2026-09-02

Yoyo I'm back with updates to the fastest inference engine with minimal quality loss for Qwen3.8-27B. The last few weeks I've been optimizing decode speed and I don't think it can be pushed further, until a newer/better drafter is invented. So I focused on prefill, which I this morning was around 1.300 per second at 4k and now is just below 2.000. The main improvement came from a custom kernel, which matches the quality of fp32 with 0.99997 similarity at int8. Try all of the improvements here:

Fable 5.1 includes changes meant to reduce token cost and false-positive restrictions from the model's safeguards.

A few weeks ago we had a bit of drama on this sub when Planka moved its SSO functionality behind a paid tier. And just a few days ago, we had a "shame" list of self-hosted sso tax apps announced here. I decided to compile the opposite, as I try to host things on my homelab that can be integrated with PocketID directly .. is a list of self hosted applications that provide SSO for free with the option to use a custom OIDC provider. Not included are apps that only support Google OAuth or similar. T

More new massive updates from turboderp: - CPU offload of MoE experts - Qwen-3.8-Flash-Next ngram disk offload - GLM-5.3-Flash - New self-calibrated optimization technique - Countless other optimizations and improvements If you have an NVIDIA card and haven't tried it lately, you might be missing out. The attached cat image was made with Qwen-3.8-Flash-Next-3.05bpw-exl3 and this prompt: Create a detailed SVG image of a cute kitten riding a magic turtle into space. Come join the crew at the exlla

Due to the nature of quadratic attention, Large Language Models (LLMs) consume a lot of memory and energy. Every new token costs more than the previous one. For each additional token, the keys and values must be stored in memory indefinitely, which is unsustainable. Several alternatives have been proposed to fix the quadratic scaling problem, one of which is retrofitting LLMs to use Linear Attention. This idea has attracted a lot of attention, given its promise to solve the quadratic scaling pro

I am absolutely blown away. Yes my setup is crap but the fact that it managed to do this in a single take is unbelievable (and I'm a developer). Hardware used: - Windows PC with 4070ti (12GB VRAM, 32GB RAM) - Macbook M5 Air (LLAMA.cpp RPC connection to Windows PC) Software used: - LLAMA.cpp (Q4KM, xhigh, 8bit KV, MTP=1) - Lmstudio Qwen 3.8 27b (Q4KM) GGUF - Deepseek harness (mode: minimal) Prompt: "please create a fully self-contained super mario game with only one short level, put everything in

How do you benchmark a web search API when the thing being tested can read the answer key? A search agent has a fetch tool. If the gold labels sit in a public dataset, the agent can download them mid-evaluation and skip retrieval entirely. A similar problem arises when the answers are already encoded in […] The post Keenable AI Open-Sources NEEDLE: A Live Search Benchmark That Rebuilds Its Query Set Every Hour appeared first on MarkTechPost .

Today we’re announcing Enterprise Frontier Safeguards (EFS), a solution that combines the privacy of zero data retention (ZDR) with state-of-the-art safeguards for detecting misuse. EFS works by storing data in cloud infrastructure controlled by the customer, not Anthropic. EFS will be rolling out to customers in phases, starting later this fall. To make the transition smooth, eligible customers will receive ZDR on Fable 5 and Fable 5.1 until EFS is ready. We developed EFS in close collaboration

Google Research has released TimesFM-3, a 330 million parameter time series foundation model that forecasts multiple related series in a single forward pass. Unlike every TimesFM checkpoint through 2.5, it is pretrained natively for multivariate forecasting, accepting multiple targets, past covariates, and past-future covariates with no task-specific fine-tuning. It takes the top average rank among pretrained foundation models on GIFT-Eval, fev-bench, and the TIME leaderboard. The weights, howev

Apple is pushing for "expedited discovery" in its legal battle against OpenAI over concerns the company is actively destroying evidence, as reported earlier by Bloomberg. In a filing on Monday, Apple alleges OpenAI only just handed over a MacBook used by a former employee at the center of the lawsuit, which contained discussions about "destroying […]

The latest release from Meta Superintelligence Lab is a powerful transcription model.

A few notes on Anthropic's new Claude Fable 5.1 - with Max thinking level I got the best SVG pelican I've had from any Anthropic model (at a hefty cost of $3.30!), which I then had it animate simonwillison.net/2026/Sep/1/c...

We present CogEvol, a family of models trained specifically for Learning Environment Generation: turning a course brief into a finished learning artifact (structured-JSON slides or self-contained interactive HTML pages) in a single pass. Across 220k production requests, CogEvol completes a slide in a median of 17 seconds and an interactive page in 59, replacing minutes-long multi-turn agent scaffolding. Reliability is enforced rather than hoped for: a production-grounded data pipeline turns real

Organizations often develop and maintain portfolios of related applications: independently deployable codebases that share substantial domain logic, interface patterns, or operational conventions. As LLM coding agents are increasingly used to generate and maintain such software, a naive application-by-application workflow duplicates shared logic across codebases and allows prolonged agentic maintenance to accumulate verbosity, dead code, and structural erosion. We introduce the Super Library Age

I was browsing HF for small LLMs and run into this model. It does not seem to be a fine tune - the model has its own architecture. There are 4B/1.7B versions - the benchmark is quite interesting (4B is neck and neck with Qwen 3.5 9B). The HF page claims both models support native 1M context size . Currently does not run out of the box on llama.cpp - pending this PR: They have a custom fork of llama.cpp that works. Anyone has tried this? Update: GGUFs (require custom fork for now):

Recent advances in large reasoning models (LRMs) have shown that reinforcement learning with verifiable rewards (RLVR) can substantially improve reasoning in mathematics and code, where outcomes can be checked automatically. Extending this progress to open-ended and agentic tasks remains difficult because reliable rewards are harder to obtain and direct human supervision cannot keep pace with the scale and complexity of model-generated experience. This paper studies how LRMs can continue to impr

You can now create decent video faster than you watch it. This is the start of... something. We’re not sure what.

连发六篇Nature期刊，复旦Neolab打通全尺度生命推演

Speed and accuracy usually pull against each other in text-to-speech. Gradium AI's new default model reports both: an 81.0% human-rated pass rate on 500 hard sentences across five languages, at 216 ms P50 time-to-first-audio on Coval. The evaluation set is open on Hugging Face under CC BY 4.0. The post Gradium AI Releases New Default TTS Model: 81.0% Hard-Case Pass Rate at 216 ms Time-to-First-Audio appeared first on MarkTechPost .

Connect Cosmic agents to any service with an MCP server Discussion | Link

Just to name a few: running Qwen3 8B on a 32GB GPU running Qwen3.6-27B Q4_K_M on 2 x R9700 quote: "each prompt was sized at 500 input and 500 output tokens" for a full system that costs $18,775?? I don't understand what they are doing. Am I reading something wrong?

A native AI agent that does real work on your Mac Discussion | Link

Google has reportedly been reaching out to a number of Hollywood's biggest studios, hoping to strike licensing agreements that would allow it to train its AI models on copyrighted material in exchange for massive piles of cash. In theory, these deals would be a win-win: a huge financial boon to the studios that would also […]

X is investigating a wave of unsolicited password reset emails that it believes may be tied to the rollout of its new payments service.

Depending on who you ask, developer platform Hugging Face was recently attacked by OpenAI - after it lost control of its own AI tools - or by a succession of AI "civilizations." Welcome to the linguistic battlefield of AI safety, where word choices can shift responsibility for a massive cybersecurity incident from a company to […]

AI model-training startup AfterQuery has reportedly raised a round that valued it at $3.2 billion, just five months after announcing its $30 million Series A at a $300 million valuation in April.

OpenAI said that the integration provides read-only access to health records for clinicians.

The startup wants to do for IT infrastructure what Cursor did for software engineering.

Claude Code limits and Infinite Slop

ChatGPT can now connect to trusted healthcare data, helping clinicians securely access patient context, medical research, and more.

While Large language models (LLMs) incorporate user personalization signals to improve usability and helpfulness, they increasingly shift from providing balanced, informative responses toward optimizing for user satisfaction when conditioned on personal context such as conversation history, inferred preferences, and user profiles. Specifically, we identify three emerging risks: (1) irrelevant personalization, where models reference personal information in unnecessary contexts; (2) preference nar

arXiv:2608.27524v1 Announce Type: new Abstract: Corporate training teams need scalable and explainable tools to improve workforce communication in multilingual settings. Existing systems often score text, audio, or video in isolation, or produce black-box outputs that are difficult to audit for coaching use. This paper presents SETU, an agentic ecosystem for corporate communication coaching aimed at recruiters, frontline sales professionals and training units who prepare for audience specific co

arXiv:2608.27471v1 Announce Type: new Abstract: Fallacies are arguments that employ invalid reasoning, making their automatic detection critical in sensitive contexts such as high-stakes political debates, where public opinion is shaped. Spotting a fallacious argument requires contextual knowledge beyond its pure surface text. This entails world knowledge pertaining to the subject matter under discussion, as well as knowledge of the relationships that exist between arguments within the argumenta

I built an online viewer and editor for .md files where the file stays on your machine. Free to use and code is available under MIT License here: You can also install it as a Chrome/Edge app so that it works for files on your desktop by double-clicking. Supports standard .md files with formatting, syntax-highlighting for code, table of contents, Mermaid diagrams, LaTeX via KaTex. Also supports exporting to .html, .pdf and .png. Please give it a try and share your feedback for improvements. Comme

Strategic deception by LLM and VLM agents has emerged as a central AI alignment and safety concern. Social-deduction games (where each player holds a hidden role and communicates with others to deduce identities) serve as the canonical testbed, particularly in multi-agent settings. Existing testbeds, however, are text-only and run on a single fixed agent configuration, missing the non-verbal sensorimotor channels treated as core by deception taxonomies and leaving it ambiguous whether an observe

CLI based Zero-Knowledge Architecture for code repositories Discussion | Link

Every model with a unified interface. Privacy with proof. Discussion | Link

I'm the developer. I built this because AI-generated conspiracy videos were getting out of control in my own YouTube feed. Weedout is a $1.99 Safari extension for macOS that removes videos YouTube labels “Made with AI” from feed, search, related videos, playlists and Shorts. It uses YouTube's label rather than AI detection, runs locally, and does not catch unlabeled videos. Source for developers who want to fork or build: Pull requests are not accepted; the repository is for independent forks. C

Hi! Every month the "Who Wants to Be Hired?" and "Who's Hiring?" Threads come through I would think to myself, "there should be a match maker for these two threads". So I had abacus.ai whip one up. The methodology is pretty simple, data is extracted from posts using an LLM, score matches based on salary, domain experience, remote/onsite, etc., and then eliminate incompatible postings, like 'looking for remote work' and 'onsite only'. The result is two views: jobs-by-user and user-by-jobs. If you

arXiv:2608.27464v1 Announce Type: new Abstract: This position paper argues that human-centered explainable AI (HCXAI) should incorporate insights from the psychology of information seeking. Drawing on Sharot and Sunstein's framework of information-seeking motives, we propose that people evaluate whether to engage with explanations based on three types of expected utility: instrumental (will it help me act better?), hedonic (will it make me feel better?), and cognitive (will it improve my underst

With Google Pics, Google is pushing deeper into the creative software market dominated by Canva and Adobe, but with a distinctly AI-first approach.

A nonprofit organization called the Fermi Explorer Mission announced today that it intends to launch a spacecraft to our nearest star system by the end of 2029. It’s a hugely ambitious mission—if all goes well, the spacecraft could take up to 80,000 years to arrive at Alpha Centauri, which is 4.4 light-years away. And the…

Tool: GeoJSON Map Viewer I was helping Natalie gather some maps of local political boundaries (for the Granada Community Services District and the Midcoast Community Council and found a need to display some GeoJSON files on a map and export that as a PNG. I asked GPT-5.6-Sol for suggestions of tools and it proactively built one. After some iterations using Claude Code for web and Fable 5.1 we got to this finished tool. As for the GeoJSON.. it turns out if you ask ChatGPT Work to provide boundari

Quantitative research agents that write their own experiments can corrupt the evidence they later learn from. A leaky feature that scores well gets stored as a successful precedent and propagated through later iterations. Prompt-level instructions and reviewer agents do not close this, because author and reviewer share the same blind spots. A team of researchers […] The post Researchers from Princeton, Ant Group and Stanford Introduce AQuA: A Two-Part Agentic Framework for Autonomous Factor Disc

Hey HN, I’m Antonio from Nori Robotics ( ). We build a $1,688 bimanual mobile robot in San Francisco for robotics developers and researchers. I started working on Nori while doing robotics research at Columbia. I was teaching robots through human demonstrations, but getting my hands on affordable hardware was difficult. Most labs have one or two expensive robots, which makes it hard to collect large datasets, run long experiments, or test across several robots. So I built my own. After seven ite

arXiv:2608.27484v1 Announce Type: new Abstract: Artificial intelligence is transforming personalized healthcare, yet fragmented clinical, self reported, and wearable evidence remains difficult to interpret and trace. We present CareGraph, an auditable hybrid AI framework that converts heterogeneous records into prioritized trends, missing context indicators, bounded next steps, discussion questions, and provenance linked explanations. CareGraph organizes evidence without diagnosing, predicting o

Lawsuit: Anthropic’s torrenting totally screwed songwriters as AI songs top charts.

TL;DR: learned how to automate SSL certificates in my homelab. Read on if you’re bored haha.