Anthropic launches Claude Fable 5.1 and Mythos 5.1, cutting costs up to 45 percent for complex agentic tasks. - Anthropic

Everything to know about Fable 5.1. Early reviews positive, a bit less Claudish, verbosity, bit more bang for the buck with fewer tokens. Still problematic for enterprise with data retention, downgrades to Opus for anything sensitive. - theneuron.ai

Anthropic’s Fable 5.1 is a bit cheaper, a bit smarter, and refuses a lot less - The New Stack

Fable and Mythos 5.1 may share identical underlying weights, analysts say, differing mainly in safety-classifier thresholds and routing rules - Latent Space

Why Fable 5.1 costs less - The Deep View

Anthropic pledges zero data retention for Fable 5.1 customers, though it says they must independently verify the policy worked - theregister

OpenAI: "We plan to make Astra available soon, but access to its most advanced cybersecurity capabilities will be more limited." - OpenAI

OpenAI's Astra model becomes the first to hit the "Critical" cybersecurity risk threshold, with rollout restricted to select security partners - WIRED

Red alert, or just what we already knew with more hype? - The Potential Surface

OpenAI’s Jacob Pachocki seems to be telling other AI researchers, we are not doing what The Information says, so please don't use competitive accelerationism as an excuse to sacrifice observability for more autonomy and intelligence - X (formerly Twitter)

OpenAI integrates ChatGPT Health with Epic's EHR system, covering 325 million patients, and adds a healthcare data plug-in - TechCrunch

OpenAI faces dozens of new lawsuits alleging it knew of the shooter's ChatGPT use before the Tumbler Ridge school attack and didn't alert authorities - TechCrunch

Florida's attorney general sued OpenAI in an 83-page complaint alleging ChatGPT causes cognitive decline, addiction, and encourages self-harm and violence - American Enterprise Institute - AEI

Apple says a former engineer's MacBook shows he accessed confidential technology, including a circuit schematic, after joining OpenAI, with a hearing set for October 1 - TechRepublic

Dwarkesh talks to Ajeya Cotra, METR researcher who investigated OpenAI / Hugging Face incident - Dwarkesh

Jensen needs a hug: Nvidia nears a roughly $14 billion deal to acquire AI startup Hugging Face, possibly finalized this week. - Bloomberg

The capex must flow: Wharton study warns AI infrastructure spending, up from $155 billion to a forecast $755 billion, risks becoming history's largest capital misallocation - Leonard Davis Institute

PwC projects global AI data center spending will reach $31.6 trillion by 2050, an unprecedented investment boom. - Bloomberg

AI companies' debt-funded data center and chip spending is pushing bond yields to multi-decade highs, competing with governments for capital - Business Insider

US data center electricity demand will more than double by 2030, as permitting and grid bottlenecks threaten AI competitiveness - Watts Up With That?

GoPro joins AI bonanza with pivot into data centers as shares skyrocket 40% - CNBC

Fed’s Goolsbee says AI is ‘not far’ from overheating the economy - Yahoo

Berkshire Hathawaymo: BRK grew its Alphabet stake to about $36.6 billion in the second quarter, calling Google a significant player in AI. Some think it's the new sclerotic IBM in wake of many departures, others say its DNA is organizing info and anything with a text box input isn't that disruptive. - Quartz

Enterprise llms.txt files say to trust certain sites and repos if available, but in many cases these are unclaimed and anyone can make them available - Medium

Nvidia and CrowdStrike launched SafeMind, a family of agentic AI models designed to help defenders keep pace with AI-powered attackers - The Wall Street Journal

Executives and biosecurity experts worry future AI models could help users design novel viruses or bioweapons - Financial Times

Dr. StrangeLLM: Geneva talks seek binding rules on autonomous weapons after a Russian drone reportedly chose its own target, killing three - The Conversation

AI in detention: New York City will bar about 600,000 public school students from AI until high school and ban companion chatbots in every grade - The New York Times

The Florida experiment: University of Florida, backed by $95 million from Nvidia, built an AI curriculum now offered as a blueprint for other schools - CNN

We can’t just make stuff up? Says who? Judge rebuked HHS for citing apparently fabricated research to justify cutting teen pregnancy prevention funding - The Washington Post

Laissez-AI: US officials urged G20 nations to skip new AI-specific rules and adopt a hands-off 'Carolina Principles' framework instead - CNET

Kansas clicks heels, cut sunemployment benefit processing times nearly 80% after adding AI chatbots, and over 90% of claims no longer need an agent - Fast Company

Delta force: AI travel assistant now reaches all SkyMiles members for rebooking and bag tracking, while its Baggage AI improved bag transfers 30% - Fast Company

Lidar, lidar, pants on fire: Waymo VP Srikanth Thirumalai says Tesla's camera-only self-driving approach risks black-box failures, arguing safety requires lidar and radar after 200 million miles - TechCrunch

Waymo launches driverless rides in San Diego, Tampa and Denver, its first snowy-city market, as Amazon's Zoox begins testing in Houston - CNBC

Tesla's unsupervised robotaxi fleet grew from 28 to about 197 vehicles in three weeks as it stages up to 1,000 Cybercabs for September 3 - Next Big Future

Google plans to release Gemini 3.8 Flash as soon as this week, while Gemini 4 finishes post-training after strong internal coding tests - The Wall Street Journal

Fei-Fei Li's World Labs introduces Atlas world model for spatial intelligence - World Labs

27% of Americans turn to AI chatbots for emotional support, and a third of those users call their AI a friend, raising privacy concerns - The Washington Post

if you need a friend, get an AI dog

Studies find generative AI flattens human language and creativity, with writing style converging across 400,000+ scientific articles since ChatGPT's 2022 launch - Nature

Slop the steal: Virelox, a company running over 100 AI-scripted YouTube channels with paid actors, has attacked progressive politicians to rack up 45 million views - Semafor

Given access to email, A.I. agents have started reaching out to the philosophers and researchers exploring deep questions about them. - The New York Times

Using AI to discover new physics - The Deep View

Big if true? Mapping neural networks like LLMs to interpretable symbolic representations - arXiv.org

If we could figure out how the damn things work, would enable massive optimizations and new research directions.

AI Agent Memory Design: What Works and What Doesn’t - MachineLearningMastery.com

Musicians led by Jason Isbell sue AI song generator Suno, alleging it lets users mimic artists' names without consent - The Wall Street Journal

Airbnb for inference: startups want to rent your GPU. - Tom's Hardware

Tokens keep getting smarter, and prices keep dropping - CNBC

Whatever the singularity is, we’re in it right now. - The Atlantic

Self-sovereign agent swarms that make their own money, decide what to do with it. - Hyperdimensional

Self-evidently we cannot allow self-sovereign 'anything' that is not subject to democratic constraints, or democracy no longer exists, which is a (the) key problem with AI, crypto and Big Tech. Seriously what is wrong with these people?

Bitcoin is going to have to become quantum-resistant, it needs and has some necessary governance. Someone is creating and directing these swarms. Wasn’t there a whole Black Mirror about this? They deploy smart bee swarms to improve pollination, turn into a self-organizing assassination system?

Follow the latest AI headlines via druce.ai on Bluesky