• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    DAIR.AI describes a survey mapping stages of AI self-improvement

    DAIR.AI says the framework makes self-improvement claims easier to check by asking which stages an agent actually automates.

    GM
    YM
    AT
    25 Sources, ,

    TLDR

    DAIR.AI highlights a survey that breaks recursive self-improvement into stages of autonomy. Its description starts with agents executing improvements designed by others, then moves through choosing their own improvement strategy, collecting their own experience, adapting to new environments and improving the process of improvement itself. DAIR.AI says the survey also compares requirements across scientific discovery, embodied agents and software engineering.

    Combined views

    536K

    25 Sources, first seen 23d ago

    Combined views

    536K

    25 Sources, first seen 23d ago

    5.3K likes
    23d ago
    first seen 23d ago
    5.3K likes
    259 comments
    3.2K saves
    703 reposts
    259 comments
    3.2K saves
    703 reposts

    Sentiment

    Positive53.6%46.4%Negative

    Summary

    Sentiment

    Positive53.6%46.4%Negative

    Many accounts welcomed RSI roadmaps and research surveys as fascinating and solid references, while others raised concerns about safety risks, runaway errors if definitions go wrong, and the impossibility of slowing development.

    Based on 97 sentiment-bearing replies from 84 accounts across 7 conversations.

    Summary

    Many accounts welcomed RSI roadmaps and research surveys as fascinating and solid references, while others raised concerns about safety risks, runaway errors if definitions go wrong, and the impossibility of slowing development.

    Based on 97 sentiment-bearing replies from 84 accounts across 7 conversations.

    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    25 Sources

    @iamtraskIf we try, AI can go a little better tomorrow than it did today
    @TheseusLabsAIThe Last AI Built by Humans: a roadmap to genuine RSI. Our review of 72 companies & teams shows AI already executes and selects improvements. The next frontier: can AI reliably improve the improver itself? 🧭 Paper: https://arxiv.org/abs/2609.11873
    @dair_aiSelf-improving agents are a top research topic right now. This new survey is a good map of the area. (bookmark it) It splits recursive self-improvement into stages of autonomy. An agent first executes improvements someone else designed. Then it chooses its own improvement strategy, collects its own experience, adapts to new environments, and finally improves the process of improvement itself. That staging makes claims easier to check. When a paper says its agent is self-improving, you can ask which of these stages it actually automates. The survey also uses a Headroom-Closed Index to show where current LLMs fall short, and compares requirements across scientific discovery, embodied agents and software engineering. Paper: https://academy.dair.ai/papers/the-last-ai-built-by-humans-toward-genuine-recursive-self-improvement-2609.11873
    @omarsar0Recommended reading. Finally, a great overview of recursive self-improvement (RSI).
    @huyberyRT @TheseusLabsAI: The Last AI Built by Humans: a roadmap to genuine RSI. Our review of 72 companies & teams shows AI already executes and…
    @teortaxesTexEAs: «the Chynese don't believe in superintelligence, they think AI is just a tool or an engineering problem… sigh, if only they were more worldly and could read Yudkowsky and LessWrong…» Meanwhile, average Chynese paper on AI:
    @hsu_steveRSI WATCH: The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement "In this report, we present an autonomy-centered roadmap for recursive self-improvement" https://arxiv.org/abs/2609.11873
    @PlinzUSA is entering a frenzy of anti AI hysteria fueled by extremist politicians, doomsday cults and irresponsible journalists, while China enthusiastically builds AI.
    @rohanpaul_aiBeautiful roadmap paper on Recursive self-improvement. Concludes, we are already seeing pieces of RSI, but full recursive self-improvement is not here yet. Most self-improving AI still cannot improve how it improves Says that most things called "self-improving AI" today only automate parts of the improvement process. Genuine recursive self-improvement would mean the AI can persistently improve not just its outputs, prompts, tools, or code, but eventually the mechanism that decides how future improvements are discovered, tested, and kept. AI is already very strong at answering knowledge and reasoning questions, but still much weaker at doing long, multi-step tasks with tools, software, and changing environments. The paper maps progress across 5 levels, from executing human-designed improvements to changing the improver, evaluator, or research policy used in later rounds. That last step makes the process recursive: a successful update changes how future updates are discovered or judged. The survey finds broad evidence for lower levels, while experience-driven learning and deployment adaptation are more domain-dependent and end-to-end L5 evidence remains concentrated in bounded prototypes.
    @menhguin@Plinz ... you cannot seriously be citing an AI generated literature review with a bajillion coauthors to make your point about building RSI, man

    25 Sources

    @iamtraskIf we try, AI can go a little better tomorrow than it did today
    @TheseusLabsAIThe Last AI Built by Humans: a roadmap to genuine RSI. Our review of 72 companies & teams shows AI already executes and selects improvements. The next frontier: can AI reliably improve the improver itself? 🧭 Paper: https://arxiv.org/abs/2609.11873
    @dair_aiSelf-improving agents are a top research topic right now. This new survey is a good map of the area. (bookmark it) It splits recursive self-improvement into stages of autonomy. An agent first executes improvements someone else designed. Then it chooses its own improvement strategy, collects its own experience, adapts to new environments, and finally improves the process of improvement itself. That staging makes claims easier to check. When a paper says its agent is self-improving, you can ask which of these stages it actually automates. The survey also uses a Headroom-Closed Index to show where current LLMs fall short, and compares requirements across scientific discovery, embodied agents and software engineering. Paper: https://academy.dair.ai/papers/the-last-ai-built-by-humans-toward-genuine-recursive-self-improvement-2609.11873
    @omarsar0Recommended reading. Finally, a great overview of recursive self-improvement (RSI).
    @huyberyRT @TheseusLabsAI: The Last AI Built by Humans: a roadmap to genuine RSI. Our review of 72 companies & teams shows AI already executes and…
    @teortaxesTexEAs: «the Chynese don't believe in superintelligence, they think AI is just a tool or an engineering problem… sigh, if only they were more worldly and could read Yudkowsky and LessWrong…» Meanwhile, average Chynese paper on AI:
    @hsu_steveRSI WATCH: The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement "In this report, we present an autonomy-centered roadmap for recursive self-improvement" https://arxiv.org/abs/2609.11873
    @PlinzUSA is entering a frenzy of anti AI hysteria fueled by extremist politicians, doomsday cults and irresponsible journalists, while China enthusiastically builds AI.
    @rohanpaul_aiBeautiful roadmap paper on Recursive self-improvement. Concludes, we are already seeing pieces of RSI, but full recursive self-improvement is not here yet. Most self-improving AI still cannot improve how it improves Says that most things called "self-improving AI" today only automate parts of the improvement process. Genuine recursive self-improvement would mean the AI can persistently improve not just its outputs, prompts, tools, or code, but eventually the mechanism that decides how future improvements are discovered, tested, and kept. AI is already very strong at answering knowledge and reasoning questions, but still much weaker at doing long, multi-step tasks with tools, software, and changing environments. The paper maps progress across 5 levels, from executing human-designed improvements to changing the improver, evaluator, or research policy used in later rounds. That last step makes the process recursive: a successful update changes how future updates are discovered or judged. The survey finds broad evidence for lower levels, while experience-driven learning and deployment adaptation are more domain-dependent and end-to-end L5 evidence remains concentrated in bounded prototypes.
    @menhguin@Plinz ... you cannot seriously be citing an AI generated literature review with a bajillion coauthors to make your point about building RSI, man