Sakana AI's Recursive Self-Improvement (RSI) Lab

I think the most impressive thing about Sakana.ai is their relentless pursuit of whatever is hype right now.

Genuinely it take a lot of work and talent to be this hype-motivated and completely ignore anything except what is popular on X at any given time.

Note: RSI is an incredibly important topic -- I just don't care to listen to Sakana on this matter -- they are the epitome of "hypebeast" https://www.urbandictionary.com/define.php?term=hypebeast

(Thanks for sharing hardmaru)

I heard about the ShinkaEvolve on a podcast where the guest had used it to evolve an agent harness for a less capable model.

I ended up borrowing the ideas from it for one of my own personal projects.

I don't know how RSI aligns with DPI ("Data Processing Inequality", which states that, basically, unless you have an infinite supply of real data, you will suffer from model collapse). Models can't keep improving themselves infinitely. See, for example, https://arxiv.org/html/2601.05280v2

[dead]

Fortunately a rational society like Japan is not as interested in outsourcing their capacity to curve fitting models as other societies.