AI — Page 22

The desktop update gives each named Bot its own model, memory and skills, with cross-Bot communication inside the same app.

The JAMA Perspective draws on simulated studies and challenges medicine's prevailing physician-led deployment model.

A tracker hidden in a roughly 1,000-book order led 404 Media to a Las Vegas site where workers cut bindings and scan pages.

Discovery produced 350 pages of prompts, including a request to show 3M was "0% at fault," before jurors assigned 30% of the blame to 3M.

MiniMax is pitching its general-purpose multimodal H3 model for game-sprite animation, though the supplied materials include no independent comparative benchmark.

The YC-backed startup says more than 40 shops use its software for booking, diagnostics, estimates, parts and payments.

WorldClaw takes USD1 payments and offers 43 models tied to Chinese companies flagged by the Trump administration, Reuters found.

The Uber Eats integration will begin in Houston and Dallas before expanding across dozens of U.S. cities.

Speko has raised $1.1M and received a $500,000 YC commitment, for roughly $1.6M total.

Bedrock's former Waymo and robotics engineers must prove autonomy can handle changing jobsites and lower contractors' cost per yard.

Philong Duong's startup says it received $100 million, but the federal program offers that amount across multiple autonomous-systems awards.

David Seaman's self-funded lab let one Mac verify context from another, then rebuild it inside a different embedding space.

Alibaba paired an Apache-licensed 27B model with downloadable flagship parameters on August 17th, but Qwen3.8-Max's license and official specifications remain unresolved.

The 27B open-weight model scored 48.0% across 60 WildClawBench agent tasks, trailing larger hosted systems while beating its Qwen3.6 predecessor.

The framework could require on-site human supervisors, kill switches, job-loss assessments and fees for fire response.

Math-AI, the open research community associated with Princeton researcher Yifan Zhang, built a local terminal assistant that stores and reuses successfully proved theorems.

Founders Demi Guo and Chenlin Meng connect Pika's creator pricing to a one-GPU streaming architecture while keeping model costs private.

Founder Spartak built the English-focused tool after AI grammar fixes made his own posts sound generic; paid rewrites run through Claude.

Decart's founders and Sequoia reportedly chose Anthropic despite a higher Nvidia offer; most of the consideration would be Anthropic stock.

The new routing can keep Sol in charge while assigning bounded work to faster, lower-cost Luna subagents.

The cost decline is helping Luis von Ahn move conversation practice into Super and reconsider Duolingo's premium Max tier.

A 741-turn, 231 MB thread loaded in 1.66 seconds in a test, with lower memory growth and 98% fewer requests.

In a 50-task run of Newsroom Reliability v0.2, SpaceXAI’s Grok 4.6 ranked first with a score of 0.78. OpenAI’s GPT-5.6 Luna Pro followed at 0.77 with the lowest reported cost among the top three, at $0.0005 per task.

Netflix engineers Ying Li, Arjun Rao and Shradha Sehgal tested an LLM-backed ranker using about 40 times fewer Phase 2 labels than the production model, reporting a 0.006% relative lift in an undisclosed core metric.