Code Generation & Program Synthesis - Weekly Roundup

This work takes an RSI-inspired approach at the harness layer, scaling auto-research loops across increasingly numerous and diverse environments for harness rollouts, and yields reusable improvements that transfer beyond their development setting.

This work identifies and formalizes this failure mode, which is term destructive resource preemption: obtaining the resources required for a requested task by terminating, overwriting, evicting, or degrading an incumbent task.

The value of execution-improved software as a resource for acquiring new policies in this setting is demonstrated, although initial references remain better on two target tasks when averaged across runs.

The red-teaming methodology and highlighting new attack vectors aim to help defenders evaluate their mitigations against the possibility of persistent malign coding agents.

ContrAgent is presented, a contract-based framework for symbolic temporal supervision of LLM agents that matches state-of-the-art LLM-judge and rule-based guardrail baselines while producing deterministic, reproducible verdicts and, in the online mode, orders-of-magnitude lower per-call latency.

This work takes an RSI-inspired approach at the harness layer, scaling auto-research loops across increasingly numerous and diverse environments for harness rollouts, and yields reusable improvements that transfer beyond their development setting.

This work identifies and formalizes this failure mode, which is term destructive resource preemption: obtaining the resources required for a requested task by terminating, overwriting, evicting, or degrading an incumbent task.

The value of execution-improved software as a resource for acquiring new policies in this setting is demonstrated, although initial references remain better on two target tasks when averaged across runs.

The red-teaming methodology and highlighting new attack vectors aim to help defenders evaluate their mitigations against the possibility of persistent malign coding agents.

ContrAgent is presented, a contract-based framework for symbolic temporal supervision of LLM agents that matches state-of-the-art LLM-judge and rule-based guardrail baselines while producing deterministic, reproducible verdicts and, in the online mode, orders-of-magnitude lower per-call latency.

This work uses teacher-successful problems to define a cheap reference for what the student can learn and measures how the likelihood of each observed token in trajectories from teacher-failed problems changes as an operational learnability signal, which can be computed once from stored trajectories and model checkpoints.

The model family shows gains in held-out scientific-code repair and across selected general-purpose benchmarks in code, reasoning, and knowledge, providing evidence of positive transfer from scientific experience to broader capabilities.

ReFigBench, a benchmark and evaluation framework built on 1,000 real overview figures retrieved from arXiv papers with full provenance, is introduced, exposing the tension between fidelity and editability as the central challenge for practical multimodal document agents.

It is observed that the natural language of problem statements affects LLM-based code generation performance and translation improved Accuracy, but its effectiveness was not consistent across datasets and model types.

ROSETTA is proposed, a hybrid CKKS/TFHE framework that overcomes inefficiency in evaluating nonlinear operations, which incur substantial overhead and dominate the decode stage and achieves up to $4.8\times$ Softmax speedup and $1.5$--$2.1\times$ end-to-end speedup over the SOTA framework CacheMir.

ExecuCritic, a joint training framework in which a coder and a critic are updated on the same execution rollouts, improves over GRPO without a critic, prompted reviewer systems and scalar reward model baselines, while requiring fewer policy gradient steps and fewer sandbox executions.

This work shows that the frozen agent LLM already carries the routing signal in its own forward passes, and that two linear maps suffice to read it out with no skill text in the context, and shows that Gavel (Glance And Verdict from a frozen LLM) reads it in two steps.

HazardAuditor is introduced, an execution-grounded framework that runs heterogeneous agents in controlled environments and normalizes their interactions into a canonical event representation for cross-framework supervision, and observes that token-level post-training objectives create a structural mismatch for generative guards, causing longer rationales to dominate gradient updates.

Empirical evidence is provided from an industrial evaluation that combining specialized agent skills with context-specific knowledge yields accurate, practically useful code reviews.

A large-scale empirical study of over six million packages across six major ecosystems to understand how prevalent cross-ecosystem packages are among all packages, whether there are distinct source code architectural patterns that cross-ecosystem packages use, and whether there are correlations between architectural patterns and project health metrics from GitHub.

This work introduces, for the first time, a personalized bilevel optimization framework that formalizes cloud-edge LLM collaboration as a dual structure: the upper level optimizes edge-side personalization, while the lower level governs cloud-side knowledge transfer, reaching cloud-edge evolving in coordination.

Experiments show that the specialized LUpp solvers proposed in this work can outperform existing state-of-the-art approaches for this class of workloads, with speedups of up to 6.5x over cuSolverDx, and up to 17.7x over MAGMA.

Registers are implemented as dedicated fixed-position tokens whose continuous hidden states are trained to carry reasoning progress across generation chunks, which are especially effective for bounded code generation, where correct programs usually span several chunks.