AI Reading for Monday March 30

Sora losers: OpenAI shut down Sora, a separate system that operated apart from its research team, and redirected its compute resources to other tasks. - The Wall Street Journal

Because it cost about $1 million per day to run and usage declined, no longer plans to fold video creation capability into ChatGPT.

OpenAI's ChatGPT App Store is not good, actually. - Bloomberg

Radio Shack clearance bin energy

Good time to start a company as things shake out a bit - interesting question is whether Anthropic will pass OpenAI in ARR this year - X (formerly Twitter)

Anthropic's March 2026 adaptive interview study of 80,508 people across 159 countries and 70 languages found most users sought time, relief and access ('give me my life back'), with ~19% seeking professional excellence and ~14% personal transformation. - TNW | Opinion

Apocaloptimism FTW: Documentary 'The AI Doc: Or How I Became an Apocaloptimist' premiered at Sundance and opened in theaters, features interviews with Sam Altman and Dario Amodei, and reports roughly 20,000 working on AGI versus fewer than 200 on alignment. - Business Insider

How the AI bubble bursts - Volpe’s Blog

iPhone 50-year anniversary edition, maybe…’AI Tim Cook’, nope. Apple executives offered contrasting forecasts while marking Apple’s 50th birthday: SVP Eddy Cue said consumers "may not need an iPhone 10 years from now," while marketing chief Greg Joswiak said iPhones will exist in 50 years and CEO Tim Cook ruled out AIs as Apple executives. - 9to5Mac

Nvidia's forward earnings multiple falls below S&P 500 for first time in 13 years - The Street

Dude, you’re getting a data center: Dell said fiscal revenue reached $113.5 billion and its AI infrastructure business grew to $25 billion, with Q4 AI-optimized server revenue up 342% to $9 billion and 12-month AI server orders totaling $64 billion (with a $43 billion backlog). - AOL

Major tech firms including Google, Amazon, Meta, Pinterest and Atlassian attribute recent mass layoffs to AI-driven productivity, while Amazon has cut about 30,000 corporate roles, Meta cut roughly 700. Makes for a better blog post and they need capital for data centers. - BBC

Companies allocate much of their AI budgets to technology and little to workforce transformation, creating process bottlenecks, adoption failures and risks, should probably reverse the ratio. - Yahoo

reprogramming silicon is cheap, humans expensive

Goldman Sachs CIO Marco Argenti said AI is improving at warp speed. - Bloomberg

Trying to make sense of Musk's pre-IPO pivot, to clean house and go for another all-in double-or-nothing bet on robots and AI - Gizmodo

Rep. Summer Lee proposed an "Artificial Intelligence Civil Rights Act" to regulate AI used in law enforcement and immigration, targeting automated decision systems like border control and surveillance amid evidence of bias and industry opposition. - Slate Magazine

more pressing than AI doom, although alignment, superpowers for bad actors with cheap drones, WMDs are a big deal. like it or not you are being surveilled constantly by your devices and AI makes it data-minable and interpretable at scale, we have to rethink what we accept as a presumably free society.

Not to mention Maven took out another school and girls’ volleyball team, which doesn’t seem to be costing anybody their job or rating a ‘concerning’ from Elon Musk. - New York Times

just keep scrolling, everybody

Iran war drives increase in cyber probes and attacks. The UAE faces 500,000–700,000 cyberattacks daily, with state-linked actors—including Iran—using AI tools for reconnaissance, phishing, malware and deepfakes, with defenses holding so far. - The Cyber Express

AI for ATC? he problem with this story is, LaGuardia already has automated runway incursion detection, had tons of alarms going off, after the incursion. - Bloomberg

never miss a good excuse to sell technology

What is OpenClaw? An open-source, self-hosted agentic AI that connects LLMs to software, executes actions and stores file-based memory, has over 250,000 GitHub stars and enterprise interest, while security firms warn of risks from default configurations and skill supply-chain abuse. - TechRadar

An AI "claw" is an autonomous agent, originating with OpenClaw, that decomposes goals across apps and tools, enables skill-based marketplaces (e.g., OpenClaw, Nvidia NemoClaw, Meta My Computer, Anthropic Claude Cowork), and raises security risks requiring sandboxes and permissions. - CNET

We’re all 10x engineers now: AI coding agents collapse development cycles from weeks to hours, let PMs code - VentureBeat

they can prototype OK, but their code will table scan terabyte tables instead of indexing them, use browser automation instead of proper APIs, lack guardrails, and they won’t be on call when it breaks.

Eric Schmidt says top engineers are shifting from line-by-line coding to directing AI-driven code generation, with many early-stage and Y Combinator-backed startups already producing most of their code using AI, accelerating product development for small teams. - Mashable India

Stanford professor Erik Brynjolfsson predicts AI will mostly complement human work, expanding software development up to tenfold and delivering 2×–10× productivity gains while creating new roles and governance needs rather than widespread job losses. - ZDNET

One definition of AGI is when you can rent robots for pretty much any job that needs a human, no place to hide and the human wage can't exceed the cost of the robot. We aren't there yet but the AI bros aren't hiding their intentions, that's where they think we are headed soon.

Research reported by Tom Claburn and a Kettle podcast discussion found that AI can generate sophisticated code but often produces errors and brittle solutions, so AI will augment developers—speeding prototyping but requiring human debugging and maintenance. - The Register

GitHub Copilot inserted an ad into a team member's pull request after being summoned to fix a typo, prompting concerns about promotional edits and trust risks. - Zach Manson

this PR sponsored by Raycast™

GNU world order: AI Agents Could Make Free Software Matter Again - George London

Can start a new wave of cloning everything proprietary or under a restrictive license.

Randomized trial of 120 Brazilian undergraduates found that after a 45-day delay (85 returned), students who used ChatGPT scored 57.5% on a 20-question retention test versus 68.5% for traditional learners, while spending 3.2 versus 5.8 study hours. - PsyPost