2026-10-23

Hackers were already probing the bug before defenders were told it existed.

Perplexity and NVIDIA introduced JOLT, a post-training method that trains one model as both teacher and student. On MATH it matched GRPO’s accuracy with 6.5x fewer processed training tokens. The post Perplexity and NVIDIA Researchers Introduce JOLT: a Joint Teacher-Student Reinforcement Learning Method appeared first on MarkTechPost .

Autonomous agents can increasingly propose ideas, modify code, and run experiments, creating an opportunity to accelerate scientific discovery using teams of collaborative AI researchers. To escape local minima in the discovery, we propose to encourage orthogonal exploration by constructing the team using agents with varying expertise. However, realizing this design requires more than diverse backgrounds. When experiments are costly, a team can test only a limited number of ideas, thereby creati

Agent harnesses have become a central component of modern agentic systems, shaping how models use tools, manage context, and coordinate long-horizon interaction. Realizing the benefits of these harnesses depends on how effectively the underlying model can make use of them, motivating reinforcement learning directly through the harness. However, existing approaches typically train through a single fixed harness, which can lead to specialization and limited transfer across harness configurations.

InclusionAI just announced Sante too. Damn, two model announcements in one day. They're on a roll. I posted about the 3.1-flash-1019 update earlier today and they've already announced Sante . Two model announcements in one day, damn. This one's built on Ling-3.0-flash , with 124B total parameters, 5.1B active per token and a 256K context window. It's a healthcare-focused version aimed at medical reasoning, literature research and tool-assisted workflows. The weights are already up on Hugging Fac

Flubbed key signing "ceremony" allowed minting of web credentials for any site.

The next competitive advantage in enterprise AI has less to do with which model you choose than with what you build around it. That’s an important shift. For the past several years, much of the AI conversation has centered on model capability: bigger models, better reasoning, larger context windows, higher benchmark scores. Those advances have…

Fast, focused, low-cost decisions for any application.

Cohere Rerank 5 ships in Pro and Fast tiers with a 32K context window, Context Focus passage extraction and calibrated relevance scores. Pro scores 82.2 on BRIGHT and costs $0.05 per 1M tokens. The post Cohere Releases Rerank 5 With Passage-Level Context Focus and Calibrated Scores appeared first on MarkTechPost .

Retrieval Augmented Generation (RAG) grounds large language models in external knowledge. Modern pipelines combine embedding, chunking, retrieval, reranking, query augmentation, context compression, and agentic verification, yet these techniques are typically evaluated one at a time. It remains unclear whether their isolated gains survive composition, whether retrieval quality and answer faithfulness favor the same design choices, and whether conclusions from small corpora hold at large scale. W

Ask most executives to picture the cost of AI and they’ll likely see GPUs—racks of accelerators with gigawatts of power—and a waitlist for silicon. But that picture leaves something out: Data storage costs. What happens to data after a model responds? Much of the back-end data that’s generated—from billions of daily prompts, images, and videos—needs…

I have debunked (some of the) benchmarks of this new and suspicious "Tov" programming language. Slight background: My friend (who is pretty interested in new stuff generally) found this thing and sent its main site to me saying things like "looks pretty cool", but my intuition told me this sounds like "a scam", esp. when it claimed it performs faster than C and Rust. I personally think apart from changing the code, they probably tried to set up some specific environments that make their programs

IBM released Granite Speech 5.0 2B, an Apache 2.0 speech recognition and translation model. It ranks #2 on the Open ASR Leaderboard with 5.68% average English WER, alongside updated Turbo CTC and 4.1 2B Plus models. The post IBM AI Releases Granite Speech 5.0 2B for Multilingual Transcription and Translation appeared first on MarkTechPost .

arXiv:2610.28686v1 Announce Type: new Abstract: Query clarification is essential for retrieval-augmented generation (RAG) systems that face ambiguous user queries. We consider multi-turn grounded conversational scenarios where a user query is either ill-formed or ambiguous in the context of the prior dialog or, more interestingly, it may be well-formed but when posed against retrieved documents there are multiple relevant interpretations. The task is to recognize both such cases and then generat

Autonomous agents powered by foundation models can reason, plan, and use tools, but their fixed weights and context limits hinder sustained learning from experience over long-term deployment. While external memory systems attempt to bridge this gap, they mostly act as passive retrieval caches, evaluated primarily by whether they can recall stored facts or past interactions. Whether memory can drive genuine continual learning remains an open question. In this paper, we introduce Memory-Centric Co

arXiv:2610.28934v1 Announce Type: new Abstract: Coding agents now complete tasks autonomously, modifying real-world state such as artifacts, software environments, and configurations along the way. Yet they show limited backup awareness: recognizing that an action will modify state that may later be needed, and preserving it beforehand. To evaluate backup awareness, we introduce RegretBench, a benchmark of 205 executable tasks grounded in regret moments reported on platforms such as X.com, GitHu

arXiv:2610.28815v1 Announce Type: new Abstract: Agentic harness evolution optimizes an agent's instructions, tools, and control flow by using its task execution experiences as evolving evidence. While more evolution tasks should intuitively improve performance, our preliminary study reveals a paradoxical failure mode: the performance of existing methods degrades as the task pool expands. Our further analysis demonstrates that this degradation stems from suboptimal evidence selection during optim

arXiv:2610.28763v1 Announce Type: new Abstract: Frontier coding agents are increasingly framed as end-to-end automated researchers, and ML for science offers a demanding test of that promise: improving a scientific model often requires domain insight beyond routine ML engineering. We introduce Agent4Sci, a long-horizon testbed for measuring how far today's agents can improve the published best models on scientific ML benchmarks, spanning 20 benchmarks across 11 domains. Rather than ranking agent

Discover NVIDIA Nsight Python, an open-source interface that allows developers to profile GPU kernels, analyze hardware metrics, evaluate cache and clock controls, and measure kernel fusion directly from Python code. The post NVIDIA Nsight Python Tutorial: How to Profile GPU Kernels appeared first on MarkTechPost .

On-policy distillation (OPD) can succeed with surprisingly few training questions, motivating a view of questions as carriers of distributional knowledge transfer. How can we assess whether a newly generated question is valuable for distillation, and learn to generate better questions---without access to its correct answer? We introduce Ask-to-Distill (\method), a framework that treats synthetic questions as distribution elicitors and optimizes the teacher supervision they expose along student-g

On October 22, Chinese media reported that DeepSeek and MiniMax have officially completed the required registration filings under China’s new temporary AI export regulations. By submitting their technologies for cross-border review, both companies successfully fulfilled the administrative requirements established by China’s Ministry of Commerce for distributing advanced AI models overseas. For DeepSeek and MiniMax, this clearance grants explicit regulatory approval to continue open-sourcing mode

I am surprise no one is talking about one of the biggest vulnerability discovered in the last 24 hours and the havoc it created. So this is not about AI. But AI (Wiz and OpenAI) found these CVEs. 70% of world's site runs on PHP. I was getting 20 calls and my phone blowing up from people I have not spoken to in 10+ years. To patch. I am not a PHP developer (I left in 2012) but know enough.. This highlights a problem with people simply NOT having a regular maintenance cadence. I had people with PH

I just downloaded Raon-1-mini 30B A4B as well as Flint 40B A2.5B and it just blows my mind that we have so many options now from so many countries. It feels like the countries are all working together and trying to contribute useful software and it’s just amazing. I can fucking TALK to my computer and have it answer like fucking Star Trek. That’s insane. Every day I use this tech I’m just more and more blown away.

Hi Guys!, Recently I was making a research and development, but I do not have time to go further with this idea (kids, work, etc.). But maybe it will be interesting for someone. I do not think it would be something big, but maybe interesting approach that can be used in some of the cases. I believe that it is not AI slop, you can use those information's and try by yourself using Claude or Astra. The idea was my own and I am using it in my production app, it saved me a lot of money (project is no

Made this from the Artificial Analysis model pages after Muse Spark Open and Beam got their scores last week. X axis is total parameters as AA lists them, log scale. Orange = released in October 2026. - Muse Spark Open (816B, 42B active) scores 39. Qwen3.8-Flash-Next (180B, 6B active) scores 40. - The Unsloth UD-Q4_K_XL GGUFs are about 506 GB for Muse Spark Open and 111 GB for Qwen3.8-Flash-Next. - Meta's recommended local setup for Muse Spark Open is two M5 Ultra Mac Studios with 512 GB each. -

OpenAI is launching its decision model to the world. Can it compete with the original Jev?

Microsoft’s new AI diagnostic system beat generalist doctors on hard cases, according to a study in Science. But most of the improvement seems to come from the AI models underneath, not from Microsoft’s system itself.

Agent Skills improve LLM agents by making task knowledge reusable across sessions. Learning transferable Skills requires evidence about a procedure and its conditions of use, which a successful task attempt can leave untested. We introduce Socratic Skill Learning (SSL), which interleaves expert teaching with Student practice. An Educator guides investigation while the Student remains the sole environment operator and Skill author, able to question, test, and adapt advice. To test reuse beyond th

Enterprise AI can generate recommendations, automate tasks, and help organizations make decisions faster. But without an understanding of how a business actually operates, even powerful models can miss the constraints and relationships that determine whether a decision is the right one. As companies look to move AI from experimentation into everyday operations, the challenge ahead…

Agentic robotics systems combine planners with robot harnesses to perform complex manipulation tasks. Recent work improves these systems through better harnesses and external memory, but relies on powerful closed-source models as planners and does not improve the planners' own capabilities. In contrast, coding agents have improved substantially by post-training the model itself. We therefore investigate whether post-training can also improve robot planners. However, applying it across robot harn

Quantitative research is an iterative loop: a researcher proposes a market hypothesis, implements it as an executable signal, tests it on historical data, and decides what to keep. Existing benchmarks score formulas, code, or individual trading decisions, leaving open whether large language models (LLMs) can complete this loop and produce strategies that work beyond the research period. We introduce Quant-Arena, a benchmark and public leaderboard for end-to-end quantitative research on real A-sh

中电信人工智能科技有限公司星辰通用人工智能实验室正式发布并开源意图理解专用模型 Xing4.0-29B-Intent

arXiv:2610.28905v1 Announce Type: new Abstract: Diffusion language models form answers through a sequence of revisable continuous states, but multi-agent systems usually discard those states and make agents communicate only after decoding text. We ask whether one independent diffusion trajectory can pass a compact hidden message into another and, crucially, whether the receiver uses what that message means. In Hidden Messages, a sender compresses its denoising state into four signed dense vector

arXiv:2610.28706v1 Announce Type: new Abstract: LLM-driven evolutionary search is a promising paradigm for scientific discovery, but existing methods remain costly and slow, often applying expensive reasoning uniformly while underusing evidence from failed, redundant, and saturated trials. We introduce Speculative Evolution (SpecEvo), a multi-LLM framework that coordinates lightweight and frontier models through continual consolidation of search evidence. Lightweight models power both parallel S

Using an LLM to generate training data for another is among the most general ways to train one, and it remains common practice. Some frontier teams have closed that route for their own models: what comes back is an answer, or a short summary, without the detailed reasoning trace behind it. That is a little ironic: they were themselves built on distilling human knowledge, and distilling an open-source model is still entirely routine; the restriction applies to their own outputs alone. Recent work

Linear-attention language models keep their context in a recurrent state of fixed size, where an inference system stores one copy of this state for every sequence it decodes at the same time. Storing the state at lower precision therefore lets more sequences share an accelerator. Existing state quantizers round the state after every generated token and report that four bits lose accuracy unless the bit width is raised or the model is fine-tuned.\par We introduce PACE, a post-hoc four-bit state q

Language-model agents that run long tasks often redo work they have already done: they re-issue a tool call they made earlier and get back exactly the same observation, a step that adds no new information. We identify this failure as a loop. Loops appear natively in Qwen, Llama, and Gemini agents and become far more common once the agent's history is compressed to fit its context window. To reduce these loops, we propose DeLoop, a self-supervised training method. DeLoop builds on a simple observ

arXiv:2610.28914v1 Announce Type: new Abstract: Frontier language-model agents excel at complex tasks, but high inference costs make transferring their capabilities to smaller models essential. Traditional knowledge distillation achieves this through parameter updates, requiring access to the student's weights and additional training compute while encoding the transferred knowledge implicitly in the parameters. We introduce D-Skill, a framework that treats the teacher as a black box and distills

arXiv:2610.28909v1 Announce Type: new Abstract: Deep research systems iteratively plan, gather evidence, synthesize, and revise, making end-to-end latency a key bottleneck for interactive use. Almost all production systems are built on autoregressive (AR) language models and inherit left-to-right decoding in planning and long-form writing. We present DiffResearch, an open multi-agent deep research framework that places a diffusion language model (dLLM) at the core of every stage, and measure how

Full 3D attention is costly in video Diffusion Transformers. Block-sparse attention reduces this cost by computing only selected attention blocks. However, existing methods either overlook the potential to optimize sparsity allocation or rely on heuristic rules or local metrics to guide it. In this work, we observe that sparsifying attention in different layers and at different denoising steps affects final-video fidelity to substantially different degrees. Yet the identities of the most sensiti

An open-source personal agent that gets you things done Discussion | Link

Hello everyone! Pangolin 1.25 is here with two highly requested features from this community: HTTP redirects and mutual TLS (mTLS) for public resources. It also includes roles for machine clients and more! Pangolin is an open-source, identity-aware remote access platform that simply and securely connects and authenticates your users to applications, infrastructure, and AI workloads. GitHub: Redirects Redirects can be attached to a public resource and share its hostname, so you can send traffic f

Scaling data, diversity and model capacity has improved embodied learning in physical environments, where diverse experiences remain grounded in shared physical laws. Whether these gains extend to virtual worlds with different visual styles, dynamics, controls, and objectives remains unclear. We investigate this question through a controlled study of continued pre-training (CPT) across three axes: data volume, diversity, and model size. Our 120 training configurations span four model sizes from

Article URL: Comments URL: Points: 68 # Comments: 38

Canva is diving even deeper into AI with the launch of an image generation model that competes directly with those from OpenAI, Microsoft, and Google. The model, called Canva Craft, uses a two-stage architecture that allows it to plan a design before generating an image, allowing for "greater precision, consistency and control," according to an […]

Meta has re-enabled Muse sharing after security researchers had reported that public conversation links could be easy to find. The researchers said they had been able to find Muse pages like a marketing company's internal revenue review and a small business owner's loan application. Researchers at Claroc reported earlier this week that Muse the web […]

Canva is making its new image model available to all users, promising better, more customizable performance at lower costs.

Explore how to audit and rebuild Datalab's OmniParseBench, test parser ranking certainty with document-level bootstrapping, and evaluate Tesseract and Indra OCR performance on a custom sample. The post A Developer’s Guide to OmniParseBench: Leaderboard Benchmarks and OCR Testing appeared first on MarkTechPost .

As fun as AI is, we built so much bad software over the last 30 years, and now the universe is punishing us for our sins. — Armin Ronacher Tags: security , generative-ai , armin-ronacher , ai , ai-security-research , llms

How Moshi turns market data into actionable insights, and why AI tools have to fit the trader’s workflow.

I've worked for a long time in the storage field during my career and ZFS is the piece of engineering I admire most in that space. I wanted a way to pay homage to it and to explore and better understand its internals. So I built Uberblok: a museum of ZFS. A deep dive into copy-on-write, txgs, vdevs, RAID-Z, snapshots, the ARC, metaslabs, the ZIL, send/receive with explanations and animated diagrams. I'm sure people here know these internals better than I do, so corrections are very welcome. I bu

Multimodal agents (MMAs) sustain interaction through reasoning, memory, tools, and feedback, most visibly as computer-use agents, while robotic systems (RSs) couple sensing and actuation under physical dynamics. We define multimodal embodied agents (MMEAs) as goal-directed systems that couple multimodal task reasoning with physical action and revise decisions from the resulting feedback. This raises a central question: what changes when a multimodal agent moves from computer-use to robot-use? We

arXiv:2610.28891v1 Announce Type: new Abstract: Many social evaluations for multiagent systems study behaviors as they emerge from environments instead of systematically intervening in the environment to elicit such behaviors. We introduce Fig (Finding Interactions in Games), an agentic system that performs targeted elicitation interventions on game environments. It takes a natural language behavior description, decomposes it into sub-behaviors, generates new social games or selects existing one

arXiv:2610.28855v1 Announce Type: new Abstract: Progress on foundation model pretraining is gated by who can afford a trial, and relational pretraining is no exception. Recipes are developed inside individual labs, reported under private protocols, and do not accumulate. Speedrun leaderboards remove that barrier by decentralizing the search rather than the training run. The problem is fixed, the stack is open, entries are ranked by wall-clock time to a quality target, and one GPU is enough to ho

Explore a comprehensive coding guide benchmarking Cactus Compute's 16.9 MB speech model against Whisper base. Learn about word error rates, CPU speed, streaming latency, and speech-to-tool integration. The post A Coding Guide to Cactus Whistle: Does a 16.9 MB Speech Model Hold Up Against Whisper Base? Word Error Rate, Speed, Streaming, Timestamps, Keywords, and Speech-to-Tool Calls on a CPU appeared first on MarkTechPost .

This classifier can process data from the public DHT to filter, map, & match torrents extremely accurately while also filtering out malware, junk, and fake torrents. you can now obtain the same level of quality torrent feed data as a public site with simply a regular dht crawler (like bitmagnet) I’ve been experimenting with how far you can push a tiny byte-level RoPE Transformer (8.3M params, 8.3 MB INT8 ONNX on CPU / I've been using this with BitMagnet & Rats Search -> sonarr / radarr stack. I’

When a speaker says ``some students passed'', listeners infer that not all of them did. Reasoning models now let users decide how long a model thinks before it answers, but we do not know what extra thinking does to such everyday inferences. We study this directly: we keep a model's weights fixed, change only how much it thinks, and record which reading it chooses. A second tool cuts the model's own reasoning at successive points and asks for an answer, showing where the reading changes. Thinkin

Title is a bit of an exaggeration but the question is more: if the frontier models are so good, why even sell them as your primary money maker? Couldn't OpenAI or Anthropic make a lot more money by just starting companies themselves that leverage the models and building their own products? Is selling tokens really the best strategy? Especially with commodification and prices declining wouldn't that indicate there is more money to be made on the application-side of the models rather than the util

Follow up to my post about the £300 Snapdragon mini PC: Quick recap of how I got here: I started on llama.cpp, then pretty quickly moved over to Janas (github.com/prabanta-dev/janas). I ported it to ARM and contributed the work back as PRs (NEON kernels, a Windows on ARM build, and a cache bug fix Maurizio has since shipped in his own release). Working inside Janas gave me a load of ideas, so I'm now writing my own engine, codename **ToRc**. It's not released, YET, but it's what I used to produc