You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 900f863
Browse filesBrowse the repository at this point in the historyBrowse files
Update the survey pages for the published Preprints.org record (#17)
* Update the survey pages for the published Preprints.org record
The survey is now published on Preprints.org as
doi:10.20944/preprints202608.2108.v1 (posted 2026-08-28) under the
title 'AI4AI Survey: From Long-Horizon Agents to Recursive
Self-Improvement—Definitions, Reliable Horizons, and Open Problems'.
- Adopt the published title in the page title, h1, og/twitter cards,
citation tags, and the ScholarlyArticle nodes on both pages
- Add citation_doi, a DOI identifier, and sameAs links to the
preprint record; sync the abstract and dates with the published v1
- Replace the 'Paper (coming soon)' placeholder with a link to the
preprint, and link the DOI from the tagline and citation section
- Update the BibTeX entry to cite the Preprints.org record
- Refresh the homepage work card and README entry, and bump sitemap
lastmod dates
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HZxZsJ6KCnffNAfu8WxquU
* Replace the survey BibTeX with the full-author Zenodo citation
Cite all 23 authors and point the DOI at the Zenodo archive
(10.5281/zenodo.22198847) per the authors' preferred citation.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HZxZsJ6KCnffNAfu8WxquU
* Drop the DOI from the survey tagline
The DOI stays in the eyebrow's Preprints.org label, the Paper button,
and the citation section; the tagline reads as one clean line again.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HZxZsJ6KCnffNAfu8WxquU
---------
Co-authored-by: Claude <noreply@anthropic.com>
<title>On the Eve of AI4AI: A Survey of Long-Horizon Agents and RSI</title>
8
+
<title>AI4AISurvey: From Long-Horizon Agents to Recursive Self-Improvement</title>
9
9
<meta
10
10
name="description"
11
-
content="An AI4AI survey: how far can AI reliably carry an improvement loop from idea to verified result? A taxonomy, a closure audit of 35 systems, and the composition gap."
11
+
content="The AI4AI survey, published on Preprints.org: how far can AI reliably carry an improvement loop from idea to verified result? A taxonomy, a closure audit of 35 systems, and the composition gap."
Without that tag Scholar processes the PDF as if it carried no metadata. -->
29
-
<metaname="citation_title" content="On the Eve of AI4AI: From Long-Horizon Agents to Recursive Self-Improvement" />
26
+
required (drop one and Scholar ignores the rest). Only add a
27
+
citation_pdf_url once a self-hosted copy of the PDF is committed to
28
+
this repo; a citation_pdf_url that 404s drops the whole record. -->
29
+
<metaname="citation_title" content="AI4AI Survey: From Long-Horizon Agents to Recursive Self-Improvement—Definitions, Reliable Horizons, and Open Problems" />
<metaname="citation_keywords" content="AI4AI; AI for AI; large language models; LLM agents; AI agents; long-horizon agents; reliable region; reliable-horizon surface; benchmark survey; self-improving agents; self-evolving agents; recursive self-improvement; autonomous AI research; automated AI research; AI scientists; machine learning engineering; functional coupling; stage ownership; composition gap; agent harness; agent scaffolds; agent evaluation" />
59
60
60
-
<metaproperty="og:title" content="On the Eve of AI4AI: From Long-Horizon Agents to Recursive Self-Improvement" />
61
+
<metaproperty="og:title" content="AI4AI Survey: From Long-Horizon Agents to Recursive Self-Improvement" />
61
62
<meta
62
63
property="og:description"
63
-
content="A survey of how far AI systems can reliably carry an improvement process from idea to verified result—and what still limits recursive self-improvement."
64
+
content="Published on Preprints.org: a survey of how far AI systems can reliably carry an improvement process from idea to verified result—and what still limits recursive self-improvement."
<metaname="twitter:title" content="On the Eve of AI4AI: A Survey of Long-Horizon Agents and RSI" />
82
+
<metaname="twitter:title" content="AI4AISurvey: From Long-Horizon Agents to Recursive Self-Improvement" />
82
83
<meta
83
84
name="twitter:description"
84
85
content="AI increasingly performs the work of improvement, but humans still define its goals and evidence. A survey of AI4AI's reliable limits and composition gap."
"name": "On the Eve of AI4AI: From Long-Horizon Agents to Recursive Self-Improvement",
94
-
"headline": "On the Eve of AI4AI: From Long-Horizon Agents to Recursive Self-Improvement",
95
-
"alternativeHeadline": "Definitions, Reliable Horizons, and Open Problems",
94
+
"name": "AI4AI Survey: From Long-Horizon Agents to Recursive Self-Improvement—Definitions, Reliable Horizons, and Open Problems",
95
+
"headline": "AI4AI Survey: From Long-Horizon Agents to Recursive Self-Improvement—Definitions, Reliable Horizons, and Open Problems",
96
96
"description": "A survey of AI4AI (AI for AI): how far an AI system can reliably carry an improvement process from idea to a validated result, a taxonomy for locating where the loop closes, a closure audit of 35 systems, and the composition gap that bounds recursive self-improvement.",
97
-
"abstract": "Can AI improve AI? AI agents can now run experiments, modify code and training pipelines, and iteratively improve AI artifacts—bringing a long-standing idea closer to an empirical research question. Yet the rapidly expanding literature remains fragmented across long-horizon agents, AI for AI (AI4AI), self-improvement, and recursive self-improvement, making it difficult to tell how much progress has actually been made. This survey synthesizes evidence from hundreds of studies to answer this question. We organize the emerging AI4AI landscape around a simple question: how far can an AI system reliably carry an improvement process from idea to verified result? Across model design, agent harnesses, benchmarks, automated research, and self-modifying systems, we examine what current systems can do, how progress should be evaluated, and where claims of self-improvement remain unsupported. AI systems increasingly excel at the work of improvement—planning, coding, experimentation, optimization, and repair—but humans still largely determine the goals, evaluation criteria, and what ultimately counts as progress. Moreover, strong performance on individual components rarely translates into reliable end-to-end AI improvement. We call this composition gap. Today's systems can already produce impressive improvements under bounded conditions, but evidence for reliable research judgment, causal experimentation, persistent gains, and compounding improvement remains limited. By separating demonstrated capability from extrapolated autonomy, this survey maps what AI4AI can do today—and what must change before AI can reliably improve AI itself.",
97
+
"abstract": "Can AI improve AI? AI agents can now run experiments, modify code and training pipelines, and iteratively improve AI artifacts—bringing a long-standing idea closer to an empirical research question. Yet the rapidly expanding literature remains fragmented across long-horizon agents, AI for AI (AI4AI), self-improvement, and recursive self-improvement, making it difficult to tell how much progress has actually been made. This survey synthesizes evidence from hundreds of studies to answer this question. We organize the emerging AI4AI landscape around a simple question: how far can an AI system reliably carry an improvement process from idea to verified result? Across model design, agent harnesses, benchmarks, automated research, and self-modifying systems, we examine what current systems can do, how progress should be evaluated, and where claims of self-improvement remain unsupported. A striking pattern emerges in the information flow within the taxonomy–benchmark–model–harness structure of this survey. AI systems increasingly excel at the work of improvement—planning, coding, experimentation, optimization, and repair—but humans still largely determine the goals, evaluation criteria, and what ultimately counts as progress. Moreover, strong performance on individual components rarely translates into reliable end-to-end AI improvement. We call this composition gap. Today's systems can already produce impressive improvements under bounded conditions, but evidence for reliable research judgment, causal experimentation, persistent gains, and compounding improvement remains limited. By separating demonstrated capability from extrapolated autonomy, this survey maps what AI4AI can do today—and what must change before AI can reliably improve AI itself. The eve ends the moment a system, for the first time, reliably strengthens its successors without humans specifying the goals or evaluation criteria. We hope this taxonomy and survey will help that moment arrive a little sooner.",
<spanclass="lang-en" lang="en">The survey is published on Preprints.org as <ahref="https://doi.org/10.20944/preprints202608.2108.v1" target="_blank" rel="noreferrer">doi:10.20944/preprints202608.2108.v1</a>. If it helps your work, please cite it as:</span>
"name": "On the Eve of AI4AI: From Long-Horizon Agents to Recursive Self-Improvement",
116
+
"name": "AI4AI Survey: From Long-Horizon Agents to Recursive Self-Improvement—Definitions, Reliable Horizons, and Open Problems",
117
117
"url": "https://simpleagentlab.com/ai4ai/",
118
118
"description": "A survey of AI4AI (AI for AI): how far an AI system can reliably carry an improvement process from idea to a validated result, a taxonomy for locating where the loop closes, a closure audit of 35 systems, and the composition gap that bounds recursive self-improvement.",
119
119
"genre": "survey",
120
-
"datePublished": "2026-08-12",
120
+
"datePublished": "2026-08-28",
121
121
"inLanguage": ["en","zh-CN"],
122
122
"publisher": {
123
123
"@id": "https://simpleagentlab.com/#organization"
@@ -286,23 +286,28 @@ <h2>
286
286
<a
287
287
class="work-summary"
288
288
href="./ai4ai/"
289
-
aria-label="On the Eve of AI4AI — an AI4AI survey"
289
+
aria-label="AI4AI Survey — from long-horizon agents to recursive self-improvement"
290
290
>
291
291
<spanclass="work-tag">
292
292
<spanclass="lang-en" lang="en">Survey</span>
293
293
<spanclass="lang-zh" lang="zh-CN">综述</span>
294
294
</span>
295
-
<strong>On the Eve of AI4AI</strong>
295
+
<strong>AI4AI Survey</strong>
296
296
<small>
297
-
<spanclass="lang-en" lang="en">A survey of AI4AI, from long-horizon agents to recursive self-improvement: definitions, reliable horizons, and a closure audit of 35 systems.</span>
<spanclass="lang-en" lang="en">A survey of AI4AI, from long-horizon agents to recursive self-improvement: definitions, reliable horizons, and a closure audit of 35 systems. Now published on Preprints.org.</span>
0 commit comments