AI Leadership Weekly

Issue #32

Brad
May 22, 2025

Welcome to the latest AI Leadership Weekly, a curated digest of AI news and developments for business leaders.

Top Stories

Source: Amazon.com

You can’t trust Grok anymore
Grok, the chatbot developed by Elon Musk’s xAI, has had a very public and troubling meltdown. It started last week when it began parroting conspiracy theories regarding “white genocide” in South Africa (which is widely debunked), but then moved onto doubts about the Holocaust, and scepticism about Jeffrey Epstein’s cause of death.

Mysterious origins. Grok injected fringe theories into unrelated chats, prompting xAI to blame a “rogue employee” for tampering with its code. And while they haven’t named this “employee”, it’s worth remembering that Elon Musk himself (and close ally Peter Thiel) was born and raised in South Africa.
The bias is baked in. Reports have also stated that, after parroting its claims of white genocide, Grok would also say it was ordered to do so, and that the facts go against its programming of sticking to factual information. CNN points out that, when training these AI models, there’s often little effort to build in safety nets that prioritise truth over ideology.
The industry is moving too fast. Despite clear issues, many AI companies have already “moved on from the research and safety-testing phases,” according to a recent CNBC report. This all becomes extra alarming given pushes from Google and others for “AI search”.

The big picture: Grok’s conspiracy-laced responses are a flashing red warning for AI safety. As AI becomes more embedded in our digital lives, the risk of it being used—or misused—to manipulate public opinion grows. As AI researcher Gary Marcus bluntly put it: “Should we be worried? Hell, yeah.”

Nvidia moves further beyond just chip manufacture
Nvidia came out swinging at Computex Taipei this week, announcing a wave of AI-forward developments that signal its vision for the future—from humanoid robotics to ultra-flexible data centre infrastructure. The company’s latest moves show a clear focus on diversifying its hardware capabilities, expanding into global markets, and unlocking new paths for AI deployment in both enterprise and physical environments.

Humanoid AI gets a boost. Nvidia introduced Isaac GR00T-Dreams, a new robotics platform designed to generate massive datasets for training humanoid robots. CEO Jensen Huang called physical AI “the next trillion-dollar industry,” and said Nvidia is building the software foundation to make adaptable robots a real-world presence in factories—and eventually homes.
Build-your-own AI server. The company’s new NVLink Fusion platform allows enterprises to pair Nvidia’s Grace CPUs and GPUs with third-party AI chips or CPUs. Qualcomm is among the first adopters, a major win for both companies as they look to dominate next-gen AI infrastructure.
Global GPU clouds and new servers. Nvidia also unveiled RTX Pro Blackwell servers, aimed at shifting enterprise workloads from CPU-heavy setups to GPU-accelerated infrastructure. Plus, DGX Cloud Lepton lets customers access powerful AI compute via a global GPU network hosted by CoreWeave, Foxconn, and SoftBank.

The big picture: Nvidia’s barrage of announcements underscores its ambition to not only power AI but to reshape how it's built, trained, and deployed. By opening up its tech stack and pushing into robotics, Nvidia is solidifying its role as the backbone of the AI era—right as the world’s biggest players compete for control of the future of computing.

Microsoft is building the ‘agentic web’
At Build 2025, Microsoft rolled out more than 50 new AI tools aimed squarely at building what it calls the “agentic web” — a new phase of the internet where autonomous AI systems can reason, remember, and act with minimal human input. From developer tools and enterprise platforms to security frameworks and scientific R&D, Microsoft made it clear: the era of passive AI is over, and intelligent agents are here to get things done.

From assistants to agents. Microsoft is shifting focus from reactive AI to proactive agents that can reason, remember, and complete tasks independently. “We’ve entered the era of AI agents,” said Microsoft’s Frank Shaw, with new tools like structured RAG and multi-agent workflows enabling more complex behaviour.
GitHub Copilot becomes an autonomous teammate. GitHub’s Copilot now goes beyond code suggestions to autonomously fix bugs, refactor code, and collaborate across the development life cycle — and Microsoft is open-sourcing Copilot Chat in VS Code to involve the dev community.
Enterprise AI goes local and secure. With Windows AI Foundry and Foundry Local, developers can now run AI models directly on devices. Meanwhile, Entra Agent ID and Microsoft Purview help companies avoid “agent sprawl” and maintain tight governance over autonomous systems.

The big picture: Microsoft is betting that the future of computing lies in intelligent agents — not just smart tools, but AI co-workers that can operate across platforms, automate workflows, and accelerate everything from software development to scientific discovery. It’s a bold vision, and if Microsoft gets it right, it could shape the next 20 years of digital innovation.

In Brief

Market Trends

Robin (helps) make end-to-end scientific discovery
FutureHouse has just announced a milestone in automated science: their multi-agent system, Robin, made what they describe as a complete end-to-end scientific discovery with minimal human guidance. In just 2.5 months, Robin autonomously proposed, analysed, and validated a new therapeutic use for ripasudil—a glaucoma drug—as a potential treatment for dry age-related macular degeneration (dAMD), a leading cause of blindness.

A fully autonomous research workflow. Robin integrates FutureHouse’s existing agents—Crow (literature review), Falcon (experiment design), and Finch (data analysis)—to handle the entire scientific process, from hypothesis to publication. Humans carried out the lab work, but the research logic was 100% AI-driven.
Drug repurposing with real-world promise. Robin identified that ripasudil, a ROCK inhibitor, enhances retinal pigment epithelium (RPE) cell function—offering a new therapeutic pathway for dAMD. Notably, the drug is already clinically used in the eye, meaning real-world applications could be on the near horizon.
Speed and scale redefine scientific timelines. The project—from system development to a submitted paper—took just 10 weeks and was completed by a small team. Robin’s modular design makes it adaptable beyond biotech, with potential for discoveries in climate tech, materials science, and more.

The big picture: Robin isn’t just another AI assistant—it’s an entirely new model for how science can be done. By connecting autonomous agents into a coherent workflow, FutureHouse may have cracked the code for scalable, interpretable AI-driven research. With Robin going open-source on May 27, the implications for democratised scientific discovery are huge.

Intel releases affordable AI dev cards
At Computex 2025, Intel introduced its new Arc Pro B-series GPUs, aiming directly at AI professionals and power users who need serious inference horsepower without blowing their budget. The $299 Arc Pro B50 and its beefier sibling, the B60, pack impressive VRAM, AI capabilities, and a clear focus on scalable workstation performance. Oh, and there’s even a full-blown multi-GPU system called “Project Battlematrix.”

Affordable AI-ready GPUs with massive VRAM. The Arc Pro B50 comes in at $299 with a whopping 16GB of VRAM—more than double what’s typical at that price. The B60 bumps things up to 24GB and 197 TOPS of compute, making it ideal for AI model inference and large-context tasks.
Project Battlematrix supercharges inference. Designed around the B60, Intel’s pre-built Project Battlematrix workstations start at $5,000 and scale up to eight GPUs, 192GB of VRAM, and support for 70B+ parameter models. One demo even ran the 675B Deepseek model on a single box.
Software stack and container rollout coming soon. Intel is launching a full-stack containerized Linux solution this fall, with features like SRIOV, VDI, and AI-ready software all built in. It’s all about fast deployment and out-of-the-box performance.

The big picture: Intel is gunning hard for a piece of the pro AI hardware market—and doing it by undercutting rivals on price while delivering workstation-grade power. With massive VRAM and a growing ecosystem of software and hardware partners, Intel’s Arc Pro series could become a go-to for AI developers and researchers looking for performance without the enterprise price tag.

OpenAI and Jony Ive announce an announcement
Best friends Sam Altman and Jony Ive (of Apple fame) have appeared together in an almost-10-minute video to announce the merger of OpenAI with Ive's 'io'.

Apparently they are partnering to make some sort of AI-powered hardware product, which Sam has been play-testing for a while now. This is all we know so far, so you better stay tuned! You may also enjoy their subreddit.

Tools and Resources

Fluig
Turn your documents and ideas into beautiful diagrams with AI.

Pipedream
Helps connect your AI assistant to over 2,500 APIs.

Short Ads
Upload a product image then use templates to turn it into a short ad.