Pith.

Artificial Intelligence · arXiv:2603.19461 v2 · submitted 19 Mar 2026

Hyperagents

Jenny Zhang, Bingchen Zhao, Wannan Yang, Jakob Foerster, Jeff Clune, Minqi Jiang, Sam Devlin, Tatiana Shavrina

The paper says

“Across diverse domains, the DGM-H improves performance over time and outperforms baselines without self-improvement or open-ended exploration, as well as prior self-improving systems.”

from the abstract · picked out as the main claim by a classifier

Pith says

Pith has not reviewed this paper.

Get a Pith Review, $5

Reviews

Reviews by people

No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review

How Pith checks papers · The full audit of this paper

The paper

Abstract

Self-improving AI systems aim to reduce reliance on human engineering by learning to improve their own learning and problem-solving processes. Existing approaches to self-improvement rely on fixed, handcrafted meta-level mechanisms, fundamentally limiting how fast such systems can improve. The Darwin Gödel Machine (DGM) demonstrates open-ended self-improvement in coding by repeatedly generating and evaluating self-modified variants. Because both evaluation and self-modification are coding tasks, gains in coding ability can translate into gains in self-improvement ability. However, this alignment does not generally hold beyond coding domains. We introduce \textbfhyperagents, self-referential agents that integrate a task agent (which solves the target task) and a meta agent (which modifies itself and the task agent) into a single editable program. Crucially, the meta-level modification procedure is itself editable, enabling metacognitive self-modification, improving not only the task-solving behavior, but also the mechanism that generates future improvements. We instantiate this framework by extending DGM to create DGM-Hyperagents (DGM-H), eliminating the assumption of domain-specific alignment between task performance and self-modification skill to potentially support self-accelerating progress on any computable task. Across diverse domains, the DGM-H improves performance over time and outperforms baselines without self-improvement or open-ended exploration, as well as prior self-improving systems. Furthermore, the DGM-H improves the process by which it generates new agents (e.g., persistent memory, performance tracking), and these meta-level improvements transfer across domains and accumulate across runs. DGM-Hyperagents offer a glimpse of open-ended AI systems that do not merely search for better solutions, but continually improve their search for how to improve.

The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.

abstract compiled · 2026-11-03 · no language-model review

How the main claim is stated: asserted 1.00

What the abstract itself states, as probabilities: claims first 0.12 · states limitation 0.23 · human participants 0.04 · controlled comparison 0.93 · randomized assignment 0.01 · states quantitative result 0.05 · compares with prior results 0.98 · mentions code or data availability 0.03

The abstract, sentence by sentence

This table labels the abstract only (arXiv listing abstract v2). The number is the probability the classifier gave the label.

SentenceRoleThe authors' own workStates the main conclusionStrength
Self-improving AI systems aim to reduce reliance on human engineering by learning to improve their own learning and problem-solving processes. background 1.00 0.04 0.03 asserted 0.97
Existing approaches to self-improvement rely on fixed, handcrafted meta-level mechanisms, fundamentally limiting how fast such systems can improve. background 0.50 0.05 0.02 asserted 1.00
The Darwin G\"odel Machine (DGM) demonstrates open-ended self-improvement in coding by repeatedly generating and evaluating self-modified variants. background 1.00 0.10 0.03 asserted 1.00
Because both evaluation and self-modification are coding tasks, gains in coding ability can translate into gains in self-improvement ability. background 0.70 0.11 0.03 asserted 0.78
However, this alignment does not generally hold beyond coding domains. gap 0.60 0.18 0.04 asserted 0.99
We introduce \textbf{hyperagents}, self-referential agents that integrate a task agent (which solves the target task) and a meta agent (which modifies itself and the task agent) into a single editable program. contribution 0.88 0.80 0.09 asserted 1.00
Crucially, the meta-level modification procedure is itself editable, enabling metacognitive self-modification, improving not only the task-solving behavior, but also the mechanism that generates future improvements. method 0.64 0.55 0.15 asserted 1.00
We instantiate this framework by extending DGM to create DGM-Hyperagents (DGM-H), eliminating the assumption of domain-specific alignment between task performance and self-modification skill to potentially support self-accelerating progress on any computable task. method 0.56 0.84 0.16 hedged possibility 0.75
Across diverse domains, the DGM-H improves performance over time and outperforms baselines without self-improvement or open-ended exploration, as well as prior self-improving systems. result 1.00 0.94 0.49 asserted 1.00
Furthermore, the DGM-H improves the process by which it generates new agents (e.g., persistent memory, performance tracking), and these meta-level improvements transfer across domains and accumulate across runs. result 1.00 0.92 0.15 asserted 1.00
DGM-Hyperagents offer a glimpse of open-ended AI systems that do not merely search for better solutions, but continually improve their search for how to improve. interpretation 1.00 0.31 0.49 hedged possibility 0.92

What kind of paper the classifier reports this is: method or tool 1.00 · evidence from simulation or computation 0.52

What was examined, and what was not

Sources read

  • arXiv listing abstract v2

Not examined

  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers

Levels run

  • P abstract

Not run

  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

How the labels were made

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-03: 1 request, 8,357 input tokens.

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON

Cited by 11 Pith papers
How the literature uses it

34 arXiv papers cite this in 61 sentences

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

Papers citing it · 38 sentences without a confident label

Compares with it 4 papers
  • We compare: (i) four non-LLM baselines (equal-weight, momentum-weighted, min-variance, inverse-momentum); (ii) five prior self-improving methods: Reflexion [ref] , ExpeL [ref] , FactorMiner [ref] , Meta-Reflexion [ref] , and EvoTool [ref] (adaptation details in <ref>); (iii) HyperAgent [ref] , a recursive self-modification framework (<ref>); and (iv) an incremental build (Stateless $→$ +Tools $→$ +Memory $→$ ).

    AEL: Evolving Agent Harness in Open-Ended Environments · Experiments Setup · 0.98

  • On the three open-ended tasks, we compare against two agent-based evolution baselines, Codex and Claude Code, and two procedure-based evolution baselines, OpenEvolve [ref] and HyperAgents [ref] .

    Harnessing Agentic Evolution · Settings · 1.00

  • We further distinguish ourselves from [ref] in that they introduce meta-cognitive ability to Darwin Gödel Machine variants and mainly note that this meta-cognitive ability enables transfer of learned cognitive strategy across tasks but does not yield statistically significant improvements over customized Darwin Gödel Machines.

    SBCO: Self-Supervised, Verifier-Grounded Harness Optimization For Planning Agents · Related Work · 0.84

Cites it as support 3 papers
  • By converting massive execution logs into actionable insights and fixing task-specific bottlenecks step-by-step, harness optimization has enabled better agent performance than hand-written harnesses [ref] .

    Towards Direct Evaluation of Harness Optimizers via Priority Ranking · Introduction · 0.96

  • Academic work has shown source-level self-rewriting to be feasible on minimal scaffolds: SICA [ref] , the Darwin Gödel Machine [ref] , and HyperAgents [ref] all demonstrate an agent modifying its own implementation to lift its own benchmark scores; Meta-Harness [ref] further shows that exposing past execution traces to the coding-agent proposer drives iterative improvement more strongly than benchmark scores alone do.

    MOSS: Self-Evolution through Source-Level Rewriting in Autonomous Agent Systems · Introduction · 0.98

  • Recent agent work such as HyperAgents is useful here because it shows that self-improvement should not be delegated to a fixed outer wrapper; a task agent improved by a fixed meta-agent inherits the limits of that meta-agent, while adding further meta-levels only moves the bottleneck upward [ref] .

    A Multi-Timescale Recursive Self-Improvement Engine for Open-Ended Persona Growth · Recursive self-improvement and bounded recursive self-evolution · 0.98

Cites it as background 13 papers

Challenges and discussion

Know something about this paper? Leave a comment.

No comments yet.

Sign in to comment