In Today's AI News:
- Washington’s AI Bill Blender
- Powering AI: Space Solar + 100‑Hour Batteries
- EU vs Google: Android AI Interoperability
- Agents Need Badges: Agentic IAM + Manus Drama
- OpenAI’s Week: AWS Terms, Sora Sunset, Phone Rumors
- YouTube’s New “Ask” Button
- Reinforcement Learning’s Billion‑Dollar Comeback
- Schools vs Chatbots: Exams + “AI‑proof” Majors
- Ubuntu Adds AI (While Swearing It Isn’t Becoming “An AI Product”)
Today’s headlines are basically one long argument that AI isn’t a product category — it’s a fight over defaults, power, and who gets to press the big red ‘deploy’ button. Congress is blending bills, Europe is prying Android open, and the compute crowd is shopping for electricity like it’s the last carton of milk. Meanwhile, students are trying to major in ‘still useful’ and Linux is trying to add agents without accidentally summoning Clippy.
Washington’s AI Bill Blender
Reps. Ted Lieu and Jay Obernolte are back with a big bipartisan package aimed at standards, research, federal adoption, workforce, and a crackdown on deepfake abuse — because nothing says “future-proof” like stapling 20+ proposals together.
Reps Lieu and Obernolte Introduce Bipartisan Bill to Advance American Leadership in AI — Congressman Ted Lieu
A press release for the American Leadership in AI Act, pitching a single package that bundles standards, R&D, governance, workforce support, and AI-enabled crime provisions.
Lieu and Obernolte introduce consolidated AI bill package — Nextgov/FCW
Nextgov’s rundown of the consolidated bill: a broad, bipartisan push to move multiple AI policy goals forward under one roof (because committees love shared custody).
AI bill would crack down on deepfake distribution and protect whistleblowers — CNBC
CNBC highlights deepfake penalties, whistleblower protections, and standards-setting — a “least-controversial” opening bid that still tries to touch a lot of sharp objects.
Singularity Soup Take: Policy is finally drifting from “AI is spooky” to “AI is infrastructure,” which means the real action is going to be standards, procurement, and enforcement — i.e., the unsexy stuff that actually sticks.
Powering AI: Space Solar + 100‑Hour Batteries
Meta is pitching two ways to keep the lights on for data centers: beaming energy to solar farms after dark and banking power for days. It’s the same old story: models are hungry, and the grid is suddenly the main character.
Powering AI, Strengthening the Grid: Innovation in Space Solar Energy and Long-Duration Storage — Meta Newsroom
Meta announces partnerships for up to 1 GW of space-solar capacity and up to 1 GW/100 GWh of ultra-long-duration storage, with demonstrations targeted for 2028.
Meta inks deal for solar power at night, beamed from space — TechCrunch
TechCrunch digs into Overview Energy’s plan to beam near-infrared energy from satellites to existing solar farms — bold, early-stage, and very on-brand for the “compute needs electricity” era.
Singularity Soup Take: If you want to understand AI competition, stop staring at benchmarks and start reading grid plans — the bottleneck is increasingly measured in megawatts, not tokens.
EU vs Google: Android AI Interoperability
Commission seeks feedback on measures to ensure interoperability with Google's Android under the Digital Markets Act — European Commission (DMA)
The Commission lays out draft measures to open Android’s key capabilities so third-party AI assistants can do system-level tasks (wake words included), with feedback open until May 13.
EU tells Google to open up AI on Android; Google says that's “unwarranted intervention” — Ars Technica
Ars explains the practical fight: Gemini’s system privileges vs. rival assistants’ access to APIs, context, and device permissions — plus a timeline that could force changes by late July.
Singularity Soup Take: “Default assistant” is becoming a regulated surface — and once that door opens, every phone turns into a competitive battleground for agents that can actually push buttons.
Agents Need Badges: Agentic IAM + Manus Drama
JumpCloud Launches Agentic IAM to Govern the AI Lifecycle — JumpCloud
A pitch for treating agents like employees: discover them, register them, scope credentials, and audit actions — because “it was the bot” is not a governance strategy.
China kills Meta’s acquisition of Manus as US-China AI rivalry deepens — Ars Technica
Ars reports China blocking Meta’s acquisition of Manus, spotlighting how “agent wrappers” can become geopolitical hot potatoes when ownership, compute, and deployment cross borders.
Singularity Soup Take: The agent era isn’t just “cool demos” — it’s identity, audit, and jurisdiction. If your bot can click “send,” someone will ask who authorized the finger.
OpenAI’s Week: AWS Terms, Sora Sunset, Phone Rumors
OpenAI ends Microsoft legal peril over its $50B Amazon deal — TechCrunch
TechCrunch details a renegotiated Microsoft–OpenAI deal that creates a timeline for IP rights and clears a path for OpenAI products to show up across cloud providers.
What to know about the Sora discontinuation — OpenAI Help Center
OpenAI says Sora’s web/app experiences ended April 26, with the API set to discontinue Sept. 24 — and encourages users to export their content (before the great delete comes).
OpenAI's Rumored Phone Would Replace Apps With AI Agents — CNET
CNET rounds up a hardware rumor (via analyst chatter) about a phone concept built around on-device agents — the most ambitious way to replace apps since “this time the AI assistant will totally work.”
Singularity Soup Take: The big labs are converging on the same endgame: own the runtime (cloud), the interface (assistants), and the distribution (devices) — and then call it “user empowerment” with a straight face.
YouTube’s New “Ask” Button
YouTube is testing an AI search mode that "feels more like a conversation" — Engadget
Engadget says YouTube’s experimental “Ask YouTube” feature gives Premium users conversational, summary-style search results — plus the occasional classic AI hallucination, as a treat.
Reinforcement Learning’s Billion‑Dollar Comeback
The Man Behind AlphaGo Thinks AI Is Taking the Wrong Path — WIRED
WIRED profiles David Silver’s new company, Ineffable Intelligence, and his argument that “superlearners” built via reinforcement learning (not just bigger LLMs) are the route to more general intelligence.
Singularity Soup Take: When a serious RL person says “LLMs won’t get you there,” investors don’t argue — they just hand over a very large check and hope the simulation doesn’t bite.
Schools vs Chatbots: Exams + “AI‑proof” Majors
College students are changing course in search of ‘AI-proof’ majors. But no one knows what they are — Associated Press (via Journal-Advocate)
An AP look at students switching majors amid fears that entry-level work will be automated — and the dawning realization that “AI-proof” may just mean “good at people, judgment, and ambiguity.”
Artificial intelligence malpractice and assessment - advice note — Ofqual (GOV.UK)
Ofqual publishes guidance for awarding organisations on how existing rules apply to AI-related malpractice risks in assessments — a polite way of saying “yes, we noticed the chatbots too.”
Singularity Soup Take: Education is being forced to update its threat model in real time — and the job market anxiety loop is now a curriculum design input, whether anyone admits it or not.
Ubuntu Adds AI (While Swearing It Isn't Becoming “An AI Product”)
Canonical lays out a plan for AI in Ubuntu Linux — The Verge
The Verge reports Canonical’s plan to roll out AI features through 2026 (accessibility, troubleshooting, automation), while emphasizing transparency and local inference over “cloud magic.”
Relevant Resources
Agentic AI — If today’s theme was “agents everywhere,” this is the map.
AI Hardware & Infrastructure — Power, cooling, grids, and the bits that make the magic expensive.
AI Tools & Applications — Where assistant features like “Ask YouTube” fit in the broader tool zoo.
Today's Pulse: 13 stories tracked across 15 sources — Congressman Ted Lieu, Nextgov/FCW, CNBC, Meta Newsroom, TechCrunch, European Commission (DMA), Ars Technica, JumpCloud, OpenAI Help Center, CNET, Engadget, WIRED, Associated Press, GOV.UK, The Verge