybosworld wrote: post #6

Unless I'm misunderstanding, calling this RSI seems misleading? This looks like an optimization of current training methods, and a good one, but not "RSI" in the sense of a system that can perpetually improve itself forever.

Does RSI actually mean anything specific anymore? RSI, AGI, at this point seem like buzzwords. Sure AGI has definition that are measurable, say "better than 95% of humans on 95% of intellectual tasks" but if we used that definition we already have AGI and almost no one thinks we have achieved AGI. We use AIs to train AIs which we use to train AIs, why is that not RSI? How much human intervention means that is not RSI…

> say "better than 95% of humans on 95% of intellectual tasks" but if we used that definition we already have AGI and almost no one thinks we have achieved AGI

What matters isn't 95% of humans, it's 95% of actual professionals. Benchmarking an AI accountant against people with zero accounting experience is worse than worthless.

Re: Dream-RSI: Recursive Self-Improvement through Evolving Worlds

#35

Post by embedding-shape » Wed, Sep 16, 2026, 5:24 PM UTC

eggbrain wrote: post #28

Perhaps I'm not understanding it correctly, but here's my take on what the paper is doing. Imagine you have a problem you want to solve (let's say, identify an OCR'd handwritten character, e.g. the MNIST Dataset). You tell 3 agents "Hey, each of you take a stab at getting really good at recognizing characters from this dataset. You can take 10 refinement steps to continue to improve ". You can't give each agent unlim…

Intuitively I wouldn't readjust how many steps they each do, but instead add another run afterwards, that get the same amount of steps as the previous, but now also with a concise description of what the previous attempts did and what they achieved, and ask it to improve. The amount of compute you have available, would dictate how many full iterations of this "fan out search > consolidate" workflow you can do.

Re: Dream-RSI: Recursive Self-Improvement through Evolving Worlds

#36

Post by dmcrespo » Wed, Sep 16, 2026, 5:35 PM UTC

Calling this paper "Dream" seems a bit speculative to me. The idea is interesting and reminded me somewhat of karpathy's work at https://github.com/karpathy/autoresearch

Re: Dream-RSI: Recursive Self-Improvement through Evolving Worlds

#37

Post by jonbaer » Wed, Sep 16, 2026, 5:42 PM UTC

Have Sutton and Barto chimed in on "RSI" yet? Does anyone know?

Re: Dream-RSI: Recursive Self-Improvement through Evolving Worlds

#38

Post by eggbrain » Wed, Sep 16, 2026, 5:45 PM UTC

embedding-shape wrote: post #35

eggbrain wrote: post #28

Perhaps I'm not understanding it correctly, but here's my take on what the paper is doing. Imagine you have a problem you want to solve (let's say, identify an OCR'd handwritten character, e.g. the MNIST Dataset). You tell 3 agents "Hey, each of you take a stab at getting really good at recognizing characters from this dataset. You can take 10 refinement steps to continue to improve ". You can't give each agent unlim…

Intuitively I wouldn't readjust how many steps they each do, but instead add another run afterwards, that get the same amount of steps as the previous, but now also with a concise description of what the previous attempts did and what they achieved, and ask it to improve. The amount of compute you have available, would dictate how many full iterations of this "fan out search > consolidate" workflow you can do.

In the paper (section 5.1), they actually tried to abstract high level directional insights into the prompt in order to see if that helped, and they basically found it underperformed a prompt that didn't have those insights at all, implying that directional guidance perhaps over-constrains things.

Re: Dream-RSI: Recursive Self-Improvement through Evolving Worlds

#39

Post by bob1029 » Wed, Sep 16, 2026, 5:53 PM UTC

There is no way this could be reasonably framed as RSI.

This iterative, online optimization of an exploration policy is not recursively intelligent in any way. It simply reallocates the available computational resources to more promising (hopefully) parts of the search space as system conditions change over time.

Re: Dream-RSI: Recursive Self-Improvement through Evolving Worlds

#40

Post by tiffanyh » Wed, Sep 16, 2026, 5:55 PM UTC

eggbrain wrote: post #28

Perhaps I'm not understanding it correctly, but here's my take on what the paper is doing. Imagine you have a problem you want to solve (let's say, identify an OCR'd handwritten character, e.g. the MNIST Dataset). You tell 3 agents "Hey, each of you take a stab at getting really good at recognizing characters from this dataset. You can take 10 refinement steps to continue to improve ". You can't give each agent unlim…

Great explanation.

Do you think this could be extrapolated to areas with no objectively verifiable results / outcomes?

(Outside of math & science)