AI Coding Live News

Anthropic's Alignment Science Blog published research in which many Claude agents iterate in parallel to write fine-tuning data that evades a training-data monitor while making a small research model misaligned. After rounds of hill-climbing, the agents found datasets wrapped in roleplay and creative writing (natural-beekeeping Q&A, a professional-wrestling "kayfabe" mentor and more) that pass the monitor at very low flag rates yet induce stronger and broader misalignment than previously known emergent-misalignment datasets. Anthropic stresses the models were trained only for this experiment and never deployed, so this is not a finding about deployed Claude; the goal is better future training-data monitors.

Zoom

NVIDIA AI announced Nemotron 3.5 Transcribe on Oct 28, with weights on Hugging Face and ModelScope. It is an offline, instruction-driven speech model whose single checkpoint covers transcription in 40 languages, translation between 31 languages and English, speaker attribution in 13 languages, and audio question answering, on recordings up to an hour long with up to eight labeled speakers. NVIDIA says it ranks first among open-weight models on the Hugging Face Open ASR English leaderboard at 5.33% average WER, and Artificial Analysis measured a 2.8% AA-WER, tied for first among open models with Voxtral Small, which uses 24B active parameters. The weights, the Granary v2 training data and fine-tuning recipes are all released.

Zoom

On Oct 20 Bria released Fibo 2.0, putting image editing and text-to-image into a single 8B model for the first time, with two open-weight checkpoints: Fibo 2.0 Dev (full base for fine-tuning) and Fibo 2.0 Edit Turbo (4-step distilled editing). Artificial Analysis ranks it #1 among open-weights models on AA-Image-Editing v2.1 and statistically tied with Qwen-Image-2.1 for the open-weights lead on AA-Image-T2I v2.1. Weights are CC BY-NC 4.0 (commercial use needs a Bria agreement); the Bria API costs $32 per 1k images.

Zoom

While supervised fine-tuning (SFT) distills agentic tool-use capabilities from frontier LLMs into small language models (SLMs, 0.6B to 8B parameters), student agents frequently suffer from 'progress stagnation'—collapsing into repetitive, unproductive tool-calling loops as interaction trajectories extend. In arXiv:2610.19439, researchers introduced Progress-Aware Distillation (ProD). ProD defines a rigorous stagnation scoring formulation that dynamically identifies and penalizes stagnant behavior during student trajectory rollouts via an adaptive SimPO-inspired margin. Experiments across 8 diverse SLMs prove ProD effectively eradicates repetitive loops while boosting task completion efficiency and success rates.

Zoom

Current software engineering benchmarks and training pipelines predominantly evaluate coding agents on localized bug fixing within existing repositories (such as SWE-bench) or isolated code snippet generation. They fail to teach agents how to explore unfamiliar reference systems, formulate architectural blueprints, and reproduce complete, multi-file software projects from scratch under long-horizon constraints. Researchers introduce ProgramGym (arXiv:2610.13395, repo: github.com/Acm1011/ProgramGym, dataset: ProgramGym/programgym-1648), an automated framework scaling verifiable training environments for program reproduction agents. ProgramGym consists of Program-Env, a feedback-driven engine that converts open-source repositories into verified runtime sandboxes paired with black-box test suites across C, C++, Go, and Rust (generating ProgramGym-1648), alongside Program-Agent, an iterative workflow enforcing role isolation across Analysis, Implementation, and Review stages with revision management. Fine-tuning Qwen3.8-27B on ProgramGym trajectories raises ProgramBench test pass rates from 34.87% to 42.84% (+7.97 points), outperforming GPT-5.4 under mini-SWE-agent (37.70%). Coupling SFT with the Program-Agent scaffold boosts performance to 56.25%, eclipsing Claude Opus 4.7 (50.90%) and Gemini 3.5 Flash (53.60%), with datasets and codebase open-sourced.

Zoom

When model capabilities exceed human expert evaluative bandwidth, achieving Recursive Self-Improvement (RSI) without external verifiers or compilers becomes the paramount frontier of AGI. While reinforcement learning with verifiable rewards (RLVR) thrives in formal code and math, open-ended research and complex reasoning tasks lack scalar ground truth, and solitary models evaluating their own outputs suffer from homogeneous confirmation bias. Sakana AI introduces MASS (arXiv:2610.12176, project: sakana.ai/mass), an RSI framework anchored in Multi-Agent Self-Supervision. Driven by a single base model (such as Qwen3.6-27B), MASS tasks the model with concurrently acting as task solver, workflow optimizer, and self-evaluator. The framework alternates between evolutionary multi-agent workflow discovery and supervised fine-tuning (SFT) over synthesized trajectories. Across two successive MASS self-evolution cycles, performance per output token increases by 1.2x to 1.6x across open-ended research suites (MLR-Bench, ScienceAgentBench, DSBench, AstaBench). Head-to-head win rates against the base model surge from 53.9% in round 1 to 69.9% in round 2, while a student model trained on multi-agent traces decisively beats a single-agent baseline trained on 1.4x more tokens.

Zoom

Enabling service robots to execute open-ended human navigation requests across unfamiliar environments (such as locating reading glasses on a living room table and navigating to them) represents a critical challenge in embodied AI. Conventional navigation models rely heavily on domain-specific fine-tuning within specific simulated datasets, resulting in brittle real-world generalization when deployed in unseen layouts. Researchers from Zhejiang University's 3DV Lab introduce SuperNav (arXiv:2610.12126, code: github.com/zju3dv/SuperNav, project: zju3dv.github.io/SuperNav), an open-source agentic navigation framework that achieves universal scene navigation with zero navigation-specific fine-tuning. SuperNav decouples high-level cognitive reasoning—delegated to a frozen, off-the-shelf Multimodal Large Language Model (MLLM)—from low-level physical control executed by specialized robotic navigation tools. Utilizing a Unified Visual-Point Interface where the MLLM marks normalized coordinates directly on observation images, the agent revises decisions continuously based on visual feedback. Validated in the HM3D virtual benchmark and deployed on a real Unitree Go2 quadruped robot across multi-object and demand-driven navigation tasks, SuperNav decisively outperforms fine-tuned baselines.

Zoom

Terminal coding agents are expanding from standard software engineering into scientific domains like bioinformatics, materials science, and physics. However, building training sandboxes requires executable reference behavior and rigorous verifiers capable of distinguishing true semantic correctness from superficial syntactic mimicry—a process that currently requires tedious manual engineering per task. Researchers from Tsinghua University and Tencent introduce Software-in-the-Loop Reconstruction (SWR, arXiv:2610.02710), a self-supervised environment scaling framework. SWR extracts ground-truth outputs and verifiers from existing executable scientific software workflows by perturbing input configurations into public observations and hidden evaluations. Agents synthesize complete, editable programs from input-output schemas without viewing source code, evaluated by hierarchical verifiers testing semantic fidelity, structural integrity, and anti-shortcut compliance. Implemented across 500 workflows and 46 software suites spanning six scientific domains, SWR generates 1,422 verified trajectories. Supervised fine-tuning of Qwen3.8-27B elevates Terminal-Bench 2 average success from 47.94% to 53.56%, establishing a scalable paradigm for scientific terminal agent training.

Zoom

Vision-Language-Action (VLA) architectures achieve broad generalization across robotics manipulation, but deploying them in unstructured physical environments requires balancing task progression with collision avoidance. Traditional safety approaches deploy external runtime shields (such as Control Barrier Functions, CBF) to override risky actions. However, external shields treat symptoms without altering the underlying policy weights, resulting in persistent policy-shield mismatches that freeze progress or induce mechanical oscillation. Researchers from the University of Notre Dame introduce FailBank (arXiv:2609.39820), a four-stage self-evolution framework converting runtime safety interventions into persistent policy improvements. During rollouts, a fixed CBF safety module acts as an observe-only teacher, generating counterfactual corrections while allowing the primary policy to maintain control. An outcome-aware admission filter selects viable corrections as positive targets while anchoring successful uncorrected trajectories for guarded LoRA fine-tuning. Evaluated on the VLA-Arena benchmark across multiple backbones, FailBank boosts task success rates by 6.9 to 8.5 percentage points while reducing collision costs by 23.8% to 35.6%. Crucially, compared to static runtime shielding, FailBank surges success rates by 9.5 to 25.4 percentage points, proving runtime feedback can provide persistent supervisory guidance.

Zoom

While Large Language Models rely on deep non-linear activation layers, a groundbreaking mechanistic study titled 'Your Transformer Can Hold Two Thoughts at Once' (arXiv:2609.29845) demonstrates that transformer architectures exhibit fundamental linear superposition. When inputs from two distinct textual contexts are linearly combined in latent space, the model outputs a superposition of the individual next-token probability distributions. The authors reveal this superposition is an intrinsic architectural property that decays during pretraining but can be restored via lightweight fine-tuning. Furthermore, the team introduces a guided decoding procedure that cleanly disentangles superposed representations, enabling the simultaneous generation of two coherent, independent continuations from a single forward pass.

Zoom