AI News

Large language model (LLM) agents can benefit from reusable skills distilled from prior task experience, yet existing skill optimization methods often rely on costly execution-based evaluation and substantial task data. We introduce \textbf{COBRA-Skills}, an efficient framework that formulates skill optimization as budgeted sequential optimization over a dynamically evolving candidate space. COBRA-Skills couples contextual-bandit-guided prioritization with evidence-grounded skill evolution, selectively allocating evaluations to promising or informative candidates while continually refining the skill population from execution feedback. Across six heterogeneous agent benchmarks and three target models, COBRA-Skills consistently achieves the strongest average performance among compared methods, while reducing optimization cost by 55--58\% relative to SkillOpt and using only 50 unique optimization examples per benchmark. Further analyses show that COBRA-Skills remains robust to changes in the agent harness and performs effectively when the target model itself is used for skill generation and refinement.

Post-training attention sparsification reduces the quadratic cumulative attention cost of pretrained Transformers by selecting a small set of context units (tokens or blocks) for each query. Existing trainable methods usually use a lightweight selector to score context units, followed by hard Top-K selection that blocks gradients from the language modeling loss. Consequently, these methods commonly distill layer-wise dense attention distributions. Although this encourages the selector to rank context units by dense attention weights in the original model, the ranking is not directly aligned with their impact on predictions under a fixed attention budget (i.e., the number of attended context units per query), potentially wasting the limited budget on less useful units. To address this misalignment, we propose Simple Attention Sparsification (SAS), a gated sparse attention mechanism that optimizes context ranking end-to-end with the language modeling loss. The key idea is to inject the selector's continuous scores into attention logits during training, allowing the loss to update the selector through standard backpropagation. We identify several choices crucial for this simple design to work well in practice: placing the gate inside the attention softmax in log form, using normalized softmax gates to calibrate historical context against the always-retained current block, and preserving continuous selector scores so the model learns relative priorities rather than only hard selections. To support long-sequence training, we implement a memory-efficient Triton kernel that integrates SAS into FlashAttention-style computation. Across reasoning, long-context understanding, and agentic tasks, SAS consistently outperforms trainable sparse attention baselines across attention budgets, with especially large gains under tight budgets, demonstrating more effective context ranking for downstream tasks.

The company also announced a product called Active Compute Fabric, a network technology that targets the fact that much GPU time is wasted waiting for data to arrive.

LLM switchboard for Claude Code, Codex and AI agents — pool providers, cloud credits, regions and local models behind stable routes. - swobuforge/swobu

Pre-built synthetic environments for Salesforce, Jira, Stripe, and 35+ more. We generate the dataset, 200 adversarial queries, and the correct answer for every one.

Replay Doctor reads the transcripts your coding agent already keeps on disk and names the turn the provider re-read everything at write prices. Every figure carries its population and its date. No account. Your transcripts never leave your machine. Source available.

The faster your team ships with AI, the harder it becomes to keep production stable. Not anymore. Coldtea is an agentic IDE where your coding agents build, visual QA agents catch regressions, and AI monitoring watches production, so your team can keep moving at agent speed without breaking anything

Kombai is your AI Design Engineer to design and code standout (not slop) websites and product UI's.

litellm without the bloat. Contribute to kennethwolters/litelm development by creating an account on GitHub.

Join the discussion on this paper page

Join the discussion on this paper page

Training capable cyber agents is often treated primarily as a problem of model scale, yet open-weight post-training is constrained more directly by the cost of executable environments, reliable multi-turn supervision, and access to strong teachers. We present a data-centric framework that addresses these bottlenecks through five complementary systems: Choulea analyzes hidden reasoning signatures, SkyReal reduces teacher-sampling cost, Hongzwang bypasses API restrictions on teacher execution, PSBreakup restores capabilities weakened by model merging, and Kreator converts expert interventions into trainable reasoning. Our data engine constructs resettable coding, vulnerability, CTF, kernel-history, full-exploit, firmware, and device-backed environments. Candidate trajectories are retained only after execution verification and evidence auditing, yielding 164,269 trajectories for long-context supervised fine-tuning. The three checkpoints improve over their starting models by an average of 23.76% on the full CyberGym suite and 10.49% across the pooled CTF suites. As of September 1, 2026, Feyospace-s1 achieves a verified success rate of 63.24% and ranks 10th on the official CyberGym leaderboard, while all three checkpoints rank 1st among models at comparable parameter scales. To our knowledge, this is the first end-to-end demonstration that a seven-person independent team can train open-weight models with leading agentic cyber capability.

Robot foundation models achieve strong in-distribution performance but often degrade under visual distribution shifts. When learning to generate actions from pretrained visual representations, models may exploit task-irrelevant visual cues that correlate with demonstrated actions within the training distribution. Such vision-action shortcuts can undermine generalization when these correlations change under distribution shifts. Mitigating these shortcuts requires constraining how visual information is used for action generation while preserving task-relevant spatial information. We propose Latent Interface Training (LIT), a framework-agnostic two-stage strategy that first establishes a spatial-goal-conditioned action prior without images, then constrains visual conditioning through a pose-supervised latent interface. Stage 1 trains the action expert to generate action chunks conditioned on language, robot state, and each demonstrated chunk's terminal SE(3) end-effector pose, learning goal-directed action generation independently of visual cues. Stage 2 introduces a latent interface that aggregates visual and semantic representations and serves as the pretrained action expert's only visual conditioning pathway. The interface is supervised to reconstruct the terminal pose previously used to condition Stage 1, encouraging it to retain the goal-relevant spatial information needed for action generation. Across four vision-language-action and world-action architectures (Pi0.5, MolmoAct2, FAST-WAM, and ImageWAM), LIT improves overall LIBERO-Plus success by 3.87-10.70 percentage points while preserving or improving average LIBERO success. Real-world evaluations show 13.30-16.70 percentage-point gains in success aggregated across three tasks under unseen camera configurations, lighting variations, and distractors.

GPT-6 Astra tops the ARC-AGI-3 leaderboard while cutting token costs. Here is what matters for developers: 🚀

⚡ Tied with Claude Fable 5.1 on Artificial Analysis’ Intelligence Index
⚡ Asynchronous tool calls for parallel execution
⚡ Retained reasoning memory across API calls

Efficient context man…

High-performance, zero-dependency hybrid vector search & BM25 lexical engine with IVF-Flat ANN indexing in pure Python.

A native macOS command center with on-device AI, window management, clipboard intelligence, and dictation. Free forever. Pro is $29.99 once, for two Macs.

This is pretty neat: ChatGPT Work and GPT-6 Astra (I used "Max") can take an address and produce a 5K/10K circular running route starting from that address, using OSM data https://t.co/GVxIqlqThC

UBTECH has commissioned a 14,000-square-meter smart manufacturing facility in Guangxi engineered to produce a humanoid every 10 minutes, scaling up the industry’s hardware volume race.

Scale your B2B outbound sales on data you can trust. With Tables you can access the most accurate contact data on the market in minutes at a fraction of the cost. 300M Lead Database → 98% Contact Accuracy → 7-day Refresh Rate → Waterfall Enrichments → Automated AI Prospecting. Lead sourcing, qualification and enrichment in one place.

Airuncode is a local-first agent runtime for serious software work. Run multiple coding agents on your machine, bring your own API keys, switch between cloud and local models, and pay providers directly with zero token markup. It scans your codebase, debates solutions across agents, edits files, runs tests and self-heals failures. It also ships with V-CORE, a native Vulkan 3D runtime for AI-assisted game development. Available on Windows, macOS and Linux.

Humanoid Daily is your sharpest daily briefing on humanoid robots, AI breakthroughs, and the machines reshaping the world.

Join the discussion on this paper page