Stories from labs, community, research and press, tagged by concept.https://justuju.org/2026-10-13T21:50:42ZJustujuurn:justuju:68e3de130cf52026-10-13T21:33:21ZNVIDIA Technical Blog

Enterprise applications have become increasingly demanding. Agentic AI has compounded the need for vast amounts of enterprise data with frequent, complex... [The agent loop]

urn:justuju:a17663b427bf2026-10-13T21:23:05ZOllama releases

launch: allow usage credits without upgrading ( #18985) [Ollama]

urn:justuju:082a179fcd582026-10-13T20:45:25ZThe Verge AI

An OpenAI engineer was arrested after the company said it discovered a "sophisticated recording device" hidden in one of its offices. [GPT (OpenAI)]

urn:justuju:70093601d1462026-10-13T20:36:44ZTechCrunch AI

A federal judge ruled that Apple can access OpenAI's interview materials involving former Apple employees, as well as private messages and testimony from key… [GPT (OpenAI)]

urn:justuju:cdd5fd8d13262026-10-13T20:10:33ZAWS Machine Learning Blog

Learn how the AWS Generative AI Innovation Center Partner Agent Factory and Anchor Browser built AI agents that autonomously pay for x402-protected web resources and APIs. Amazon Bedrock AgentCore payments checks budgets, signs transactions, and retries requests without exposing wallet credentials to the agent runtime. [Bedrock, Vertex, Azure AI Foundry, The agent loop]

urn:justuju:72146ea92bf42026-10-13T19:59:31Zr/LocalLLaMA (top today)

I'm excited to see our Lemonade embedded SDK getting used out there! [Gemma, Llama, Mistral, llama.cpp, Ollama]

urn:justuju:25432b6b093a2026-10-13T19:43:07ZGoogle DeepMind

[Gemini (Google)]

urn:justuju:c839b9474fae2026-10-13T19:42:42ZThe Verge AI

More AI agents will soon be able to browse and purchase from Shopify listings, marking an expansion of the company's agentic commerce ambitions. [The agent loop]

urn:justuju:4a4a81eb8e4e2026-10-13T18:00:49ZThe Decoder

Google is making Gemini 4 Argon available worldwide through its API, the Gemini app, and Google Antigravity. [What 'frontier' means now, Gemini (Google), Consumer AI assistants]

urn:justuju:fa1a37fdc4162026-10-13T17:30:00ZTechCrunch AI

University of Michigan spinout Laplace Intelligence uses AI agents and interactive simulation to stress-test robots and self-driving vehicles before deployment. [The agent loop]

urn:justuju:4826ab478a6a2026-10-13T17:17:25ZAWS Machine Learning Blog

Managing AI model access across dozens or hundreds of AWS accounts creates friction when each account must accept AWS Marketplace Private Pricing Offers individually. This post shows two ways to activate MPPO entitlements in AWS License Manager: a lightweight script for a one-time fix, and CDK automation for ongoing… [Bedrock, Vertex, Azure AI Foundry]

urn:justuju:d1d86d368e3b2026-10-13T17:15:27Zr/LocalLLaMA (top today)

I'm sure some here read this at the time and remember, but I recently found this hilarious Guardian article from 2019 through Wikipedia: Some choice quotes: "OpenAI... says its new AI model, called GPT2 is so good and… [GPT (OpenAI)]

urn:justuju:b7825d7053512026-10-13T17:14:08Zr/LocalLLaMA (top today)

Gguf's: Free Demo: "LFM2.5-24B-A2B is a 24B-parameter reasoning model with only 2B active parameters, built for fast, private agents on-device. It is the largest and final model in the LFM2.5 family, which now spans two orders of… [Mixture of Experts (MoE), Reasoning / thinking models, Small & on-device models]

urn:justuju:550c4c02e9ff2026-10-13T17:05:48ZHugging Face Blog

[Small & on-device models, Mixture of Experts (MoE)]

urn:justuju:12e59a3a5c8b2026-10-13T16:07:55Zr/LocalLLaMA (top today)

[Agent memory]

urn:justuju:9587599774902026-10-13T16:03:31ZAWS Machine Learning Blog

Orca Security built Discovery AI, a natural language query agent on Amazon Bedrock that converts plain-English questions into Sonar queries. The post shares the multi-stage architecture and the results: 90% lower query cost, 93% fewer input tokens, and 75% lower latency than the initial baseline. [Bedrock, Vertex, Azure AI Foundry]

urn:justuju:d1f8258ec2122026-10-13T16:03:16ZHacker News (best, AI-filtered)

[Gemini (Google)]

urn:justuju:8b0463614a7b2026-10-13T16:01:58ZGoogle DeepMind

[Gemini (Google), World models & physical AI]

urn:justuju:316153953adf2026-10-13T16:01:52ZGoogle DeepMind

[Gemini (Google)]

urn:justuju:2bad0427cfe42026-10-13T16:00:02ZArs Technica AI

Developers will get 50 percent discounted API rates through November. [Gemini (Google), Consumer AI assistants]

urn:justuju:66cb1e68848f2026-10-13T16:00:00ZGoogle AI Blog

TBD [Gemini (Google)]

urn:justuju:24ef7cf40bd22026-10-13T16:00:00ZThe Verge AI

After months sitting on the sidelines of the frontier AI race, Google has finally launched its long-awaited Gemini 4 Argon model, claiming it can compete with… [Gemini (Google), What 'frontier' means now]

urn:justuju:576253ed89472026-10-13T15:54:49ZAWS Machine Learning Blog

ShipStation Global's engineering team built Deckhand, an AI IT help desk agent for Slack on Amazon Bedrock AgentCore. It answers routine IT questions from a curated knowledge base, reads error-message screenshots, and files Jira tickets. In its first three months it resolved 97.5% of channel queries with a 96%… [Bedrock, Vertex, Azure AI Foundry]

urn:justuju:c4d6082c07a12026-10-13T15:52:20ZAWS Machine Learning Blog

Grok 4.7 is now available on Amazon Bedrock. In this post, we build a tool-calling agent on Grok 4.7 and evaluate it on three reasoning tasks at four reasoning effort levels, measuring correctness and token cost per completed task. We show how to tune the trade-off per request and monitor it in production. [Bedrock, Vertex, Azure AI Foundry, Grok (xAI), Reasoning / thinking models, Tool / function calling]

urn:justuju:8c17f1f1ac302026-10-13T15:49:56ZHacker News (best, AI-filtered)

[Claude (Anthropic)]

urn:justuju:ba8a6ffbbe0f2026-10-13T15:17:09ZAWS Machine Learning Blog

AWS engineers share kernel optimizations for fused Mixture-of-Experts (MoE) models on Amazon SageMaker AI: Swap-AB for WGMMA utilization, Triton backend routing for FP8 quantization, and eliminating Triton JIT recompilation. Together they deliver up to 10% higher throughput and up to 9% faster Time to First Token on… [Mixture of Experts (MoE), FP8, FP4, NVFP4 & MXFP4, Quantisation basics]

urn:justuju:1ac263d2b9b02026-10-13T15:02:59Zr/LocalLLaMA (top today)

The 3090 Founders edition has a reputation for hot VRAM. Since LLM token generation maxes out VRAM's read capacity, I checked VRAM temps with Nvidia's stock fan/power curves and discovered that VRAM… [VRAM math, Multi-RTX home rigs]

urn:justuju:4e49d23a4ed62026-10-13T14:49:31Zr/LocalLLaMA (top today)

Here is how i serve 40 people (10 of them developers) with local AI services. n8n for automatisation whisper and Vexa AI, a bot joins Teams and makes transcripts, n8n later extracts decisions and todos OpenWebUI, ChatGPT-like webinterface for everyone, most agentic features are disabled, websearches via Tavily are… [n8n, Dify & workflow tools, GPT (OpenAI)]

urn:justuju:01fed557964c2026-10-13T14:19:40ZNVIDIA Technical Blog

NVIDIA is making the full NVIDIA Run:ai platform available as open source software, giving customers, developers, and ecosystem partners access to... [AI datacenters & energy]

urn:justuju:e890334130962026-10-13T14:08:21Zr/LocalLLaMA (top today)

Dynamic vision, in action Basalt is an inference engine for 3.8 Flash Next, forked from Strata and heavily tuned for Blackwell. It supports vision, and for this to work you need to have the image encoder either in VRAM or CPU all the time, having it on VRAM takes valuable space from experts that would otherwise stay… [NVIDIA Hopper → Blackwell → Rubin, Qwen (Alibaba), VRAM math]

urn:justuju:cd375fccf76f2026-10-13T14:04:51Zr/LocalLLaMA (top today)

Follow-up to last week's post about streaming MoE experts from disk in a browser tab. That ran on hand-written engines for two models. This week the same disk tier sits under llama.cpp's own WebGPU backend, so it works for the MoE models llama.cpp already runs, with no per-model engine. The video is one real recording… [llama.cpp, Qwen (Alibaba), Gemma, Llama, Mistral, Mixture of Experts (MoE), GGUF & k-quants]

urn:justuju:48866c22a6372026-10-13T13:50:19ZTechCrunch AI

The new X Subscription combines X's paid features with AI tools like Grok and Cursor, offering plans ranging from $8 to $200 per month. [Grok (xAI), AI IDEs (Cursor, Windsurf, Copilot)]

urn:justuju:7b4ab1a30d912026-10-13T13:49:31Zr/LocalLLaMA (top today)

I built a desktop app harness that keeps local meeting notes and can use them to autocomplete my typing as well as complement my dictation with additional context via reading my screen or using the previous meetings notes. It's been a bit of a struggle to get it to work well and also to not be too intrusive and laggy… [Apple Silicon Macs, Gemma, Llama, Mistral]

urn:justuju:8fb9aa71a73e2026-10-13T13:16:08Zr/LocalLLaMA (top today)

Anthropic says we can run Gemma4 26b a4b with only 14 smart tvs. [Claude (Anthropic)]

urn:justuju:bc6b0eca02362026-10-13T13:00:00ZTechCrunch AI

American Express is introducing a new framework designed to ensure AI agents only make purchases their users have authorized, potentially preventing costly… [The agent loop]

urn:justuju:ffc748cb8d882026-10-13T12:00:00ZOpenAI News

Higgsfield brings creators longer videos with GPT-6 Astra, precise visuals with GPT-Image-2.5, and a path to voice filmmaking with GPT-Live. [GPT (OpenAI)]

urn:justuju:b7472565c79d2026-10-13T11:50:07Zr/LocalLLaMA (top today)

So I wanted to try the horsey challenge myself on the latest engine i been testing which is a strata fork by sergqwer. I seen a post earlier and seen i believe it was a iq3 i think and it was pretty dope so i thought i try out the NVFP4. I wasn't expecting to share but i… [FP8, FP4, NVFP4 & MXFP4, Qwen (Alibaba)]

urn:justuju:d647463237212026-10-13T11:50:03Zr/LocalLLaMA (top today)

This is an update to my previous post which can be found here: Since my last post I have continued working on the conversion process and also converting new models. My latest model is the Qwen2.5-1.5B-MoE-A0.9B. I have also released GGUF versions for both the 0.5B and… [Mixture of Experts (MoE), llama.cpp, GGUF & k-quants]

urn:justuju:0f8702d20d782026-10-13T11:27:14ZHacker News (best, AI-filtered)

[Vision, speech & generative media]

urn:justuju:7517cf52afea2026-10-13T11:11:44Zr/LocalLLaMA (top today)

Got this couple. It was difficult purchase in this economy but paid only roughly $6100 equivalent for each in my country. I justified the purchase for learning (so I try to convince myself) and coding workflow improvements. I started looking into recipes for GLM 5.3 flash but surprisingly there is very little chatter… [NVIDIA DGX Spark]

urn:justuju:625fccaa6d732026-10-13T11:06:46ZThe Decoder

Anthropic says it has cut the extraction of Claude's internal reasoning tokens by 90 to 99 percent. [Claude (Anthropic), Distillation, pruning & synthetic data]

urn:justuju:459735cd30fd2026-10-13T10:33:28ZThe Decoder

Small AI models may be just months away from spreading on their own across poorly secured consumer devices like smart TVs and robot vacuums, according to a new… [Claude (Anthropic)]

urn:justuju:c48f77cfce802026-10-13T10:00:00ZMIT Technology Review AI

MIT Technology Review Explains: Let our writers untangle the complex, messy world of science and technology to help you understand what’s coming next. [Open weights vs open source]

urn:justuju:d5672e7604792026-10-13T09:57:21Zr/LocalLLaMA (top today)

My project is optimising qwen flash next on Mac Mini m5 64gb through SSD streaming. I recently asked LocalLLama for guidance on how to optimise token output. Today a popular method to do this is a lookahead guesser which fetches experts one layer ahead of the current layer. One popular suggestion was building a… [Qwen (Alibaba)]

urn:justuju:ec132ecc411f2026-10-13T09:37:04ZLatent Space

MSL’s emergence as a frontier lab is finally complete. [What 'frontier' means now, Claude (Anthropic)]

urn:justuju:19274c938d072026-10-13T08:21:05Zr/LocalLLaMA (top today)

[NVIDIA DGX Spark, AMD Strix Halo (Ryzen AI Max)]

urn:justuju:e5a7015739f12026-10-13T08:17:38ZOpenCode releases

release: v2.0.28 [OpenCode]

urn:justuju:324d401eae582026-10-13T05:26:17ZOllama releases

What's Changed llama: update llama.cpp to b11521 for decision models and EmbeddingGemma 2 by @jmorganca in #18917 server: reuse verified manifests and loaded models in inference routes by @dhiltgen in #18947 mlx: compact restored recurrent state after speculative rollback by @dhiltgen in #18955 launch: include… [Ollama, Gemma, Llama, Mistral, llama.cpp]

urn:justuju:c4b53f7c6d1f2026-10-13T04:13:29Zr/LocalLLaMA (top today)

[FP8, FP4, NVFP4 & MXFP4]

urn:justuju:64c0e2c4c4872026-10-13T04:03:55Zr/LocalLLaMA (top today)

Didn't realize the V100 were still so fast, anyone have first hand experience on these? [Multi-GPU parallelism, NVIDIA DGX Spark, Benchmarks that matter]

urn:justuju:aa4b8849242e2026-10-13T01:01:00Zr/LocalLLaMA (top today)

A lot of projects have come up recently consisting of specific hardware config specialized forks of existing conferencing engines, but they all seem to be made using Claude or OpenAI's biggest frontier models. So given that open source is about 4-6 months behind, can we reasonably expect that local runnable models… [What 'frontier' means now, Claude (Anthropic)]

urn:justuju:c6054984e0072026-10-13T00:27:09Zr/LocalLLaMA (top today)

[Claude (Anthropic)]

urn:justuju:d87f52db72f52026-10-12T23:18:32Zr/LocalLLaMA (top today)

After running a benchmark test on Exo in Opencode and being amazed by the result, I thought I'd try other models on Opencode. Many were surprisingly terrible. Especially some mid level models. Earlier posts: [Qwen (Alibaba), Benchmarks that matter, OpenCode]

urn:justuju:653273f51bcd2026-10-12T22:42:31Zr/LocalLLaMA (top today)

The newly named D-Series GPU has up to 480GB of LPDDR5X memory, 1.5TB/s of memory bandwidth Intel rates it for up to 1.3 PetaFLOPS of FP8 performance, with support for formats ranging from FP4 to FP64. It features a total of 32 Xe3P cores, 256 XMX engines, 32 MB of unified L2 cache. No cost estimates currently, I hope… [FP8, FP4, NVFP4 & MXFP4, Memory bandwidth is king]

urn:justuju:aafb1b36fddb2026-10-12T20:49:49ZvLLM releases

Signed-off-by: Matthew Bonanni mbonanni@redhat.com Co-authored-by: Claude Opus 5.5 noreply@anthropic.com [vLLM & SGLang, KV cache & prompt caching, Claude (Anthropic)]

urn:justuju:811c660e23802026-10-12T20:14:03Zr/LocalLLaMA (top today)

I get a lot of hate whenever I tell people I run a Q4 quant of Qwen3.8 27B with 262k KV 80t/s with vision, all while running 2 screens on windows on the same card. People seem to think I am lying or making it up. I am really just trying to help people. I am saddened whenever I see fellow 16GB VRAMers stuck on 60k… [VRAM math, Qwen (Alibaba), GGUF & k-quants]

urn:justuju:e02c3eb04ac62026-10-12T20:13:48Zr/LocalLLaMA (top today)

[Claude (Anthropic)]

urn:justuju:865d5757827e2026-10-12T19:25:28ZServeTheHome

Intel today has unveiled its official performance specifications and branding for its upcoming "Crescent Island" data center AI inference cards. [AI datacenters & energy]

urn:justuju:e738abb5d3be2026-10-12T18:58:37ZOpenCode releases

Co-authored-by: uyencss uyencss@users.noreply.github.com Co-authored-by: Aiden Cline aidenpcline@gmail.com [OpenCode, Cline, Aider, Roo & co]

urn:justuju:2da9ef0d9c242026-10-12T18:48:29ZThe Decoder

Meta's new Muse Wonder models rank in the top three on six out of ten benchmarks, with the biggest gains in agentic tasks. [What 'frontier' means now, The agent loop, Benchmarks that matter]