2026-10-21

A reliable source has indicated an announcement is scheduled for 22 October 2026 at 18:00 CEST regarding 2x PHP RCEs with a CVSS score of 10.0. I have no evidence beyond "source: trust me, bro", but if you're running PHP in production, be ready to update as a priority. These CVEs reportedly relate to remote code execution in all affected PHP versions from PHP 5 onwards. These are the two reserved CVEs for the reported upcoming announcement: CVE-2026-109618 CVE-2026-109619

Anthropic is reportedly struggling to secure enough compute from cloud providers that increasingly view it as a competitor, even as its infrastructure spending balloons ahead of a planned November IPO.

Google has joined Amazon in releasing a developer tool that makes it easy for PyTorch users to move their apps off Nvidia GPUs.

From GPU trading to physical settlement, independent markets, and programmable compute

Cloudflare has previewed Clairic, a System One extraction model on Workers AI that returns typed values with confidence scores in one forward pass. It costs $0.10 per million input tokens and charges nothing for output. The post Cloudflare Releases Clairic, a System One Extraction Model on Workers AI appeared first on MarkTechPost .

arXiv:2610.23866v1 Announce Type: new Abstract: We introduce GardenBench, a benchmark with 11 test suites that evaluate the capabilities of Large Language Models (LLMs) across a range of tasks. By systematically prompting LLMs with various lightweight queries, we document day-to-day changes in model performance, including run costs and times, that other benchmarks miss. This allows us to compare the accuracy and features of responses across model providers and versions. We account for daily vari

arXiv:2610.24161v1 Announce Type: new Abstract: TorchTPU is an open-source collaboration between Google and Meta that introduces a native and principled PyTorch backend for Google TPUs. It enables seamless out-of-the-box model execution and TPU migration while delivering industry-competitive inference and training performance through eager execution optimizations, torch.compile integration, and native hardware acceleration.

arXiv:2610.24194v1 Announce Type: new Abstract: Tool-using LLM agents repeatedly face decision points under uncertainty, choosing between autonomous action, additional information gathering, and human escalation. These decisions require balancing task success against interaction cost and the risk of irreversible errors. Learning to make these trade-offs is challenging because existing interaction trajectories record only the outcomes of actions taken, leaving the consequences of alternative acti

arXiv:2610.23921v1 Announce Type: new Abstract: There is growing interest in automating business activities with Agentic Artificial Intelligence (AI) due to latter's seeming ease of use. Never has it been easier, or costlier, to do less with more. However, little is known about when agents are preferable to established alternatives such as local computation, Representational State Transfer (REST), the Simple Object Access Protocol (SOAP), and the Model Context Protocol (MCP), particularly when d

OpenAI只掏出一个6.1 Terra…

from Ornith: Ornith-1.5-122B Chirp Chirp! 🐦 We are introducing Ornith-1.5, a major step toward building foundation models through end-to-end self-improvement. Ornith-1.5 extends Ornith-1.0 (which was developed on top of Qwen3.5 and Gemma4 with additional continued pretraining, mid-training, and post-training) by expanding the self-improvement loop from scaffold and rollout optimization to jointly optimizing task generation, scaffold construction, and solution rollouts. Rather than relying on a f

This demo from ISTA-DASLab recovers low quant accuracy while retaining low VRAM usage and fast decoding. It needs a Nvidia Blackwell GPU to run, so I have not tested it. For context longer than ~4k there should be no speed penalty compare to running without the prefiller model. Graph shows MMLU-Pro accuracy. Huggingface🤗:

While LLM-based agent systems have demonstrated remarkable proficiency in reasoning and solving complex real-world tasks, they also come with substantial token consumption and inference latency, limiting their practical deployment. To enhance the efficiency of agent systems, recent efforts have focused on transferring agentic capabilities from large-scale models to small language models (SLMs) via supervised fine-tuning-based distillation. Nevertheless, as interaction trajectories grow longer, i

Eight SSL Intermediate Certificates of eMudhra have been publicly leaked. All your servers https traffic can be broken right now. Browsers are fine as they have revoked the certificates but as you can see, curl and many other libraries still accept those. Proof of concept and check if you are affected: ``` curl --resolve 'q3k.google.com:443:135.181.235.221' q3k.google.com body { font: courier, sans-serif; font-size: 12px; text-align: center; margin-top: 100px; } oh, this is bad ``` No download c

A few of you pointed out in my last post that comparing Flash-Next and Opus with different harnesses and effort levels wasn't fair. So here is a more controlled comparison. Same task for every run: a real feature from the LlamaStash roadmap (save llama-server's KV cache before an unload and restore it when the model comes back). Six combinations, each run twice, so 12 PRs. Flash-Next ran locally on my Strix Halo laptop (Flow Z13) via Halogen (~Q4 quant). Opus 5.5, Flash-Next, and Kimi K3 scored

Meta is suspending the sharing feature while it investigates its privacy.

The proposed policy seems to offer protections against AI run amok but actually forbids federal agencies from blocking a model's release without the permission of a judge.

OpenBSD 8.0 lands with Intel Arc support, ARM64 hibernation, and 64-vCPU guests

My new book is out today... what happened since the last one

The goal of a world model is to capture the dynamics of an environment. This has the great appeal that it allows agents to plan and simulate a course of actions with the help of a world model. For these uses to be reliable, however, the learned dynamics must behave well beyond the specific state-action combinations observed in expert rollouts. Despite this, the relationship between data coverage, dynamics generalization, and downstream planning performance remains underexplored. We study this re

System service that watches your card, swaps and loads whatever you want, whenever you want. Don’t have to swap anything manually, collisions avoided, VRAM fully dedicated to each function while it’s running (no VRAM squatters). I had issues everyday (Linux) where a node process would orphan, comfyUI auto starts and holds 500MB that strata can’t claim. I also wanted to be able to switch natively and not have to think about starting each engine, making sure strata or llama wasn’t stuck loaded, or

Exa 3.0 is a search API upgrade from Exa. It adds a larger embedding model, Exa Lite at $4 per 1,000 requests, and Dynamic Highlights, and leads 8 benchmarks Exa ran itself. The post Exa 3.0 Search API Tops Perplexity, Brave and Parallel on 8 Benchmarks appeared first on MarkTechPost .

Congrats to DeepSeek V4.1 Flash 1017, MiniMax M3.1 Flash, and Mercury 3 (details below) but they’re not going to be in super wide use just yet.

arXiv:2610.24061v1 Announce Type: new Abstract: As teams of AI agents take on larger projects and long-standing open problems, agent collaboration offers another dimension for scaling inference-time compute. To understand what makes such collaboration effective and scalable, we develop Software World, a framework for simulating multi-agent coworking in realistic software ecosystems. It comes with a pipeline to create environments of real, interdependent Python packages of different topics, sizes

arXiv:2610.23962v1 Announce Type: new Abstract: We propose a controlled Large Language Model (LLM) workflow for explaining risk score differences using similar historical transactions. The workflow combines reference retrieval and contrastive SHapley Additive exPlanations (SHAP) evidence with reference-based LLM reasoning and deterministic validation. Experiments on a test dataset show that the workflow produces more consistent, evidence-grounded and traceable diagnoses than the same LLM without

arXiv:2610.24107v1 Announce Type: new Abstract: Context-to-LoRA methods encode documents as low-rank parameter updates, allowing language models to answer without retaining the source text in their decoding context. How reliably do these updates preserve facts and support their use under different instructions? Controlled diagnosis of existing context-to-LoRA methods separates fact discrimination, target retention, and response adaptation. Competing sources influence answers, while target-only c

Long-horizon, semi-verifiable agentic tasks often rely on rubric-based evaluation, but the evidence needed to assess these tasks can be spread across the environment state and the execution trace. A judge may overlook this evidence or accept unsupported claims, including an agent's own reports of success. When rubric verdicts serve as reinforcement-learning rewards, such errors can reinforce undesirable behavior. We introduce Sauron, an evidence-first agentic verifier that investigates both stat

I've recently discovered that you can use Touch ID (rather, the Mac's Secure Enclave) for SSH authentication. I'm the author of the post, feel free to ask any questions and if there are any experts I'd love to hear your thoughts!

Existing Multi-token prediction (MTP) methods are auxiliary by design: the multi-horizon loss mostly trains modules attached to the backbone, and each added horizon costs another module and another pass. We propose Efficient Causal Self-Recursion (ECHO), which removes both costs. The backbone's own final layer predicts every horizon, so the supervision lands on the parameters that generate the next token, and Dynamic Depth Sampling (DDS) keeps deeper horizons affordable by drawing the training d

Coding agents keep working on runs that will fail verification, and asking a frontier model to watch for this is expensive in its own right. We introduce StopLoss, a small external monitor that learns failure recognition from completed trajectories and their verifier outcomes. The name is inspired by the stop-loss order in finance: StopLoss ends a run early when the run is likely to fail, which saves resources. Final outcomes label earlier local windows, so a fine-tuned language model can score

At Gradium we just made our first open-source model release with Phonon, a collection of small Text-to-Speech models designed to run on device. The inference stack is fully open-source and targets 10+ different backends (CPUs, GPUs, NPUs), most of them in Rust. We hope that the community will build lots of different things on tops of our models and code, any feedback is very welcome and we're looking forward to many more open-source releases!

Introduction A little over a month has passed since Spark-X2.5. In the rapidly evolving world of AI agents, that is enough time to rethink, refine, and move the frontier forward. Building on the Spark-X2.5-4B, we introduce Spark-X2.6-4B , the latest generation of our compact, general-purpose language models for on-device deployment. Spark-X2.6-4B retains exactly the same architecture as Spark-X2.5-4B, achieving its capability gains through improved post-training. It delivers substantial improvem

Meta changes AI scraping policy but won’t add language respecting robots.txt.

Hysteria over the agent's appearance in a 7,500-person group reveals that internet experts have outdated notions of privacy.

My latest vibe coded game experiment is a Command & Conquer style RTS (Real Time Strategy) game called Battle Drones ( play it here ). Unlike my previous experiments , this one is actually fun to play . I'm somewhat embarrassed to admit that I got there by telling Claude to "Make it a LOT more fun". It also supports multiplayer between nearby devices using audio to synchronize game state! Here's what I've built so far, and some of the things I learned along the way. Battle Drones How I built thi

OpenAI says it's adding Orchestra's team and technology to work on Rosalind, its AI model for life sciences research, and to its related Rosalind Workbench tool, a harness that aims to help researchers plan experiments and analyze results.

I just wanted to let you know, both Immich and Paperless-ngx (two of my most favorite selfhosted apps) are now at version v3.3.3. (I know, I have too much time on my hand)

We propose KronLSTM, a linear attention architecture based on the Kronecker Delta Rule, a novel update rule for associative matrix memories. KronLSTM extends the xLSTM family with a memory-revision mechanism that corrects targeted associations while modifying the memory only along a single direction in matrix space. This minimally invasive memory revision reduces interference with previously stored information compared with delta-rule updates, making it particularly well suited for long-context

A judge ordered Abdulrahman Alfozan to remain in custody, likening the Saudi government’s offer to secure the former OpenAI engineer’s release to a “blank check.”

arXiv:2610.24177v1 Announce Type: new Abstract: Video generative models are an emerging tool for embodied agents: they can serve as scalable visual planners, generating videos that depict how to complete previously unseen tasks. However, executing these plans typically requires a video-to-action or "inverse dynamics" model trained on large, human-collected video-action datasets, severely limiting scalability. We introduce Mirror, a self-supervised approach that learns video-to-action mappings th

Hi HN, I'm Peter Bishop and I've been working in graphics and web tech for over 25yrs. I've recently launched Death Plan Switch to solve a major problem in my succession plan. What if I go missing or die? The right people are told and my business can continue to operate for my clients. There's a backstory to this project, I was hospitalised in 2012 and lost my business back then and I couldn't get the data to my technical 2IC in time without frustrating my clients. Death Plan Switch solves this

arXiv:2610.24133v1 Announce Type: new Abstract: We ask how much an autoregressive transformer retains of its prompt after generating $t$ tokens. Consider two different prompts followed by the same continuation text, so that any difference in their hidden states comes from the prompt alone, and we find that the distance between their hidden states decays approximately as a power law $t^{-\alpha}$, where $\alpha=1$ is the rate at which a fixed prompt's share of uniform attention shrinks and smalle

arXiv:2610.24090v1 Announce Type: new Abstract: Does persistent memory accumulated from previous attempts help an LLM agent solve new problems, and does memory design affect that benefit? We study programmatic Chinese-character drawing across five agent conditions attempting the same 668 items. Accumulated memory produces modest aggregate gains despite substantial growth in the memory stores: free-form notes improve first-attempt success by 4.6 percentage points, while a design combining free-fo

OpenAI is acquiring Orchestra Bio to bring AI-powered operations to life sciences R&D, helping biopharma teams accelerate drug development and advance new treatments.

Existing world models can generate action-conditioned video from a single image. Many focus on camera movement, allowing users to explore the generated scene. However, navigation alone does not ensure that an action follows persistent rules when its outcome depends on state that is invisible in the current frame. We present Stateful Harness, a framework that separates world evolution from video rendering to support stateful and rule-governed interaction. It maintains an evolving world state unde

Open models, built for agents. Discussion | Link

Safety in Multimodal Large Language Models (MLLMs) depends on unsafe multimodal data for post-training alignment or activation steering. However, such data is hard to collect, must be gathered anew for each modality, and its collection and distribution raise ethical and legal concerns. In this work, we relax this constraint and investigate whether textual refusal directions, extracted directly from the LLM backbone, generalize across modalities (i.e., image, video). Preliminary findings confirm

Lifelong agents continually accumulate and reuse experience to improve after deployment, yet in practice they often operate under tight generation, interaction, and memory budgets. Under such stringent joint constraints, we find that existing lifelong-memory methods can become ineffective, with several sophisticated approaches offering little advantage over simple retrieval or even no-memory agents. This reveals a more fundamental problem: resource scarcity makes each memory decision more conseq

Smarter search that gives AI agent richer data at lower cost Discussion | Link

当AI越来越懂你，如何让智能真正属于你？

A guardrail classifier is reported as if unsafe were a property of text. It is not: it is a property of a policy, and the policy enters the model through its labels. We exploit an accident of an industrial deployment in which the same ~53k prompts were labelled twice, under two operational definitions ("is this a request for harmful content?" and "would a moderator block this?"), and used to train ten guardrail arms sharing architecture, codebase and optimiser family, at two model scales. First,

Harness the agent. Then harness its self-improvement. Discussion | Link

Hey, everyone! I'm so happy to finally share this with the community. I started VoiceManager way back in March to support a game concept and my AI assistants. I had the idea for this Trek inspired app to build pieces for a future project. Low latency streaming speech to speech running locally, streaming over one WebSocket. The default configuration uses a special audio.cpp sidecar with Nemotron and Breeze TTS 2. VM supports turn based conversations with barge-in. Barge-in is off by default becau