2026-10-22

Startup Recursive has released Flint, a free, laptop-sized coding model built largely by its own AI research system, as an early test of AI that improves itself.

An open source “bomb” drops from China

Mistral AI released Voxtral 3, a family of 3 transcription models, 1 text-to-speech model and a LiveKit-based voice agent SDK. Voxtral Small Transcribe 3 scores 2.1% AA-WER, ranking #3 on Artificial Analysis. The post Mistral AI Releases Voxtral 3: Speech-to-Text, TTS and a Voice Agent SDK appeared first on MarkTechPost .

Google, Red Hat and IBM released llm-d 1.0, an open-source Kubernetes-native stack for distributed LLM inference. It adds multi-tenant flow control, zero-downtime upgrades and agent-aware routing across vLLM, SGLang, TPUs and GPUs. The post Google Cloud Ships llm-d 1.0 With Cache-Aware Routing for TPU and GPU Fleets appeared first on MarkTechPost .

Long-horizon tasks require LLM agents to continually draw on information from earlier interactions. However, retaining the full history increases context costs, while compressing it risks losing details needed later, and the relevance of historical information often becomes apparent as the task progresses. To address these challenges, we propose FlowState, which treats execution state as memory that can be retained and revisited across requests, unifying current decision-making with the reuse of

I came across OpenBMB’s engineering write-up on running large MoE models locally, and the systems work behind it is worth highlighting. Instead of shrinking the model or skipping experts, they tackled the real bottleneck: getting expert weights from flash storage to computation fast enough. OpenBMB’s MiniCPM-V 4.7 uses a technique called Lookahead Expert Scheduling (LES) to run a 35B-A3B MoE model locally without changing its actual routing or dropping experts. The key idea: predict which expert

If you run the LiteLLM proxy in Docker and let Watchtower or anything similar auto-update it, check it now. The 1.107.0 release on PyPI and the official image litellm/litellm:1.107.0 carried an infostealer plus a backdoor that polls the attacker's server for new code every hour. The latest , main-latest and main-stable tags pointed at it, so auto-updaters pulled it in. The image has since been removed from Docker Hub and GHCR. Original analysis from Aikido: What it took: every provider API key s

Two bugs found with AI are behind major issues today for 4chan, DeviantArt, Wikipedia, and the blog where the bar around the corner posts pictures of people's dogs. The "critical vulnerabilities" were discovered in the PHP programming language, which underpins much of the internet, including any site built on WordPress. But it's far bigger than […]

All websites using PHP should update to the latest security patch.

Unconfirmed report claims PHP2Shell exploit remotely executes code on PHP sites.

The company is suspending the feature after security researchers discovered how easy it was to access sensitive information shared with the AI assistant.

Claude Mods is an early form of the Attention Interface. It flips the agent harness paradigm from the harness facing the machine, to the harness facing the human.

arXiv:2610.26887v1 Announce Type: new Abstract: Large language model (LLM) agents are evolving from conversational assistants into Cowork Agents capable of autonomous planning, tool use, and complex task execution. Financial analysis requires these agents to integrate cross-document evidence and perform multistep numerical reasoning, but expert task authoring is costly, while direct LLM synthesis struggles to produce challenging tasks with reliable answers. We introduce Navalia, a framework for

Mixture-of-experts (MoE) models often exceed a single accelerator's memory capacity. Offloading experts to host memory creates a severe PCIe I/O bottleneck because data-dependent routing keeps transfers on the critical path. We present Specter, an inference system that uses quantization to combine speculative decoding with expert offloading. Its lightweight quantized draft model runs efficiently on device while preserving high-fidelity expert activation patterns. This provides accurate multi-ste

arXiv:2610.27032v1 Announce Type: new Abstract: Large language model (LLM)-based agents have recently demonstrated outstanding performance in single-turn tasks, yet they continue to struggle with long-horizon tasks, particularly in environments without prior training. A key limitation of single LLMs is their difficulty in processing long contexts, which stems from weaknesses in continuous decision-making and reasoning capabilities. Although existing multi-agent systems (MAS) equipped with long-t

Computer-use agents are limited by the interfaces humans give them. An agent may type a spreadsheet cell by cell when a single line of code would do, or print all two thousand lines of a file through a shell to find one function. Both the interfaces and the knowledge of when to use each are built by hand, which cannot keep pace with new applications, workflows, and models. We argue that interface design should itself be an agent capability. We introduce Agent S4, an autonomous computer-use scien

arXiv:2610.26995v1 Announce Type: new Abstract: LLM-based agents increasingly rely on reusable skills. However, scaling skill libraries makes efficient and accurate skill use difficult. Full skill loading incurs high token and runtime overhead. In contrast, flat semantic retrieval treats skills as independent text entries, ignoring dependencies, execution order, state transitions, and data flow constraints. We propose SPEM, a lightweight structure-aware procedural memory framework for efficient

We present Épi, an autonomous research agent that sustains investigations across model sessions and evolves its methods through accumulated experience. Épi combines persistent research history, hierarchical memory, branching and independent review to maintain research continuity, learn from experiments, and explore and revisit alternative approaches. Its experiments run across session boundaries, and a fixed verifier evaluates their outputs. Épi produced 120 record-setting results, comprising 20

OpenAI outlines a blueprint for empowering cyber defenders with frontier AI, stronger safeguards, modernized cybersecurity, and more resilient public institutions.

Recent works show that LLMs can generate kernels with strong performance on server GPUs. Edge deployment presents distinct challenges because the hardware is heterogeneous, the software stacks are fragmented, and the resource budgets are tight. LLM kernel generation across diverse edge backends has yet to be studied systematically. We introduce EdgeKernelBench, which combines a unified IR deployment framework with a hierarchical benchmark for edge workloads. The unified IR captures reusable comp

Agent scaffolds increasingly delegate small decisions (whether to stop, verify, or change approach) to cheaper models. We ask whether a typed decision model, Jev 1.13, consulted only at deterministic decision points, makes a Claude Sonnet 5.5 terminal agent cheaper without making it worse. In a pre-specified study on Terminal-Bench 2.1, we hold one agent loop fixed and vary only the controller: none, deterministic rules, a small LLM (Claude Haiku 5.5) answering the same questions, Jev, and a pla

Hi HN! We’re Srikar and Kevan, founders of Alter – . We’re building access control for agents that can take actions on behalf of multiple users. If you’ve built an internal agent people use through Slack, you may have experienced this scenario: everyone signs in with their own account, but the agent connects to the CRM or billing system with one shared API key, so the agent can do anything that the key can do. User-level authentication in Slack does nothing to control what the agent can do in an

Coding agents can answer analytical questions over a data lake by discovering, cleaning, and joining the relevant data. However, they discard this work after each question and redo it from scratch for the next one. Tool evolution, which automatically builds and refines reusable tools across tasks, is a natural fit for this repetition. Since a ground-truth label for a data-lake question requires an expert to solve it first, we consider the unsupervised setting. We propose TEAL (Tool Evolution for

I saw this today on X and decided to try it, I havent seen it posted here yet. I ran it on a single 3090 benchmarked it with qwen 3.8 27B before and after the optimization and saw these results: Observed prompt-processing throughput improved 22–30% ; generation speed improved 1–5% . All tests passed without GPU errors. Edit: Keep in mind performance gains can be higher, my 3090 is power limited to 230 watts including during my tests

Raon-1 mini is a 30B-A4B Mixture-of-Experts (MoE) language model, designed and trained from scratch by KRAFTON AI . It demonstrates strong capabilities in Korean conversation and agentic tasks . As part of the Raon open model family, it is released under the Apache 2.0 license . Highlights Reasoning : Built-in thinking, with leading results on the math and science benchmarks shown below. Instruction following and search : Leads the comparison models on the instruction-following and search benchm

I bought a used PC with an RTX 3090 24 GB, 128 GB DDR4 and an Intel i9-9900X last year for under €2,000. With the release of DeepSeek-V4.1-Flash-1017 I decided to try running it on my machine. It is of course the smallest quant available (1-bit), but I was still happy to see it working at 7.2 tok/s . The answer has a physics error in it (wondering if someone can spot it), and I don't see much practical use for it. Qwen models actually work pretty well on my setup. Still, it felt really nice to s

Recursive makes the case that the benefits of using AI to build AI outweigh the risks.

At Nvidia’s GTC in Berlin, Huang unveiled the company’s new plug-in data center—the DSX Pod—as the key to unlocking Europe’s AI infrastructure lag. Oh, and autonomous humanoid robotics.

China just ordered 3mil RTX 5500 Pro for $11.4k a pop. Probably for llm inference and AI video gen. I suppose this card is now out of reach for local llm.

Training high-quality diffusion models is computationally expensive, and inference-time guidance often adds further sampling cost. In this work, we introduce Swift, a simple yet effective self-guided training method that uses a frozen weak denoiser to improve a stronger model without adding inference overhead. For each noisy input, Swift extrapolates the clean regression target away from the weak denoiser's prediction, amplifying components that remain underfit. The weak guide can be an early tr

Diffusion Transformers (DiTs) achieve high-fidelity image and video generation, but they suffer from substantial inference costs due to repeated transformer evaluations across many denoising steps. Feature caching reduces this cost by reusing features across denoising steps. Recent forecasting methods extend this idea by extrapolating features to skipped steps instead of reusing them. However, existing forecasting methods extrapolate the features they predict uniformly, even though the spatial-f

arXiv:2610.26904v1 Announce Type: new Abstract: Recent studies have proposed reinforcement learning with verifiable rewards (RLVR) methods that enable LLMs to generate high-quality RTL circuits from natural-language specifications. Existing approaches optimize circuit PPA (Power, Performance, Area) by aggregating multiple objectives into a single scalar reward, making it difficult to simultaneously generate diverse Pareto-optimal solutions across different trade-offs such as low power, low laten

Drafting an answer allows a reasoning model to reconsider each decision in the context of the others. Yet repeated diffusion denoising does not ensure that individually plausible token updates form a better answer together. We introduce Draft Refinement Language Models (DRLMs), which use a single denoiser to propose parallel revisions and evaluate a candidate draft. We train the denoiser to reconstruct every token from uniformly corrupted drafts, teaching both preservation and repair. Its predic

Listen now (44 mins) | Paying for outcomes, building with agents, and the last mile of enterprise AI adoption

arXiv:2610.27027v1 Announce Type: new Abstract: Multi-agent LLM systems improve performance on clinical tasks, and that improvement is often read as evidence of a more trustworthy system. However, the two are not the same claim. A clinician acts on the stated reasoning and not only on the conclusion, so whether collaboration reduces hallucination, and what decides when it does, has to be measured in the reasoning too. Yet most evaluations score only the final answer. We evaluate four systems (a

arXiv:2610.26964v1 Announce Type: new Abstract: Large language models (LLMs) struggle with classical planning, but benchmarks rarely distinguish structural difficulty from the difficulty of interpreting symbolic vocabulary. We examine seven STRIPS domains with BFS-optimal plan lengths. For predicting structural difficulty without solving an instance, the zero-training delete-relaxation heuristic $h_{\mathrm{add}}$ outperforms learned alternatives on held-out domains (mean Spearman $|\rho|=0.931$

arXiv:2610.26924v1 Announce Type: new Abstract: General-purpose agents are increasingly expected to perform professional analytics work, which requires sustained reasoning, iterative investigation, and reliable evidence. Training these capabilities requires diverse and challenging tasks. However, generating tasks directly from raw data often overlooks deeper questions that emerge through analysis and provides limited evidence for verification. To address these limitations, we introduce DataPraxi

I held off as long as I could, but finally gave in and set up strata. I am pretty blown away. 3.8 27b at Q6 kv8 was my driver before and that works super well, but Flash Next in strata IQ4 xs is really winning me over. Its not perfect, but it feels like models I was using professionally last year, which were/are good enough to do just about anything. You all are a bad influence.

Decision models support routing, guardrails, and output verification in LLM systems. However, item accuracy alone cannot establish whether models make correct decisions across controlled variations or assign reliable confidence to their predictions. We introduce DecisionPivot, a diagnostic benchmark comprising 48 tasks and 40,033 items across seven decision roles. Its core track contains 7,800 items from 13 generated task families, organized into contrast groups with computed answers verified by

Rubric-based reinforcement learning (RL) trains policies for open-ended generation with natural-language criteria as rewards, but these rewards can become ineffective as training progresses. We identify two failure modes. Under non-discriminative rewards, scores collapse to uniformly low or high values due to insufficient exploration or a mismatch between task difficulty and policy capability. Under spurious rewards, higher scores do not correspond to better responses. Both reflect a training co

Continuous flow language models offer a path to parallel generation with tools for control and acceleration unavailable to discrete diffusion, yet effective techniques for building them at the billion-parameter scale remain less established. We present Lafa (Large flow language model), a fast, high-performing flow language model scaled to 8B parameters in a data-efficient manner. Lafa is a blockwise flow model with causal attention on the prefix and bidirectional, asynchronous flow within each b

I see a number of websites that have been deliberately taken offline because of the PHP2Shell vulnerability (CVE-2026-109618 and CVE-2026-109619). How can I easily check whether any of the apps I self-host are affected?

Article URL: Comments URL: Points: 906 # Comments: 409

Article URL: Comments URL: Points: 72 # Comments: 18

A rush of new data centers could bring air pollution roaring back to levels not seen since the 2000s in Texas, which has become ground zero for the physical infrastructure behind generative AI. New and proposed data centers are demanding a tremendous amount of electricity. How developers and utilities meet that demand - with gas […]

Reliable multi-view forecasting requires videos from different cameras to describe the same evolving 3D scene. To this end, we introduce Video-Point Model (VPM), a world model that conditions on multi-view video histories and a language instruction to jointly forecast future videos from arbitrary viewpoints and the 3D trajectories of persistent scene points. We extend a pretrained world model with camera-conditioned multi-view processing that integrates complementary observations across cameras,

As Amazon blocks AI agents from shopping on its site, Walmart is taking the opposite approach, introducing new technology that lets AI agents make purchases on customers' behalf, with controls over spending and transaction approvals.

It’s the latest in a growing wave of bets that AI companies are making on brain data.

OpenEvidence is booming despite questions about accuracy and ads.

I built a quiz game around TypeSafe's new Jev model, which has vision capabilities. Jev looks at a picture and a question and picks an answer (or taps a place) in about a quarter of a second. I've always loved quiz games and thought this would be a fun way to see what it can do and how fast it is. I made the cards from scratch, and Jev gets exactly the picture you see. More cards will come over time. Play against Jev on your own (there's a leaderboard for each mode), or start a live match and ch

Samsung sells out 100% of all 2027 HBM capacity - HBM prices expected to jump 200% next year, DRAM prices expected to see a double-digit price increase" src=" title="Be prepared, GPU price increase incoming --> Samsung sells out 100% of all 2027 HBM capacity - HBM prices expected to jump 200% next year, DRAM prices expected to see a double-digit price increase" />

arXiv:2610.26942v1 Announce Type: new Abstract: Automated theorem provers often rely on sampling language models, but proof assistants like Lean reject each failed step with a precise error message --- information that pure sampling discards. To overcome this, we introduce a bounded, online step-level repair module. Our search samples $K$ candidate tactics at each step and repairs a failed tactic at most $R$ times. Instead of discarding failed tactics, we feed the compiler's error message to a p

arXiv:2610.26941v1 Announce Type: new Abstract: Large language models (LLMs) are enabling agents to move beyond predefined workflows toward autonomous planning and execution, creating new opportunities for smart manufacturing. However, dynamic environments, complex physical constraints, and stringent reliability requirements hinder the direct deployment of general-purpose agents in smart manufacturing. This article presents a systematic overview of LLM-driven industrial agents. First, we analyze

Omni LLMs can naturally locate key audio and visual evidence and integrate information across modalities for audiovisual reasoning. Existing audio-visual reasoning benchmarks that mainly rely on multiple-choice questions usually score models solely based on answer correctness, making it difficult to assess whether a model truly forms a reasoning chain from evidence to the correct answer. In addition, textual or unimodal shortcuts may allow models to answer correctly without analyzing the necessa

Am I missing something or did 'linuxserver.io' get hacked? All I am seeing on their main page is an otter floating down the river drinking from a 5 gallon water jug. If I am wrong, please let me know. I just thought it was weird and hadn't seen anything online about it. Edit: removed the link

I’ve seen vastly more questions than answers about how people are using their local AI so I figured I’d make a post about my experience. It seems most people are looking for frontier-level coding expertise for huge projects with their setups, and that anything less is not enough. I disagree and don’t really care about that stuff. TL;DR: I used a frontier model to build the infrastructure for my local agents to persistently do tedious but outstandingly useful work on my massive collection of book