recursive self-improvement
AI that improves AI: the loop where a model writes the code or the data that makes the next one.
FilterEveryone, all time25 from 16 people
- RN
Ramez Naam quoted
Author and clean-energy investor who has been publishing…
Our readingTo achieve runaway recursive self-improvement, conceptual breakthroughs are likely required.
Their wordsBut to get very large benefits, and runaway RSI, I expect we’ll need conceptual breakthroughs.
↗Can a Million AI Agents Create Superintelligencerameznaam.com
- 3 weeks earlier
- RN
Ramez Naam quoted
Their wordsGiven our best current data, the AI self-improvement loop would need to be roughly 5–10× stronger to sustain itself, let alone run away.
- RN
Ramez Naam quoted
Their wordsI do expect autonomous self-improvement to arrive at some point. I’m skeptical that it leads to Type 5 - runaway super-intelligence - without a major conceptual breakthrough.
- 3 days earlier
- MY
Matthew Yglesias quoted
Writes Slow Boring, a near-daily newsletter on American…
Their wordsWe have always known that recursive self-improvement is dangerous.
- 2 days earlier
- TC
Tyler Cowen quoted
Economist at George Mason University
Their wordsThe policy creates a big incentive for some other parts of the world to invest a lot more in compute.
↗Banning self-recursive improvement in AI models?marginalrevolution.com
- 3 days earlier
- NL
Nathan Lambert quoted
Machine-learning researcher on open language models
Their wordsRSI is poised to make modern LLMs vastly cheaper. Trends that have shown LLMs get exponentially cheaper at a given intelligence are likely to accelerate.
↗Why I still haven’t bought into true RSIinterconnects.ai 2nd of 3 in this piece
- 5 days earlier
- ZM
Zvi Mowshowitz quoted
Writes Don't Worry About the Vase
Their wordsInside Anthropic and OpenAI, internal models are improving even faster. Over the summer there was a step change, as Mythos and Astra started kicking off the early stages of recursive self-improvement (RSI).
- 1 day earlier
- ZM
Zvi Mowshowitz quoted
Our readingRecursive self-improvement has already started, because AI research itself is now being automated.
Their wordsOpenAI and Anthropic have entered the RSI era. OpenAI Research Progress Is Accelerating Due To OpenAI Research Progress OpenAI has been opening up lately about many things. The most important issue of all is that of the acceleration and automation of AI R&D.
- 7 days earlier
- JP
Jakub Pachocki quoted
Chief Scientist at OpenAI
Our readingThe current speed of AI progress could be sustained all the way into recursive self-improvement.
Their wordsBased on internal results, I have a strong expectation that this speed of progress could be sustained into recursive self-improvement.
- 5 days earlier
- TL
Tobias Lütke quoted
Co-founder and CEO of Shopify
Their wordsTraining tiny models for special purpose use cases works so incredibly well if you have a great self improving recursive flywheel.
- 8 days earlier
- JC
Jack Clark quoted
Co-founder and head of policy at Anthropic
Their wordsThe lesson here is that we as a species might write a bunch of gold-label helper systems, like Hawkeye, and then machines will use this to bootstrap above and beyond our own capabilities.
↗Import AI 470: No rights for machines; automating environment generation with SPADE; and building better GPU kernels with Hawkeyejack-clark.net 2nd of 3 in this piece
- 7 days earlier
- JC
Jack Clark quoted
Their wordsMy guess is we'll reach human parity on DiG-bench by middle of 2027, at which point we should expect things like recursive self-improvement to seriously kick off.
↗Import AI 469: Science AI; RSI simulator; and Zuck’s technological pessimismjack-clark.net 1st of 3 in this piece
- JC
Jack Clark quoted
Their wordsThe better AI systems get at this, the more likelihood we can assign to the idea that AI systems will imminently become capable of building themselves.
↗Import AI 469: Science AI; RSI simulator; and Zuck’s technological pessimismjack-clark.net 2nd of 3 in this piece
- 6 days earlier
- RG
Ryan Greenblatt quoted
Chief scientist at Redwood Research
Our readingOnce AI can do AI research, a single year should deliver four or five years worth of AI progress.
Their wordsMaybe my sort of median expectation is something like uh four or five years of AI progress in a single year.
↗Ryan Greenblatt – What happens once AI can automate AI research?youtube.com 1st of 26 in this recording
- RG
Ryan Greenblatt quoted
Their wordsI would say that I expect like full automation of AR&D perhaps somewhere around like 2031 2030 and then getting to like the like beats all humans on the job milestone. Maybe I expect median around 2033
↗Ryan Greenblatt – What happens once AI can automate AI research?youtube.com 2nd of 26 in this recording
- 1 day earlier
- JC
Jack Clark quoted
Our readingToday's AI systems are significantly under-elicited for their ability to automate AI R&D.
Their wordsPosts like this highlight how we are under-eliciting today's AI systems for their ability to automate AI R&D - especially striking is how the company can jump the performance of Opus 5 by 10 absolute percentage points with a better harness.
↗Import AI 468: 23 RSI ideas; PostTrainBench+; and how trust and transparency interplay with AI racingjack-clark.net 1st of 3 in this piece
- FC
François Chollet quoted
Creator of the Keras deep-learning library and the ARC-AGI benchmark
Their wordsCoding isn't yet another application domain -- it's the meta-skill required for AI to automatically develop its own training material, via symbolic world models. That's how the RSI loop actually kicks off.
- 7 days earlier
- CP
Chamath Palihapitiya quoted
Founder and CEO of the venture firm Social Capital
Their wordsLooking at the results and capabilities from the various labs over the past few weeks I would say we are firmly in this loop now.
- CP
Chamath Palihapitiya quoted
Their wordsRecursive self improvement will dramatically increase capability very quickly from here.
- 4 days earlier
- JD
Jeff Dean quoted
Chief Scientist at Google DeepMind and Google Research
Their wordsUm but I think there's no uh you know real impediment to making that be a much more automated loop where the model itself decides it's going to explore or maybe with a nudge from some people uh at the various highest level like oh why don't you try some new ideas around model architectures that incorporate this and then it will go run lots of experiments uh see which ones work and then those will get incorporated at a much more rapid rate
↗Jeff Dean: The 1% Rule for Building in AIyoutube.com 21st of 22 in this recording
- 4 days earlier
- JH
Jensen Huang quoted
Co-founder and CEO of NVIDIA since 1993
Their wordswe we kind of have coarse level uh recursive self-improvement already. And the fact that every time you use it, it improves the markdown files. Uh every time you use it, it updates its uh long-term memory.
↗Jensen Huang: The Mindset That Built NVIDIAyoutube.com 9th of 20 in this recording
- 3 weeks earlier
- LW
Lilian Weng quoted
Machine-learning researcher
Their wordsRecursive structure alone is not enough. The base model must be capable enough to improve the mechanism.
↗Harness Engineering for Self-Improvementlilianweng.github.io 2nd of 3 in this piece
- 5 months earlier
- JL
Jan Leike quoted
Alignment researcher
Our readingRecursive self-improvement has already started, because AI research itself is now being automated.
Their wordsWe are starting to automate AI research and the recursive self-improvement process has begun.
↗Alignment is not solvedaligned.substack.com 4th of 5 in this piece
- 6 months earlier
- DH
Demis Hassabis quoted
Co-founder and CEO of Google DeepMind
Their wordsI mean potentially that's possible. I would say I'm not sure it's even desirable because that's a kind of hard takeoff scenario.
↗Demis Hassabis: Future of AI, Simulating Reality, Physics and Video Games | Lex Fridman Podcast #475youtube.com 9th of 20 in this recording
- 6 months earlier
- NL
Nathan Lambert quoted
Their wordsAnd for that reason, there's physical constraints to things like AGI, like recursive improvement to kill us all type stuff. For the physical reasons and for how humans have figured things out before, I'm not too worried about AI takeover.
↗DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters | Lex Fridman Podcast #459youtube.com 43rd of 44 in this recording