2026-09-23

I thought we were supposed to be slowing down the frontier.

We talked to Google’s Oscar winning “Giganerd” about automating science, solving climate change, and how future generations can contribute to science in the age of superintelligent AI

Researchers say another looming threat hangs over some of America's most important critical infrastructure.

We present onPanda, an interactive tool for efficiently annotating LLM alignment data and agent trajectories. onPanda adopts token-level correction as its core interaction: while reading a model response, the annotator locates the first inappropriate token and either picks a substitute from the model's candidate tokens or types the correct text via free-form editing. The system then truncates everything after that position and continues generation from the corrected prefix, repeating this locate

Professional graphic design is a long-horizon agentic task in which structured, editable artifacts emerge from many interdependent actions, yet outcomes admit no reliable programmatic oracle. We introduce a continual adaptation framework in which a frozen frontier model operates professional design software through more than 230 tools, while an external procedural memory of natural-language skills accumulates and refines reusable design procedures from experience. The memory widens by acquiring

9月22日，在2026杭州云栖大会企业级Agent实践峰会上，基元律动联合创始人兼CTO韩凯发表演讲《从Harness到RSI飞轮》。

This post is written by a human and I'd appreciate it if you treated it as such. Thanks. So, I've been noticing a pretty clear interest in developing as good a coding and agentic tool-calling model as possible, especially at smaller sizes, sub-50 gigs. However, I'm finding that at least for my use of AI, if I really want to move away from big providers, I am going to require a model that has better world knowledge than the current offerings. Qwen 3.8 27B is a truly fantastic model for tons and t

Repository-level software engineering (SWE) comprises heterogeneous task categories, whose progress under pooled agentic reinforcement learning can be uneven: gains in some categories coincide with regressions in others, while aggregate resolution obscures these changes. Motivated by this category see-saw, we develop a category-aware expert-training and policy-integration framework. Executable task construction and SWE Labeler, an evidence-grounded multi-axis labeling system, organize the traini

Agentic memory is becoming essential for long-horizon AI agents, yet many existing systems rely on autoregressive LLMs to control how memories are organized, retrieved, and used, placing expensive generation on the critical path of memory operations. We introduce \method, a new agentic memory architecture inspired by System-One/System-Two cognition. System One captures fast, lightweight decision-making, whereas System Two performs slower, deliberative reasoning. Jev-Mem brings this division of l

Agent harnesses, the external systems that mediate model-environment interaction, can substantially improve agent performance, but their gains remain tied to the harness at deployment. Because the best harness varies across domains, instances, and models, a general-purpose agent must either settle for a suboptimal shared harness or route among an ever-growing set of specialized ones. We therefore study agent harness distillation: using a domain- or instance-optimized harness as training-time gui

Sparse on-policy distillation (OPD) allocates teacher supervision to a small subset of tokens in student-generated trajectories. However, useful teacher guidance can yield a noisy update when its gradient is estimated from a sampled next token. We study this estimation problem at a fixed prefix in information geometry and propose an information-efficiency ratio (IER) based on a signal-to-noise decomposition. IER characterizes relative gradient estimation error under an optimal scalar baseline. A

Skills can improve the performance of Large Language Model (LLM) agents by providing task-specific procedural guidance, while skill optimization further improves their effectiveness through iterative refinement. However, existing skill optimization methods typically represent skills as unstructured natural-language instructions, creating two key challenges: 1) Unstructured skills often lack explicit workflow-level guidance and contain substantial redundancy, making them difficult for LLMs to exe

Building general-purpose agents for industrial deployment requires integrating multiple capabilities, each typically acquired at a distinct stage of training. Yet there is currently no well-established recipe for Agent Continual Learning (ACL), with little understanding of the trade-offs among existing integration paradigms. To address this gap, we introduce ACLArena, a framework for comprehensively studying, analyzing, and evaluating ACL. We first build a sequential training pipeline and conduc

A 421M-parameter model just played Flappy Bird on my desktop CPU (OpenVINO int8) Running on my Intel Core i7 12th gen CPU Converted laya system one model to OpenVINO and quantized to int8

AntLing open sourced the Ming-Image-0.1-Design family: • Ming-Image-0.1-Design, 6B • Ming-Image-0.1-Design-Layer, 6B • Two open-source Agent Skills: the Ling UI Design Skill and the Image-to-Editable-PPT Skill Ming-Image-0.1-Design ranks #1 among open-weight models on Artificial Analysis’s UI/UX Design leaderboard

After turning a string of spectacular mathematical results into a reputational crisis, OpenAI is consulting human mathematicians to help it figure out a less disastrous path forward. On Monday, the company announced a new independent panel of mathematicians tasked with advising it and other AI companies on their interactions with mathematical research and the wider […]

GPT‑6 Astra allowed Parallel’s agents to research and synthesize labor-market data in half the time and at half the cost vs. prior models.

Hugging Face CEO：「太棒了」

Manipulation policies must know which objects matter and where they are, yet the pretrained backbones that current robot foundation models build on, from language in vision-language-action models (VLAs) to video generation in world-action models (WAMs), do not directly require this metric grounding, leaving it to be learned implicitly from robot demonstrations. We propose Grounded Action Models (GAMs), a new paradigm of robot foundation models built with 3D grounding. GAM can be conditioned usin

Business intelligence (BI) is a cornerstone of enterprise decision-making and is widely used by enterprise users in software such as Power BI and Tableau. In traditional BI workflows, users need to prepare data by (1) identifying relevant tables, (2) performing data transformations, and (3) building join relationships, before they can (4) answer their business questions. These steps can be complex and time-consuming, making BI challenging. Given the strong capabilities of large language models (

Natural interaction in digital and physical environments requires continuous perception and timely responses. Spoken dialogue relies on acoustic and linguistic cues, while video interaction also requires grounding the conversation in evolving visual context. We present Realtime-Venus, a proactive full-duplex interaction system with two separately trained 9B models: Realtime-Venus-Omni for audio-visual interaction and Realtime-Venus-Audio for spoken interaction. Each model serves as a complete co

• Ming-Image-0.1-Design, 6B • Ming-Image-0.1-Design-Layer, 6B • Two open-source Agent Skills: the Ling UI Design Skill and the Image-to-Editable-PPT Skill Ming-Image-0.1-Design ranks #1 among open-weight models on Artificial Analysis’s UI/UX Design leaderboard.

better make no mistakes I've been running Qwen 3.8 Flash Next and it's a great driver for Hermes and Pi. I told it to add CUDA_DISABLE_PERF_BOOST=1 to reduce my server's idle power draw

Assign work to AI agents, like any teammate Discussion | Link

Don't know how they did it, but for under 10GB model, the results are astonishing. I am running it on Unsloth Studio. They just released the update, so if you are not seeing the option, I recommend updating your Unsloth Studio. Cheers!

Rabbit, the company behind the underwhelming R1 device, is rolling out a standalone AI agent that you don't need its hardware to use, as reported earlier by Wired. The startup says its new OS3 "agentic operating system" runs in the cloud but operates locally across Windows, Mac, and Linux devices. According to Rabbit, you can […]

British Columbia sues OpenAI, demands Tumbler Ridge shooter’s ChatGPT logs.

Qualcomm said that its new top chip can run 30B mixture-of-expert model locally.

In the hope of uncovering new details about ancient life, researchers have developed a large language model that fills in the gaps in papyrus fragments.

Two years after trying to sidestep mobile apps with dedicated AI hardware, Rabbit is launching OS3, a cross-platform agent that lives on the screens you already use.

The English hospitals recovered patient care data only.

Autonomy-1 will have a small, transformer-based AI model taking charge of a space probe.

Visual FoxPro stopped at version 9 in 2007. A surprising amount of it is still running, in 32 bits, because rewriting a 20-year-old business app is how you lose the business. A customer wanted to keep milking their app for the foreseeable future, so here it is: the same language on a new runtime (Rust, compiled to wasm, checked against the real vfp9.exe), tables no longer stopped at 2 GB, the old 32-bit .fll add-ins still loading, and lambdas, JSON and an HTTP server bolted on for good measure.

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Don’t be fooled by this summer of AI hype —Timnit Gebru, executive director of the Distributed AI Research Institute (DAIR), and Emily M. Bender, professor of linguistics at the University of…

For the people that are unaware or haven't seen the news yet.

Human video datasets offer an abundant and diverse source of interaction data that can complement expensive real-robot data. To bridge the human-to-robot embodiment gap, existing approaches either robotize videos in task-matched settings or address observation and action alignment separately at scale. In this work, we systematically examine whether robotized human videos can serve as an effective and scalable source of supervision for VLA pretraining. To this end, we develop a robotization pipel

OpenAI outlines priorities and principles for rigorous, secure, and independent third-party AI safety assessments of frontier models and safeguards.

Existing approaches to persona simulation with Large Language Models (LLMs) mostly rely on shallow character descriptions that fail to sustain coherent character behavior across extended interactions. We introduce Deep Persona, a psychologically grounded, three-layered architecture that organizes personas into hierarchical levels of observable expression, latent beliefs, and core motivational drives, for constructing highly convincing role-playing agents. Governed by the principles of scripted d

Basically the title.We did not get a new moe model with qwen 3.8 and Alibaba did not announce any small moe models on apsara.I know we might get an announcement later but ngl I kinda lost hope

The deterministic memory layer for AI Discussion | Link

We need studies on conversational AI (CAI) at scale to understand human behavior and shape CAI design. However, fragmented reporting of systems and study configurations hinders replication, extension, and knowledge accumulation. We present Gricea, an open-science platform representing studies as configurable, deployable research artifacts that researchers can run, inspect, share, and reuse. Informed by a formative analysis of prior CAI research, Gricea couples study procedures, participant-facin

I have been successfully hosting Vaultwarden for the last year or so and have had no major issues to write home about. It serves myself and my mum, but I'm intending to expand that to other family. What I'm interested in is understanding if I might be better of just using the official Bitwarden for self hosting instead of Vaultwarden. I'm concerned that if the Vaultwarden maintainer is let go from Bitwarden or changes priorities or whatever, then I'm at risk. One of the initial reasons for choos