AutoTTS discovers width-depth test-time scaling controllers through agentic search in a pre-collected trajectory environment, yielding better accuracy-cost tradeoffs than hand-designed baselines on math reasoning tasks at low cost.
hub Canonical reference
ThetaEvolve: Test-time Learning on Open Problems
Canonical reference. 88% of citing Pith papers cite this work as background.
31
Pith papers citing it
Background
88% of classified citations
abstract
Recent advances in large language models (LLMs) have enabled breakthroughs in mathematical discovery, exemplified by AlphaEvolve, a closed-source system that evolves programs to improve bounds on open problems. However, it relies on ensembles of frontier LLMs to achieve new bounds and is a pure inference system that models cannot internalize the evolving strategies. We introduce ThetaEvolve, an open-source framework that simplifies and extends AlphaEvolve to efficiently scale both in-context learning and Reinforcement Learning (RL) at test time, allowing models to continually learn from their experiences in improving open optimization problems. ThetaEvolve features a single LLM, a large program database for enhanced exploration, batch sampling for higher throughput, lazy penalties to discourage stagnant outputs, and optional reward shaping for stable training signals, etc. ThetaEvolve is the first evolving framework that enable a small open-source model, like DeepSeek-R1-0528-Qwen3-8B, to achieve new best-known bounds on open problems (circle packing and first auto-correlation inequality) mentioned in AlphaEvolve. Besides, across two models and four open tasks, we find that ThetaEvolve with RL at test-time consistently outperforms inference-only baselines, and the model indeed learns evolving capabilities, as the RL-trained checkpoints demonstrate faster progress and better final performance on both trained target task and other unseen tasks. We release our code publicly: https://github.com/ypwang61/ThetaEvolve
hub tools
citation-role summary
background 7
baseline 1
citation-polarity summary
years
2026 31representative citing papers
Frontier coding agents surpass published Nature-family SOTA on only 17.8% of 90 sealed scientific tasks, mostly by recasting problems as supervised ML rather than inventing methods.
FunctionEvolve recovers 107 exact symbolic forms out of 129 synthetic tasks (82.9% SA@50) by using expression-tree structure for evolutionary search, parent selection, mutation, and coefficient scoring with LLMs.
EMO-STA evolves a shared program archive across task families then adapts candidates to targets, outperforming matched-compute single-task evolution in most of eight families while reducing overfitting on low-data tasks like ARC.
Evolutionary coding agents achieve most benchmark gains through a small subset of edit types and by cycling previously deleted code lines rather than developing new algorithmic structures.
Current AI agents fail to reliably invent transferable ML methods on a controlled 140-task benchmark, and more test-time compute or context does not remove the scientific-judgment bottleneck.
Agentic-imodels evolves scikit-learn regressors via an autoresearch loop to jointly boost predictive performance and LLM-simulatability, improving downstream agentic data science tasks by up to 73% on the BLADE benchmark.
DeInfer reduces parallel inference communication cost for decomposed LLMs by up to 78% by moving collective operations into the low-rank latent space and redesigning KV-cache reconstruction for static graph compatibility.
k-server-bench formulates potential-function discovery for the k-server conjecture as a code-based inequality-satisfaction task; current agents fully solve the resolved k=3 case and reduce violations on the open k=4 case.
TTT-Discover applies test-time RL to set new state-of-the-art results on math inequalities, GPU kernels, algorithm contests, and single-cell denoising using an open model and public code.
LLM formal provers must shift from competition solvers to research agents that handle open-ended, under-specified frontier mathematics under machine-checked rigor.
Evolution Fine-Tuning trains LLMs on 156K trajectories spanning 371 tasks to achieve 10.22% average improvement on 22 held-out optimization tasks and match SOTA on select circle-packing problems when combined with test-time RL.
REVES augments LLM post-training by decoupling revision and verification signals from successful multi-step trajectories, reporting +6.5 point gains on LiveCodeBench over RL baselines.
Visored is a controlled-natural-language prover for LLM math that automates omitted routine steps and emits checked Lean output, with early miniF2F results showing LLMs can use it without prover-specific training.
An economy of agents using auctions and wealth accumulation produces emergent multi-step reasoning that outperforms monolithic baselines on five agentic tasks.
LLMs can forecast GPU kernel performance accurately enough to serve as selective surrogates, allowing kernel searches to consider more candidates and recover faster kernels under fixed GPU evaluation budgets.
Bidirectional Evolutionary Search augments autoregressive expansion with evolutionary recombination operators and dense backward subgoal feedback to produce better candidates than standard best-of-N or tree search for language model self-improvement.
Priority ranking offers a low-cost direct evaluation for harness optimizers that correlates with their real multi-step optimization performance, supported by the Shor dataset of 182 scenarios.
EvoLib improves black-box LLM test-time performance by maintaining an evolving, self-scored library of reusable skills and insights, without parameter updates or ground-truth feedback.
Shepherd provides a reversible execution trace substrate for LLM agents that enables meta-agents to inspect and transform runs, yielding reported gains on coding and terminal benchmarks via supervision, counterfactual repair, and RL credit assignment.
An LLM-driven agentic system evolves microarchitectural policies for cache replacement, data prefetching, and branch prediction, producing designs that match or exceed prior state-of-the-art in IPC on standard benchmarks.
SimpleTES scales test-time evaluation in LLMs to discover state-of-the-art solutions on 21 scientific problems across six domains, outperforming frontier models and optimization pipelines with examples like 2x faster LASSO and new Erdos constructions.
TEMPO scales test-time training for large reasoning models by interleaving policy refinement on unlabeled data with critic recalibration on labeled data via an EM formulation, yielding large gains on AIME tasks.
TurboEvolve improves LLM program evolution by running parallel islands with LLM-generated diverse candidates that carry self-assigned weights, an adaptive scheduler, and clustered seed injection to reach stronger solutions at lower evaluation budgets.
citing papers explorer
-
LLMs Improving LLMs: Agentic Discovery for Test-Time Scaling
AutoTTS discovers width-depth test-time scaling controllers through agentic search in a pre-collected trajectory environment, yielding better accuracy-cost tradeoffs than hand-designed baselines on math reasoning tasks at low cost.
-
NatureBench: Can Coding Agents Match the Published SOTA of Nature-Family Papers?
Frontier coding agents surpass published Nature-family SOTA on only 17.8% of 90 sealed scientific tasks, mostly by recasting problems as supervised ML rather than inventing methods.
-
FunctionEvolve: Structure-Guided Symbolic Regression with LLMs
FunctionEvolve recovers 107 exact symbolic forms out of 129 synthetic tasks (82.9% SA@50) by using expression-tree structure for evolutionary search, parent selection, mutation, and coefficient scoring with LLMs.
-
Evolutionary Multi-Task Optimization for LLM-Guided Program Discovery
EMO-STA evolves a shared program archive across task families then adapts candidates to targets, outperforming matched-compute single-task evolution in most of eight families while reducing overfitting on low-data tasks like ARC.
-
What Do Evolutionary Coding Agents Evolve?
Evolutionary coding agents achieve most benchmark gains through a small subset of edit types and by cycling previously deleted code lines rather than developing new algorithmic structures.
-
MLS-Bench: A Holistic and Rigorous Assessment of AI Systems on Building Better AI
Current AI agents fail to reliably invent transferable ML methods on a controlled 140-task benchmark, and more test-time compute or context does not remove the scientific-judgment bottleneck.
-
Agentic-imodels: Evolving agentic interpretability tools via autoresearch
Agentic-imodels evolves scikit-learn regressors via an autoresearch loop to jointly boost predictive performance and LLM-simulatability, improving downstream agentic data science tasks by up to 73% on the BLADE benchmark.
-
Co-evolving Agent Architectures and Interpretable Reasoning for Automated Optimization
DeInfer reduces parallel inference communication cost for decomposed LLMs by up to 78% by moving collective operations into the low-rank latent space and redesigning KV-cache reconstruction for static graph compatibility.
-
$k$-server-bench: Automating Potential Discovery for the $k$-Server Conjecture
k-server-bench formulates potential-function discovery for the k-server conjecture as a code-based inequality-satisfaction task; current agents fully solve the resolved k=3 case and reduce violations on the open k=4 case.
-
Learning to Discover at Test Time
TTT-Discover applies test-time RL to set new state-of-the-art results on math inequalities, GPU kernels, algorithm contests, and single-cell denoising using an open model and public code.
-
From Solvers to Research: Large Language Model-Driven Formal Mathematics at the Research Frontier
LLM formal provers must shift from competition solvers to research agents that handle open-ended, under-specified frontier mathematics under machine-checked rigor.
-
Evolution Fine-Tuning: Learning to Discover Across 371 Optimization Tasks
Evolution Fine-Tuning trains LLMs on 156K trajectories spanning 371 tasks to achieve 10.22% average improvement on 22 held-out optimization tasks and match SOTA on select circle-packing problems when combined with test-time RL.
-
REVES: REvision and VErification--Augmented Training for Test-Time Scaling
REVES augments LLM post-training by decoupling revision and verification signals from successful multi-step trajectories, reporting +6.5 point gains on LiveCodeBench over RL baselines.
-
Visored: A Controlled-Natural-Language Prover for LLM-Generated Mathematics
Visored is a controlled-natural-language prover for LLM math that automates omitted routine steps and emits checked Lean output, with early miniF2F results showing LLMs can use it without prover-specific training.
-
Economy of Minds: Emerging Multi-Agent Intelligence with Economic Interactions
An economy of agents using auctions and wealth accumulation produces emergent multi-step reasoning that outperforms monolithic baselines on five agentic tasks.
-
GPU Forecasters: Language Models as Selective Surrogates for Kernel Runtime Optimization
LLMs can forecast GPU kernel performance accurately enough to serve as selective surrogates, allowing kernel searches to consider more candidates and recover faster kernels under fixed GPU evaluation budgets.
-
Self-Improving Language Models with Bidirectional Evolutionary Search
Bidirectional Evolutionary Search augments autoregressive expansion with evolutionary recombination operators and dense backward subgoal feedback to produce better candidates than standard best-of-N or tree search for language model self-improvement.
-
Towards Direct Evaluation of Harness Optimizers via Priority Ranking
Priority ranking offers a low-cost direct evaluation for harness optimizers that correlates with their real multi-step optimization performance, supported by the Shor dataset of 182 scenarios.
-
Test-Time Learning with an Evolving Library
EvoLib improves black-box LLM test-time performance by maintaining an evolving, self-scored library of reusable skills and insights, without parameter updates or ground-truth feedback.
-
Shepherd: Enabling Programmable Meta-Agents via Reversible Agentic Execution Traces
Shepherd provides a reversible execution trace substrate for LLM agents that enables meta-agents to inspect and transform runs, yielding reported gains on coding and terminal benchmarks via supervision, counterfactual repair, and RL credit assignment.
-
Agentic Architect: An Agentic AI Framework for Architecture Design Exploration and Optimization
An LLM-driven agentic system evolves microarchitectural policies for cache replacement, data prefetching, and branch prediction, producing designs that match or exceed prior state-of-the-art in IPC on standard benchmarks.
-
Evaluation-driven Scaling for Scientific Discovery
SimpleTES scales test-time evaluation in LLMs to discover state-of-the-art solutions on 21 scientific problems across six domains, outperforming frontier models and optimization pipelines with examples like 2x faster LASSO and new Erdos constructions.
-
TEMPO: Scaling Test-time Training for Large Reasoning Models
TEMPO scales test-time training for large reasoning models by interleaving policy refinement on unlabeled data with critic recalibration on labeled data via an EM formulation, yielding large gains on AIME tasks.
-
TurboEvolve: Towards Fast and Robust LLM-Driven Program Evolution
TurboEvolve improves LLM program evolution by running parallel islands with LLM-generated diverse candidates that carry self-assigned weights, an adaptive scheduler, and clustered seed injection to reach stronger solutions at lower evaluation budgets.
-
AI-Driven Research for Databases
Co-evolving LLM-generated solutions with their evaluators enables discovery of novel database algorithms that outperform state-of-the-art baselines, including a query rewrite policy with up to 6.8x lower latency.
-
EurekAgent: Agent Environment Engineering is All You Need For Autonomous Scientific Discovery
EurekAgent achieves new state-of-the-art results on mathematics, kernel engineering, and machine learning tasks by engineering agent environments for autonomous scientific discovery, including a 26-circle packing result at under $11 API cost.
-
PACEvolve++: Improving Test-time Learning for Evolutionary Search Agents
PACEvolve++ uses a phase-adaptive reinforcement learning advisor to decouple hypothesis selection from execution in LLM-driven evolutionary search, delivering faster convergence than prior frameworks on load balancing, recommendation, and protein tasks.
-
Grokability in five inequalities
Five improved inequalities were found with AI help: better Gaussian perimeter bounds for convex sets, sharper L2-L1 moments on the Hamming cube, a strengthened autoconvolution inequality, improved g-Sidon set bounds, and an optimal balanced Szarek inequality.
-
Training-Free Test-Time Contrastive Learning for Large Language Models
TF-TTCL lets frozen LLMs adapt online by distilling textual rules from contrastive reasoning trajectories generated via multi-agent augmentation and applying them through retrieval-based steering.
-
Algorithmic algorithm development with LLMs: A Case Study on LLM-Usage for Contraction Order Optimization in Tensor Networks
Case study applies verifier-guided LLM evolutionary agents to contraction-order optimization in tensor networks and concludes that human validation remains essential.
- Evolving Ensemble of Agents