Skip to content
AI NEWS JOURNAL
AI, IN PERSPECTIVE

Search AI news

Search headlines, summaries and articles to connect the dots.

Newest first

AttentionSwitch: Jailbreak Defense for LLMs via Joint Learning of Sparse Attention-Head Selection and Scaling

arXiv:2611.08752v1 Announce Type: new Abstract: Large language models have achieved remarkable performance across a wide range of tasks, yet they remain susceptible to jailbreak attacks that circumvent safety mechanisms and elicit harmful or unintended outputs. Although safety can be improved through model fine-tuning, such approaches can be computationally expensive and may degrade previously acquired capabilities through catastrophic forgetting. Recent jailbreak defenses have therefore explored targeted interventions on specific model components associated with unsafe behavior. However, existing approaches typically decouple two inherently related decisions: where to intervene and how to modify the selected components. Such sequential optimization may be suboptimal because the optimal intervention location can depend on the intervention applied there, and vice versa. In this work, we introduce AttentionSwitch, the first jailbreak defense to jointly optimize intervention sites and strengths within a frozen LLM. Specifically, AttentionSwitch learns a sparse set of attention heads together with query-conditioned multiplicative scaling factors, allowing the outputs of the selected heads to be either attenuated or amplified without altering the directions of their representations. Across popular language models and jailbreak attacks, AttentionSwitch outperforms existing defenses, with particularly strong results when training data is scarce.

Forcing a Letter Suppresses Refusal: Mechanistic Evidence from WMDP

arXiv:2611.08639v1 Announce Type: new Abstract: Safety benchmarks for large language models (LLMs) commonly use multiple-choice question (MCQ) format. However, prior studies documented that MCQ format reduces refusal rates, which may significantly affect the safety behavior of the model. We provide a first mechanistic explanation of this effect on the Weapons of Mass Destruction Proxy (WMDP) dataset across two models - Llama-3.1-8B and Qwen3-8B. First, we use the tuned lens technique to show that when processing MCQ the model commits to an answer-letter token very quickly, which is in contrast to the gradual refusal build-up in open-ended formats. Additionally, we observe that activations on the MCQ format lie roughly 3x further from the refusal direction, compared with their open-ended counterparts. Moreover, ablating the attention heads that carry the MCQ format answer increases the refusal, which is a direct evidence that the MCQ answer-selection competes with refusal. These findings indicate that MCQ safety scores primarily measure knowledge capabilities and do not assess the safety-mechanisms.

Paper Unicorns Are Cheap. Building A $1B Revenue AI Company Takes A Multimarket Cap Table

AI startups should be judged by durable revenue, full-price renewals and healthy margins after delivery costs, with embedded customer workflows providing stronger defensibility as model access gets cheaper, rather than their paper valuation, writes guest author Ilman Shazhaev, founder and CEO of Dizzaract, who suggests that founders should build global investor bases that support customer acquisition and market entry while weighing venture growth expectations against sovereign-backed investors’ localization, governance and other obligations.

ResServe: Accurate and Efficient FP4 LLM Inference with Residue Compensation

arXiv:2611.08713v1 Announce Type: new Abstract: Block-scaled 4-bit floating-point (FP4) quantization reduces large language model inference costs but can degrade task quality. Higher-precision computation can mitigate this loss at the expense of some of the efficiency gains from FP4 inference. Narrow matrix computations common in decoding often leave spare hardware capacity for low-cost precision recovery. ResServe combines an FP4 activation base with an FP4-quantized residue representing information lost during quantization. It uses this spare capacity to compute the residue's contribution and combines results on-chip, avoiding extra result transfers. We apply this principle to linear layers and decode attention, while selective compensation limits added work when processing input prompts. On NVIDIA B200, compensated linear kernels using the NVFP4 format improve numerical precision while retaining latency close to uncompensated execution for the evaluated operations with few rows. In the evaluated NVFP4 inference configurations, ResServe adds at most 3.4% request-latency overhead over W4A4 and reduces request latency by 22.4-53.3% relative to W4A8. Across five models, code-generation accuracy improves over same-format uncompensated inference, with mean relative gains of 15.04% for NVFP4 and 53.57% for MXFP4.

Show HN: Open-source customer-facing agents

Hi, Dmitri from Radioso here, an open-source platform to design AI agents for customers and visitors in any channel, in the style of Fin/Ada/Decagon. The repo is https://github.com/radioso-ai/radioso and our website is https://radioso.aiToday, your choices are:1. Buy a very expensive service for your vertical (usually starting with $$ five figures monthly) 2. Get a cheap off-the-shelf chatbot that mostly answers FAQs 3. Build your own, and then also build the inbox, human handoff, audit trail and integrations around itAt Radioso we wanted an open-source option that combines the best of all 3. You can integrate it with your own software via API/SDK or your CRM, or run it standalone, either self-hosted or in our cloud.The engine runs multi-step interactions and takes actions: calling your APIs, updating records, booking things. Maybe you’ve tried Parlant; Radioso is the alternative “with batteries included”: RAG, audit trail, customer service inbox, and integrations like Slack and email.Under the hood it’s plain TypeScript + PostgreSQL, works with OpenAI/Anthropic/other models, and is licensed under Apache 2.0 which includes all features (The only thing we keep under EE license is the multi-tenant layer we use to run our own cloud).We have existing customers running it for technical Q&A, course booking, online product recommendations and goods returns, among others, some reaching 92% of conversations automated. We’re currently working on voice support so we can offer a call center.Our ambition is to be “the WordPress of customer-facing AI” :), where people can contribute their favorite plugins and benefit from the ecosystem.Longer term, we think browsing the web will mostly mean talking to your own agent, and when you need a service or a purchase, your agent will talk to the vendor’s agent to get things done. For this, Radioso already publishes an A2A Agent Card and we’re working on the rest. Your Radioso services are also callable over MCP, for customers who already use it.Thanks for your attention. We’d especially love to hear which integrations you’d want, whether you’d self-host it, and what’s stopped you from deploying a customer-facing AI agent so far. Tell us what you think. Comments URL: https://news.ycombinator.com/item?id=50423820 Points: 1 # Comments: 0

b12202

The llama.cpp v1.220.2 release introduces enhanced BF16 precision support across multiple components, aligning with reference implementations by moving specific operations like GLM4 family projections, MiMoVL vision matmuls, and Qwen3-TTS speaker encoder conv1d to BF16 where appropriate. Backends such as CUDA, Hexagon, OpenCL, WebGPU, and Vulkan now uniformly handle GGML_PREC_BF16 similarly to F32, while CPU, SpaceMIT, and ET backends extend flash attention support to BF16 with F32 accumulation. Flash attention tests now include BF16 cases, and the KQ multiplication in Llama is updated to request BF16 precision. A TODO remains in clip.cpp for future work.

Show HN: Throng, a project first, terminal 2nd IDE for command line people

I've been beavering away at this since July, and its finally getting to a point where its stabilising.It started Windows only, but I recently converted over to Linux, so I've been ironing out the kinks in the last week. You can install it via apt or rpm, or .exe or .msi on Windows.Many more features to come, including native agent support and more preview features (images, PDF, document formats). Claude works pretty well in it, but I've not tried many other agents yet.I rarely leave throng now - 99% of my work happens inside it. I stopped using Visual Studio around 10 months ago (in favour of VSCode), and now I don't even use VSCode. Comments URL: https://news.ycombinator.com/item?id=50412226 Points: 1 # Comments: 0

b12180

The release adds the Kimi-K3.1-Pro text model to llama.cpp, replacing full-attention MLA layers in K3 with DSA (Dynamic Sparse Attention) layers featuring a lightning indexer. The indexer uses an additive mask over the cache and reuses sparse flash attention paths, while preserving existing KDA layers and MoE components. The model's architecture includes four indexer tensors per DSA layer, hybrid memory for caching, and maintains quantization integrity by excluding indexer parameters from quant overrides. Verified against HF checkpoint with identical PPL beyond context length, and supports multiple platforms including macOS, Linux, Windows, Android, and iOS.

Why a new report on Anthropic automating AI development is alarming researchers

Anthropic is using AI to build more powerful AI. And it appears to be working remarkably well, with the company's models improving roughly twice as quickly as a year ago, with little indication of slowing down. That's according to a new report published Thursday by the AI safety research organization METR, following an 11-week investigation […]

Zyphra ZYFA: FP8 FlashAttention for AMD MI355X, Up to 1.76x Faster

Zyphra released ZYFA, open-source FP8 and BF16 prefill attention kernels for AMD Instinct MI355X. On real model activations, its FP8 kernels run 1.16-1.76x faster than AMD’s AITER, with 62-78% lower error at head dimension 128. The post Zyphra ZYFA: FP8 FlashAttention for AMD MI355X, Up to 1.76x Faster appeared first on MarkTechPost.

b12142

A fix was implemented to verify the combined K+V f16 scratch size in the Flash Attention dequantization path for Vulkan, addressing issue #30048.

LowRAM: Low-Rank Keys and Amortised Values for Composable KV Cache Compression at Test-Time

arXiv:2611.07025v1 Announce Type: new Abstract: Autoregressive decoding relies on the KV cache, whose memory footprint grows linearly with context length and caps batch size and throughput. Existing post-training methods exploit redundancy across tokens, layers, and feature dimensions, or lower numerical precision. We identify a complementary axis: within each attention head, values are largely predictable from their corresponding keys. We exploit this with value cache amortisation, which fits a lightweight per-head predictor at test time and stores residuals only for the tokens it reconstructs worst, giving $4\times$ near-lossless value compression. The predictor can be initialised by meta-learning for substantially reduced prefill overhead, or fitted from scratch in settings that allow a one-off cost, such as multi-query inference. We pair value amortisation with cross-layer low-rank key compression to produce LowRAM, an asymmetric compression pipeline that stays close to lossless across three model families. By reconstructing only a selected subset of KV pairs at each decoding step, LowRAM can keep decoding overhead minimal and achieve up to $5\times$ higher peak decode throughput at $4\times$ compression. LowRAM further composes with quantisation and eviction, reaching $26.9\times$ compression at $94.5\%$ of baseline performance on long contexts. These results show that treating keys and values as structurally distinct enables efficient, aggressive, and composable KV cache compression.

Migrating vLLM to the LibTorch Stable ABI

How vLLM migrated its CUDA build and compiled dependencies to the LibTorch Stable ABI, enabling compatible PyTorch upgrades without rebuilding vLLM.

New NVIDIA Cosmos 3 Open Models for Interactive World Simulation

NVIDIA has introduced four new open-weight checkpoints for its Cosmos 3 model family—Cosmos3-Nano-Transfer-Auto, Cosmos3-Nano-Sim-Auto, Cosmos3-Nano-Sim-Transfer, and Cosmos3-Nano-Sim-Bimanual—enabling interactive world simulation and multi-sensor, multi-view driving scene generation. Each model is a post-trained variant of Cosmos 3 Nano with 16B trainable parameters and is based on OpenMDW1.1, allowing commercial and non-commercial use. The checkpoints support applications in robotics and autonomous driving by generating synchronized camera and LiDAR data, enabling streaming simulations, translating simulator controls into photorealistic video, and predicting visual outcomes from robot actions. They are designed to be adapted using custom data and training recipes, with support for offline data generation, causal distillation, and interactive rollouts. Resources include technical reports, demos, and notebooks for inference and adaptation.

Not All Loops Cost the Same: Phase-Dependent Depth for Serving Looped Transformers

arXiv:2611.07211v1 Announce Type: new Abstract: Looped transformers scale inference-time reasoning by repeatedly reusing a shared core, increasing computation without increasing parameter count. Serving prefilling and decoding with a uniform loop budget couples their costs. We present LoopDuet, a phase-dependent depth and training-free serving approach that selects the two budgets jointly while preserving decoding depth at prefilling boundaries. Our vLLM implementation gathers tokens requiring further recurrence into compact batches, turning the selected allocation into executable savings with existing weights and attention kernels. At runtime, a queue-age controller adjusts prefilling depth to retain quality as demand changes. Experiments on the full HotpotQA dataset with one A100-80GB demonstrate two benefits of this flexibility. First, separate phase budgets improve the quality--throughput trade-off: Huginn delivers 19\% more throughput than the fastest tested uniform setting meeting the same accuracy criterion, while a higher-throughput operating point gains 6.04 EM points over uniform depth eight at 97\% of its throughput. Second, the same execution mechanism retains more quality under load: on two evaluated OLMo arrival traces, adaptive prefilling gains 0.92--1.35 EM points over the most accurate tested fixed-depth controls at the same 100\% attainment of a one-second time-to-first-token target. Together, these results establish LoopDuet as a practical way to turn flexible model computation into higher serving capacity and better quality under load.

NUMA-Torch: Compiler-Driven NUMA-Aware LLM Prefill on Multi-Socket CPUs

arXiv:2611.07105v1 Announce Type: new Abstract: For sufficiently large transformer-based Large Language Models (LLMs), whose parameters cannot be retained in on-chip caches, prefill throughput on multi-socket CPUs is often limited by sustainable DRAM bandwidth rather than peak floating-point throughput. In the standard PyTorch compilation (i.e., TorchInductor) path, model weights are allocated without considering the Non-Uniform Memory Access (NUMA) topology: specifically, the affinity between memory banks and compute cores is disregarded. This causes a substantial fraction of weight accesses to traverse the inter-socket interconnect, severely degrading the effective memory bandwidth available to each socket. We address this bottleneck through a compiler-level solution that makes TorchInductor NUMA-aware. Treating each NUMA socket as an independent tensor-parallel partition, we introduce NUMA-aware Tensor Parallelism (NUMA-TP). This approach transparently shards weights across sockets and schedules compute kernels such that thread pools and memory allocations remain strictly local to their respective sockets using CPU affinity. We evaluate the approach on GPT-2 (117M, 345M, 762M) and BERT (Small, Large), which exercise standard multi-head attention and MLP, as well as Gemma 3 270M, which combines Multi-Query Attention with a GeGLU-gated MLP. By strictly enforcing data locality, our approach successfully mitigates cross-socket traffic, achieving prefill speedups of up to 2.63x on a dual-socket Haswell platform, up to 1.87x on a dual-socket Cascade Lake platform, and up to 2.27x on a four-socket Haswell-EX platform.

WeLM-Image: A Unified Image Generation Model towards Visual Intelligence

arXiv:2611.07392v1 Announce Type: new Abstract: We present \textbf{WeLM-Image}, a unified image foundation model that integrates multimodal understanding into text-to-image generation, instruction-guided editing, and multi-reference synthesis. Designed for single-GPU deployment in WeChat's large-scale consumer applications, the model processes text, reference images, and target latents as one interleaved sequence. A Mixture-of-Transformers (MoT) backbone combines shared attention with independent understanding and generation parameters to limit optimization interference. Equipping each path with Mixture-of-Experts (MoE) modules provides sparse capacity, with 31.7B stored parameters and approximately 5.4B generation parameters activated per target latent token. A generation-oriented data engine combines a knowledge-grounded visual vocabulary with large-scale visual clustering to balance semantic coverage and editing tasks, while training-time evaluation guides iterative data refinement. Bilingual annotations and verified editing supervision support progressive training up to native 1.5K resolution. We optimize generation and editing separately with rubric- and checklist-based rewards, then consolidate the specialized policies through noise-aware on-policy distillation. In the reported overall comparisons, the 1536-resolution model ranks first among the evaluated open-source models on PRISM-Bench, Qwen-Image-Bench, and UnicBench. MeanFlow-based few-step distillation yields WeLM-Image-Turbo, which generates a $1024\times1024$ image in 4.98\,s on a single NVIDIA H20 GPU using eight guidance-free steps.