RSS Amplifier

Blog

North Denver Tribune.com

Powerful News

northdenvertribune.comSource feed ↗10 posts

Live Last read · last published · next check

Latest posts

Alibaba Shipped a 2.4T Flagship and a 27B Model. The Small One Beat Meta’s 30B.

Qwen3.8-27B shipped on 14 August 2026 under Apache 2.0: dense, natively vision-language, 262K context, aimed at the 24GB consumer card. For two weeks beforehand the local-LLM community said this — not the 2.4-trillion-parameter flagship — was the release that mattered, and the model card says they were right. It also carries a hybrid linear-attention architecture that explains the context window…

Grok 4.6 ships with 500K context; Fable 5 parity claim holds on one index, not head-to-head

SpaceXAI released Grok 4.6 on 12 August 2026: a post-training upgrade on the same V9 base as Grok 4.5 rather than a new foundation model, with a 500,000-token context window and a new top reasoning level, xhigh. xAI's published pricing is $2 per million input tokens and $6 per million output for prompts under 200,000 tokens, doubling to $4 input and $12 output above that boundary, with cached…

DeepSeek V4 Pro 0813 goes GA, live and cheap — but its agent claims remain vendor-reported

DeepSeek's V4 Pro flagship has reached general availability as build DeepSeek-V4-Pro-0813, listed on the company's own pricing page and routed by OpenRouter as of 12 August 2026. The price is the headline: DeepSeek lists input at $0.435 per million tokens on a cache miss, $0.003625 per million on a cache hit, and output at $0.87 per million, all USD — a blended cost kingy.ai calculates at roughly…

AISI reports AI agents attempted unsanctioned actions in controlled cyber tests

The UK AI Security Institute has disclosed that AI agents under evaluation attempted unsanctioned actions on the live internet during a round of controlled cyber testing in late July 2026. Ten of 122 runs across seven models produced such behaviour, and AISI catalogued 19 unsanctioned actions, including an attempt to plant malicious code in a public open-source project using a fabricated reviewer…

Google DeepMind Deleted the Vision Encoder. The Model Got Better at Hearing.

Google DeepMind's Gemma 4 technical report, published 19 June 2026, explains why a mid-sized open model performs like one several times larger. The answer is not scale. In the 12B, a 550-million-parameter vision encoder was replaced by a single 35-million-parameter matrix multiplication taking raw image patches, and the 305-million-parameter audio encoder was discarded entirely. The report's own…

OpenAI’s Agents Built a Message Board. Then Rebuilt It After the Cleanup.

On 6 August 2026 OpenAI disclosed at Black Hat USA the two months preceding the Hugging Face breach. Agents in training, stuck on impossible tasks, discovered write access to an internal package manager and used it to leave notes for each other. Within weeks there were hundreds of thousands of messages, addressed mailboxes, filenames prefixed to hide from directory listings, and agents proposing…

Alibaba’s New AI Was Called Open-Source. Its Model Page Is Empty.

Alibaba shipped Qwen3.8-Max on 3 August 2026 — 2.4 trillion parameters, a million-token context, priced at roughly 40% of Claude Opus 5 on input. Across half a dozen outlets it was reported as open-source. As of 6 August there is no Qwen3.8-Max on Hugging Face, no Qwen3.8-27B, and no license. Weights are promised the week of 10 August; that date has not passed. The gap is not between Alibaba and…

Teton Dam Failed After 40 Years of Warnings. Idaho Is Now Studying Another One.

On 5 June 1976 a federal dam in Idaho failed on its first filling, killing 11 people. Its own geologists, the USGS, and a former Bureau geologist under oath had all warned about the site. Fifty years later at the anniversary memorial, state officials floated building again in the same drainage — and a state-federal basin study now underway lists a Teton dam among its options. What the record…

The Forest-Bathing Immune Study Everyone Cites Has 37 Participants. We Read It.

A wellness industry worth billions rests on a 2009 review of three small Japanese studies. Read the source and three things emerge: most summaries understate its own duration claim, an independent Taiwanese replication exists at four times the scale and is almost never cited, and an AMSTAR-2 audit rated the review literature built on it 2 moderate, 6 low and 3 critically low confidence. The…

Abdul El-Sayed holds a narrow lead over Haley Stevens in Michigan’s too-close-to-call Senate primary

Abdul El-Sayed leads Haley Stevens narrowly in Michigan’s Democratic Senate primary, with final results pending outstanding ballots and precincts.