30
What Does an LLM-Agent Leaderboard Rank Actually Compare?
A methodological study shows close LLM-agent leaderboard rank gaps on SWE-bench and similar benchmarks often do not support superiority claims.
The paper defines an estimand-aware pairwise procedure for comparing agents, checking common support and applying explicit uncertainty rules and practical margins. Across SWE-bench, AgentRewardBench, and tau2-bench, close rank differences are frequently unresolved, and proxy labels or utility rules can change which system is selected. The authors argue a leaderboard score summarizes a released evaluation but does not by itself justify pairwise superiority conclusions.
20
30
30
30
30
30
30
DeepSeek-v4.1 Flash: Pushing the Limits of KV Cache Compression
DeepSeek-V4.1 Flash is a 552B-parameter multimodal MoE model with 1M-token context achieving 4x KV cache compression for long-horizon agent workloads.
A detailed analysis of the DeepSeek-V4.1 Flash technical report describes a 552B-parameter multimodal mixture-of-experts model supporting contexts up to 1 million tokens. Its Causal Encoder-Decoder (CED) architecture activates 8B parameters during prefill and 16B during decode, and reportedly delivers about 420 tokens/s. Joint optimization of architecture (CSA2 cross-layer compression), FP4 KV cache precision, and deployment strategy cuts runtime KV cache to roughly 1/4 and persistent KV cache to about 1/8 of DeepSeek-V4-Flash at the same sequence length, targeting storage and bandwidth bottlenecks in long-horizon agent serving. The author notes all DeepSeek-V4 Pro models were taken offline following the release.
74
s-MDM: Generative Virtualization of Multi-Device Hardware Variations for Portable DL-SCA
Researchers present s-MDM, a generative framework synthesizing virtual device profiles to improve cross-device portability of deep learning side-channel analysis.
The poster introduces the Synthetic Multiple Device Model (s-MDM), a zero-target-trace generative framework addressing performance degradation of deep learning side-channel analysis on unseen hardware. It combines a structured cVAE generator, Walsh-Hadamard leakage anchors, continuous style modulation, and decoupled leakage-style-domain critics to synthesize virtual source-device profiles offline. Benchmarked on 32-bit AES_PTv2 traces, s-MDM achieves consistently low key rank on layout- and acquisition-shifted Pinata targets where physical baselines are unstable.
20
The Model Proposes, the Code Disposes: A Pre-Registered Ablation of a Verifier-and-Acceptance Stage in an LLM-Orchestrated Offensive-Security Agent
Pre-registered ablation finds a model verifier stage in an LLM offensive-security agent suppresses findings; removing it eliminated suppression with precision tradeoff.
The paper evaluates a verifier-and-acceptance stage in an LLM-orchestrated offensive-security agent via a pre-registered 20-run confirmatory ablation and a 2x2 factorial study with 40 runs on vulnerable lab targets. Removing the stage eliminated pre-report suppression (median 2 vs 0 findings, p = 0.00003) but reduced model-blinded shipped precision (0.471 vs 0.353, p = 0.0087). Suppression was attributed to the model verifier rather than deterministic acceptance rules, and an instrumented canary recorded zero external contacts in all 60 runs. The full design retained 93.8% of model-adjudicated true candidates but failed its pre-registered non-inferiority floor of 0.90.
38
Label-Guided Knowledge Distillation for 3D-CNNs in Action Recognition
LGKD uses ground-truth labels to guide feature distillation for 3D-CNNs, combining sample-wise and class-wise distillation for action recognition.
The paper proposes Label-Guided Knowledge Distillation (LGKD) for 3D-CNNs, noting that most video feature distillation methods are simple adaptations of image techniques that neglect temporal-dimension differences. LGKD combines sample-wise distillation, which uses label information and the teacher's probability distribution to guide features impacting temporal accuracy, with class-wise distillation employing a prototype network to capture relational knowledge among same-category samples. Experiments on the UCF101 and HMDB51 action recognition benchmarks achieve competitive results.
15
Evaluating Context Segmentation in Locally Deployable SLMs for Cybersecurity CTF Tasks
Context segmentation framework boosts memory-constrained gemma-4 agents on picoCTF, solving 18.52% of tasks standard execution fails, highlighting local SLM offensive risk.
The paper introduces context segmentation, a two-level agentic framework that divides long-horizon CTF exploitation tasks into contextually isolated sub-problems to counter context bloat and cognitive degradation from accumulated tool-call outputs. It evaluates memory-constrained gemma-4 models on the picoCTF dataset; the E4B model achieves competitive rewards with superior token efficiency compared to brute-force retries. It solves 18.52% of tasks that standard agentic execution fails to complete. The work frames locally deployed open-weight SLMs as an escalating risk since they bypass proprietary API guardrails; code is released on GitHub.
32
Cyber Command turns to veteran of intelligence agencies for top AI role
US Cyber Command appoints Rear Adm. Ronzelle Green as chief AI officer as its AI budget grows from $5M to $138M.
US Cyber Command has named Ronzelle Green, previously head of research and development at the National Geospatial-Intelligence Agency, as its new Chief Artificial Intelligence Officer, replacing Brig. Gen. Reid Novotny. The command's 'AI for Cyber Operations' budget request grew from $5 million in fiscal 2026 to $138 million in fiscal 2027, aiming to let cyber operators process data and respond to threats faster than humans alone. Sources say Green's priority will be consolidating fragmented AI pilot programs across the organization, in contrast to the NSA's more established Artificial Intelligence Security Center created in 2023.
40
Why the current tech backlash feels different
The Verge's Decoder mailbag discusses the current tech backlash, arguing AI hype overstates verifiability outside software engineering.
Nilay Patel's Decoder mailbag episode addresses listener feedback on the widely discussed 'software brain' essay. He argues AI hype is concentrated on software because code is verifiable through compilation, while domains like drug discovery, math and science lack equivalent verifiability. The episode also touches on AI backlash, surveillance, data centers and upcoming midterm coverage.
18
Meta debuts its Muse AI agent. Will consumers trust it?
Meta launched Muse, a consumer AI agent powered by Muse Spark that connects to users' apps to execute tasks like emailing, booking travel, and payments.
Meta introduced Muse, a personal AI agent for US users that connects to email, calendars, payments, shopping, and other services to execute tasks such as booking travel, lowering bills, and completing purchases via Link by Stripe. The agent runs in a dedicated Muse Secure VM with a separate Sentinel agent kept apart at the system level, and Meta claims it cannot see passwords or payment data and does not share conversations with ad systems. Muse is free to start, with Power ($20/month) and Maximum ($100/month) subscription tiers, and is available on the web, iOS, Android, and WhatsApp, with Meta AI glasses support planned. The launch follows Meta's $18 billion multistate consumer-harms settlement and comes as rivals like Gemini Spark and Claude Cowork push agentic AI.
74
Difficulty-Adaptive Tree-Structured Policy Optimization for Expanding Reasoning Coverage in RLVR
Researchers propose DATPO, a difficulty-adaptive tree-structured RLVR training method that expands reasoning coverage (pass@k) and improves test-time scaling on math benchmarks.
The paper identifies three rollout design principles for RLVR: difficulty-adaptive rollout expands pass@k, tree-based rollout beats parallel sampling, and sentence-entropy-guided forking overcomes token-level branching localization. DATPO combines difficulty-adaptive tree search with a sibling-diversity advantage term to promote semantic diversity during training. On mathematical reasoning benchmarks, DATPO outperforms baselines in pass@k, directly translating to superior test-time scaling performance.
30
The democratization of cyber warfare — and what it means for CISOs
CSO Online argues AI is democratizing cyber warfare, citing Taiwan's first largely autonomous AI-driven attack that hit 85 government accounts.
CSO Online argues AI is accelerating the democratization of cyber warfare by collapsing cost and skill barriers, with effects already reaching the private sector. It cites the August disclosure of the first largely autonomous AI-enabled attack on Taiwan government infrastructure, where up to eight agents operating simultaneously compromised at least 85 government accounts and exfiltrated more than 2,500 personnel records before expanding to the nuclear safety agency and energy companies. The piece places this in a historical arc from crossbows to cheap drones and cites Kane Gamble's 2015 solo social-engineering compromise of senior US intelligence officials as evidence that low-resource attackers were already viable.
25
BeaconKV: Key-Value Cache Compression Guided by Beacon Queries for Efficient Large Reasoning Model Inference
BeaconKV introduces training-free KV cache compression using beacon queries, cutting long-reasoning inference memory up to 5.8x while preserving accuracy.
The paper shows recency-based KV cache compression assumptions fail in long-horizon reasoning because Thought Revisiting Tokens (TRT) re-attend to distant context such as early task-solving plans. TRT queries cluster into a small number of similarity groups, which BeaconKV exploits by maintaining compact beacon query representatives to anticipate revisited KV pairs without storing full query history. The training-free method achieves up to 5.8x memory reduction and over 4.3x throughput improvement across four open-source large reasoning models while nearly preserving full cache accuracy.
33
Learning 3D Editing without Paired Supervision via Generative Prior Distillation
New framework distills 2D editing and VLM priors into a feed-forward 3D editing model without paired 3D training data.
The method, PriorEdit3D, learns feed-forward instruction-guided 3D editing by distilling knowledge from foundation models instead of using ground-truth 3D pairs. Through a differentiable rendering pipeline it supervises a 2D visual prior from an image editing model at the main view and a Vision-Language Model semantic prior at novel views for instruction fidelity and identity preservation. A 3D-aware Distribution Matching regularization constrains outputs to the manifold of realistic 3D assets defined by a pretrained image-to-3D teacher. Experiments report superior instruction fidelity and cross-view consistency over state-of-the-art baselines, with code released on GitHub.
30
[AINews] Muse Spark 1.3 matches GPT-5.6-Sol, confirming Meta Superintelligence as the newest Frontier Lab, >90% discount for training
Meta's Muse Spark 1.3 reportedly ranks as the world's #3 model, matching frontier models from OpenAI and Anthropic with planned open weights.
The Latent Space AI News roundup leads with Muse Spark 1.3, promised in Zuckerberg's letter, which ranks #3 worldwide per AAII, is slated for open weights, and uses a pricing model over 90% cheaper when users opt in to training. The issue also covers the rumored Gemini 3.8 Flash launch and analysis arguing OpenAI's rumored looped-transformer 'Astra' architecture is a modest tweak rather than a breakthrough. Additional coverage includes ByteDance Seed's HarnessDev harness-evaluation benchmark, a retrieval-invoked actual-use evaluation method, Stanford's revamped agent engineering curricula, and Photon 2.1 adding TTS models and NVIDIA B200 support.
74
Enoki: Efficient Multi-Level Hallucination Detection
Researchers introduce Enoki, an open information extraction framework unifying claim-level and span-level hallucination detection in LLMs at lower inference cost.
Enoki extracts text-anchored relational facts, verifies them against evidence, and projects unsupported facts back onto hallucinated spans, so claim-level verification and span-level localization share one representation without separate alignment. It supports LLM-based, encoder-based, and rule-based extraction regimes to balance accuracy against inference cost. Experiments show it stays competitive with strong claim-level systems while using fewer resources and outperforms them on fine-grained span- and entity-level localization. The authors also release EnokiQA, a dual-granularity dataset with aligned claim-level verification and span-level localization annotations.
50
To See a World in a Living Context: Unified Indoor-Outdoor Urban World Generation
Researchers introduce HoloWorld, a unified text-driven framework generating coherent indoor-outdoor 3D urban worlds, improving average AQS over SOTA by 7.68%.
HoloWorld is a text-driven 3D generation framework that unifies indoor and outdoor urban world generation using a continuously updated cross-scale world context. It autoregressively generates urban exteriors with consistent spatial organization, grounded in 3D building instances and footprints, then produces building-specific interiors with geometry-constrained layouts that inherit exterior appearance. The authors claim it is the first framework to unify indoor and outdoor generation within one coherent 3D urban world, reporting a 7.68% average AQS improvement over prior SOTA and the highest average RDR score.
34
unsloth/Qwen3.8-Flash-Next-GGUF — new model trending #21 on Hugging Face
Qwen released Qwen3.8-Flash-Next, an experimental 125B-parameter open-weight MoE previewing the Qwen4 architecture, with Unsloth shipping optimized GGUF quants.
Qwen released Qwen3.8-Flash-Next, an experimental open-weight preview of the architecture planned to underpin Qwen4. The model has 125B parameters with 6B activated, 512 experts (10 routed plus 1 shared), Qwen Sparse Attention (QSA), Gated DeltaNet, Gated Residual, and n-gram embeddings, with 262,144-token native context extendable to 1,000,000 tokens. Unsloth provides Dynamic 3.0 GGUF quantizations, and multi-token prediction (MTP) delivers 1.3-1.7x faster inference via llama.cpp or Unsloth Desktop.
76
Arrested man allegedly impersonated NSA elite hacking unit, Supreme Court chief justice
Colorado man Joshua Culver indicted for impersonating NSA's Tailored Access Operations chief and Supreme Court Chief Justice John Roberts in Indiana court cases.
Joshua Culver, also known as Maverick Young, was arrested in Colorado after a July Indiana indictment on four counts of falsely impersonating an officer of the court and one count of using a forged judge's signature. He allegedly posed as an NSA officer in September to pressure the Tippecanoe County sheriff's office, and later presented a forged document purportedly from the head of the Tailored Access Operations unit demanding case dismissal and warrant quashing. The indictment also alleges he used a forged signature of Chief Justice John Roberts on a dismissal order in Grant County, Indiana.
25
GLM-5.3: How Chinese labs keep stride with the frontier
Z.ai released GLM-5.3, a ~750B-parameter model with frontier agentic coding scores, with open weights on Hugging Face planned in two weeks.
Z.ai announced GLM-5.3, initially available only in its coding plan, with API access and open Hugging Face weights promised within two weeks. The roughly 750B-parameter model, one-third the size of Moonshot AI's Kimi K3, surpasses Kimi K3 on many benchmarks and beats Claude Fable 5 or GPT-5.6-Sol on some, placing it at the frontier of agentic coding benchmarks. GLM-5.3 reuses the GLM-5.2 base model with substantially extended post-training based on more RL environments, more diverse tasks and more compute. The post also analyzes how Chinese labs keep pace with the frontier, arguing release speed matters more than distillation.
80
30
30
30
30
50
30
30
30
30
30
30
30
30
30