A new METR study found experienced developers completed tasks 20% SLOWER when using AI coding tools. Not faster. The research challenges one of the most repeated assumptions in the industry right now. theaivaults.substack.com/p/washington...
Will you let AI handle your todo list?
Microsoft built a to-do list that runs itself. Tell Copilot Tasks what you need once, set a schedule, and it handles the multi-step execution in the background - even when your computer is off. Early access is free. theaivaults.substack.com/p/washington...
... and an MSR study on developer productivity that literally fell apart because the coders refused to participate without their AI tools.
This week's episode is live. Give it a listen! open.substack.com/pub/theaivau...
[3/3]
This week's AI Vaults podcast also covers the $650 billion data center buildout that is creating bipartisan backlash in local communities, a Memphis power grid that xAI built in secret until the EPA stepped in... [2/3]
The Trump administration banned Anthropic from all federal contracts. Hours later, OpenAI signed a Pentagon deal with the exact same ethical limits Anthropic drew the line over... [1/3]
Perplexity just launched "Computer" - Up to 19 AI models to execute your goals. Claude handles reasoning, Gemini handles research, Grok handles speed. Read about the update alongside the weekend-breaking news about Anthropic, OpenAI, and the US Gov't.
theaivaults.substack.com/p/washington...
tldr: This is how you get Skynet.
I wrote about what this week actually means -- for AI, for the military, and for the question of whether the machines we're building understand what death is.
They don't. That's the problem.
open.substack.com/pub/theaivau...
#AI #NationalSecurity #Anthropic
Hours after Anthropic was banned, OpenAI signed a Pentagon deal.
With the same restrictions, Anthropic was blacklisted for demanding.
Sam Altman knew exactly what he was doing.
AI models were given nuclear decision authority in 21 war game simulations.
They chose nuclear escalation 95% of the time.
Not once did any model choose to de-escalate.
NOT. ONCE.
AI models were given nuclear decision authority in 21 war game simulations.
They chose nuclear escalation 95% of the time.
Not once did any model choose to de-escalate.
NOT. ONCE.
Disney, Paramount, Netflix, Sony, Warner Bros., and the MPA all sent cease-and-desist letters to ByteDance after Seedance 2.0 let users generate realistic video of named celebrities in seconds. The AI copyright war just moved to video. open.substack.com/pub/theaivau...
Accenture just told 550,000 employees that AI adoption is now tied to their promotions. At the exact same time, the European Parliament banned the same tools from official devices entirely.
Listen here: open.substack.com/pub/theaivau...
Corporations are making AI use a condition of promotion. The European Parliament just banned AI on official government devices. The rules for how AI gets used in institutions are being written right now, and nobody is agreeing. open.substack.com/pub/theaivau...
The Pentagon gave Anthropic a Friday deadline: remove your AI safeguards or face the Defense Production Act. Anthropic said no. Claude is currently the only AI model approved for the military's most sensitive work. open.substack.com/pub/theaivau...
Scary:
Accenture is now tracking weekly AI tool logins for senior staff. Regular adoption is a formal requirement for leadership promotion this cycle. KPMG and Meta announced the same thing the same week. Check out the story in my newsletter- open.substack.com/pub/theaivau...
Waymo just opened driverless robotaxi service in Dallas, Houston, San Antonio, and Orlando. That is 10 U.S. cities running fully autonomous paid rides. Tesla autonomous ride-hailing city count: zero. open.substack.com/pub/theaivau...
What if AI's success is what crashes the economy? A viral analyst thesis explains the "Ghost GDP" scenario: AI raises corporate profits while eliminating the workers who spend money. If there isn't anyone to trickle the money down to, what happens to the world? open.substack.com/pub/theaivau...
Can we get Barry to bring back the Google Glass? I think it'll catch on this time.
🇨🇳 China is leading the open-source AI revolution! Alibaba's Qwen models are surpassing Meta's Llama in downloads. Cost-efficiency is driving global adoption. #OpenSourceAI #ChinaAI #TechNews open.substack.com/pub/theaivau...
🔧 Tool of the Week: Claude Cowork! Automate your workday with Anthropic's new desktop app. Seamless AI integration with legacy systems. #AICoWork #Automation #Productivity open.substack.com/pub/theaivau...
Google's NotebookLM reacts to the biggest stories in the industry: OpenAI's speed-first model, Anthropic's desktop agent, China winning the open source race, & a serious warning that the whole industry might be one disaster away from a trust collapse.
Listen here: open.substack.com/pub/theaivau...
🚀 OpenAI just dropped GPT-5.3-Codex-Spark! 15x faster coding, real-time assistance, and powered by Cerebras. The future of dev is here! #AI #Coding #OpenAI open.substack.com/pub/theaivau...
It is also National Drink Wine Day 🍷
Exactly what it sounds like. No further explanation required. [http://daysoftheyear.com], [http://holidaycalendar.io]
Today is National Pluto Day 🪐
Anniversary of Pluto’s discovery on Feb 18, 1930. Still discovered, still orbiting, still demoted. [daysoftheyear.com]
Anthropic's Claude Sonnet 4.6 delivers flagship AI performance at 1/5th the cost! 1M token context, massive computer-use improvements. Game-changer for enterprise AI. #ClaudeAI #Anthropic #EnterpriseAI open.substack.com/pub/theaivau...
Kimi K2.5: 1T param multimodal swarm at $0.60/M tokens. Self‑host + crushes Opus pricing.
Tool spotlight: theaivaults.substack.com/p/big-money-... #AITools
Cosmic pickleball?
I've never played, guess this is a great way to learn!
