Tech-Reader AI Digest for Wed May 6 2026

-

 

Tech-Reader AI Digest

Wednesday, May 6, 2026

#AI #TechNews #Digest



Story 1: Mira Murati Testifies — The Candidness Gap, "Creating Chaos," and the 2023 Board Crisis Explained

What happened: Former OpenAI CTO Mira Murati delivered video testimony Wednesday in Elon Musk's lawsuit. Murati — who was briefly CEO of OpenAI after its board temporarily forced out Altman in 2023 — testified that CEO Sam Altman sowed distrust among top executives. "My concern was about Sam saying one thing to one person and completely the opposite to another person," she said. She said Altman was "creating chaos" and, at times, was deceptive with her and others.

Murati provided a specific example: she testified that Altman misled her regarding safety protocols for a new model, claiming the legal team had cleared it when they had not. She described Altman pitting executives against one another and undermining her role as technology chief.

Legal analysts are calling this the Candidness Gap — the bridge between the 2023 board crisis and today's testimony. When OpenAI's board fired Altman in November 2023, it cited that he was "not consistently candid" with leadership. Murati's testimony confirms that characterization wasn't an ivory-tower safety panic or a boardroom power struggle. It was a response to actual, documented executive deception — specifically about safety protocol clearances on an unreleased model.

Also Wednesday: Shivon Zilis wrapped her testimony. The February 2018 text to Musk — "Do you prefer I stay close and friendly to OpenAI to keep info flowing?" — remains in the jury's possession. (Source: Reuters / US News / NBC Bay Area / Bloomberg)

Why it matters: Murati's testimony is the trial's most important third-party voice. She has no stake in Musk winning. She was fired alongside Altman, returned when he was reinstated, then left voluntarily. Her description of Altman's leadership — chaos, deception, contradictory messaging to different executives — is the same description the board gave when it fired him in 2023. The jury now has independent executive corroboration of the board's original judgment.

Aaron's take — The November 2023 board crisis always looked like a failed coup that ended with Altman reinstated and the rebels defeated. Murati's testimony reframes it entirely. The board fired Altman because they experienced exactly what Murati is now describing under oath. That's not a Musk talking point. That's a former CTO who ran the company for five days telling a federal jury what she saw from the inside.

Story 2: Anthropic Signs SpaceX Colossus 1 — 220,000 GPUs, Orbital Compute, and Exascale Diplomacy

What happened: Anthropic signed an agreement with SpaceX to use all the compute capacity at the Colossus 1 data center in Memphis, Tennessee — announced at Anthropic's "Code with Claude" developer conference in San Francisco. This gives Anthropic access to more than 300 megawatts of new capacity and over 220,000 NVIDIA GPUs — including H200s and the latest Blackwell GB200s — coming online within the month.

The immediate product impact: Claude Code's five-hour rate limits are doubled for Pro, Max, Team, and Enterprise plans effective Wednesday. Peak-hour throttling for Pro and Max removed entirely. Claude Opus API volumes raised significantly.

The irony is complete: Musk merged xAI with SpaceX earlier this year. In February, he called Anthropic "misanthropic" on X. Wednesday he posted: "No one set off my evil detector." He said he spent time with senior Anthropic team members last week and was "impressed — everyone I met was highly competent and cared a great deal about doing the right thing." In a separate post, Musk confirmed xAI will be dissolved as a separate company and rebranded as SpaceXAI.

Analysts are calling the shift Exascale Diplomacy — Musk prioritizing the SpaceX IPO, which is targeting a $1.75-2 trillion valuation, over his personal rivalry with Dario Amodei. Anthropic as a named compute customer strengthens SpaceX's pitch as an AI infrastructure company, not just a launch and Starlink business.

The two companies are also exploring orbital AI compute capacity — space-based data centers powered by near-constant solar energy with passive vacuum cooling. The technical key is Starship's payload capacity: the mass-to-orbit economics of Starship are the only thing that makes heavy GPU infrastructure viable in space. SpaceX is the only organization with the launch cadence and constellation operations experience to make this a near-term engineering program rather than a research concept.

Anthropic's compute stack now spans: Amazon (up to 5GW), Google/Broadcom (5GW coming 2027), Microsoft/NVIDIA ($30B Azure capacity), Fluidstack ($50B US infrastructure), and now SpaceX Colossus 1 (300MW, 220,000 GPUs). (Source: Anthropic blog / CNBC / Semafor / CoinDesk / WCCFTech)

Why it matters: The company declared a "supply chain risk" by the Pentagon in March is now renting compute from the company that owns the Pentagon's most classified AI infrastructure — while the man who called it "misanthropic" describes its team as impressive. Exascale Diplomacy is the right framing: business logic overrode personal rivalry at the exact moment SpaceX needs a marquee AI customer for its IPO.

Aaron's take — "No one set off my evil detector." That's Elon Musk — who sued OpenAI, called Anthropic misanthropic, and is currently in a federal courthouse — describing his week with Anthropic's team. The orbital compute ambition is the long game. Musk doesn't rent Colossus to Anthropic for 300 megawatts. He rents it because Anthropic is the credibility that makes SpaceXAI's IPO story land.

Story 3: Microsoft Copilot Cowork Goes Mobile — Skills, Plugins, and the Shift From Assistant to Operator

What happened: Microsoft is expanding Copilot Cowork with mobile support, reusable Skills, and deeper integrations. Cowork now runs on iOS and Android so users can delegate work from their phone — on a commute, between meetings, or away from a desk — and come back to a finished outcome. The cloud runs the work. The laptop doesn't need to stay open.

Skills are a reusable set of instructions that guide Cowork on how to complete a task or workflow. Instead of improvising process every time, teams encode how work should be done — structure, tone, approval path — and Cowork executes consistently. Over time Skills function as a shared automation layer across teams, reducing repetitive work and standardizing process at the organizational level.

The Copilot Cowork / Anthropic connection is explicit: Copilot Cowork is designed in close collaboration with Anthropic, and Claude Opus 4.7 is now available as a selectable model inside Cowork alongside the default model routing. New third-party connectors include HubSpot, LSEG, Moody's, Notion, Miro, monday.com, and S&P Global Energy. Mobile beta available now for Frontier participants. General availability targeted for July 2026. (Source: Microsoft 365 Blog / Neowin / Digital Trends / Thurrott)

Why it matters: Copilot Cowork going mobile completes the task delegation loop. You think of work on your phone, delegate it, and come back to a finished outcome. The Skills framework is what makes it enterprise-grade — instead of every employee improvising process, organizations encode how work should be done and the AI executes consistently. Claude Opus 4.7 sitting inside Cowork as a selectable model means Anthropic is embedded at the model level inside Microsoft's enterprise AI product.

Aaron's take — Microsoft designed Cowork in close collaboration with Anthropic. Claude Opus 4.7 is selectable inside it. Anthropic just signed a 220,000 GPU deal with SpaceX. Musk called Anthropic's team impressive. The alliances, the rivalries, the infrastructure — all of it restructured on a single Wednesday in May.

Quick Hits — The Rest of Today's AI World

Anthropic / Claude

  • Beyond Story 2 — Claude Code five-hour rate limits doubled effective today. Peak-hour throttling removed for Pro and Max. Anthropic's "Code with Claude" developer conference underway in San Francisco. (Source: Anthropic blog)

Gemini (Google)

  • No new announcements today. (Source: Google)

VS Code / GitHub Copilot

  • No new announcements. Token-based billing June 1 transition remains standing news. (Source: GitHub)

Replit

  • No new announcements. (Source: Replit)

Perplexity

  • No new announcements today. (Source: Perplexity)

Microsoft Copilot

  • Copilot Cowork mobile launch — see Story 3. Claude Opus 4.7 now selectable inside Cowork. July 2026 general availability target. (Source: Microsoft 365 Blog)

xAI / Grok

  • xAI dissolving as separate company — rebranding as SpaceXAI per Musk's X post today. Colossus 1 deal with Anthropic announced simultaneously — see Story 2. Trial continues. (Source: CNBC / Musk X post)

OpenAI

  • Trial Day 7 continues. Murati and Zilis testimony — see Story 1. No new product announcements today. (Source: Reuters / NBC Bay Area)

Palantir

  • No new announcements today. Q1 2026 blowout earnings remain standing news — 85% growth, $1.5M revenue per employee, net income quadrupled. (Source: CNBC)

Reflection AI

  • No new announcements today. (Source: Breaking Defense)

Ollama

  • No new announcements today. (Source: Ollama)

DeepSeek

  • V4-Pro and V4-Flash live since April 24. No new announcements today. (Source: DeepSeek)

Alibaba / Qwen / Z.ai

  • No new announcements today.

Inflection Pi / Mistral

  • No major news today.

That's your AI world for Wednesday, May 6. Back tomorrow. — Aaron

Aaron Rose is a software engineer and technology writer at tech-reader.blog. 

Catch up on the latest explainer videos, podcasts, and industry discussions below.


Popular posts from this blog

Insight: The Great Minimal OS Showdown—DietPi vs Raspberry Pi OS Lite

-
Image
Insight: The Great Minimal OS Showdown—DietPi vs Raspberry Pi OS Lite When you're building a headless server, IoT device, or lightweight project box, the last thing you want is bloatware eating your precious resources. Enter the world of minimal operating systems—where every megabyte matters and efficiency reigns supreme. Two contenders dominate this space: DietPi and Raspberry Pi OS Lite. Both promise lean, mean computing machines that boot straight to the command line. But scratch beneath the surface, and you'll find they take fundamentally different approaches to the "less is more" philosophy. The Minimalist's Dilemma Picture this: You've got a Raspberry Pi 3B+ sitting on your desk, destined to become a home media server. Do you go with the familiar comfort of Raspberry Pi OS Lite, or venture into DietPi's optimized territory? The choice isn't just about personal preference—it's about understanding what "minimal" means to each operatin...
Read more

Running AI Models on Raspberry Pi 5 (8GB RAM): What Works and What Doesn't

-
Image
  Running AI Models on Raspberry Pi 5 (8GB RAM): What Works and What Doesn't The Raspberry Pi 5, with its upgraded processing power and 8GB RAM option, brings new possibilities for running AI models at the edge. While it remains a low-power alternative to high-end GPUs, the improvements over previous models make it an intriguing choice for machine learning projects. However, not all AI models run smoothly on the Pi 5, and understanding its strengths and limitations is key to making the right deployment decisions. The Raspberry Pi 5 as an AI Workhorse? At first glance, the Raspberry Pi 5's quad-core Cortex-A76 processor, clocked at 2.4 GHz, seems like a step toward AI workloads. Paired with 8GB of LPDDR4X RAM, it offers a substantial improvement in memory bandwidth and processing efficiency compared to its predecessors. But raw specs only tell part of the story. AI models demand specialized hardware acceleration, and while the Pi 5 can handle some tasks, complex deep learning mo...
Read more

Raspberry Pi Connect vs. RealVNC: A Comprehensive Comparison

-
Image
Raspberry Pi Connect vs. RealVNC:  A  Comprehensive Comparison The Raspberry Pi has revolutionized the way we interact with computers, providing a low-cost, highly adaptable platform for countless applications. One critical aspect of using a Raspberry Pi, especially for remote management and control, is the ability to access it from another device. Historically, RealVNC has been the go-to solution for this purpose. However, the recent introduction of Raspberry Pi Connect offers an integrated alternative. This article will compare Raspberry Pi Connect and RealVNC, exploring their features, requirements, and practical applications. Introduction to Raspberry Pi Connect Raspberry Pi Connect is a new VNC service directly integrated into Raspberry Pi OS 12. Designed for simplicity and ease of use, Connect aims to provide a seamless remote desktop experience for Raspberry Pi users. Currently in beta, Raspberry Pi Connect is free to use and requires a Raspberry Pi 4, 400, or 5 running...
Read more