Good morning. Carlos Alcaraz, my favorite player, just became the youngest man to win Grand Slams on all three surfaces, defeating world No. 1 Sinner at Roland Garros. At 22, he's already rewriting tennis history—and proving that when it comes to clay, there's something in the Spanish DNA that just gets it done.

— The Deep View Crew

In today’s newsletter:

• 💻 AI for Good: AI foot scanner could prevent heart failure hospitalizations

• 💤 A solution to AI hallucinations

• 🚂 AI giants are steamrolling the “wrapper” startups

💻 AI for Good: AI foot scanner could prevent heart failure hospitalizations

A wall-mounted AI device that monitors feet for fluid buildup could help prevent heart failure hospitalizations by providing a 13-day advance warning before symptoms become severe.

What happened: Researchers from Torbay and South Devon NHS Foundation Trust tested an AI-powered foot scanner on 26 patients across five NHS trusts. The device, developed by Cambridge-based startup Heartfelt Technologies, detects oedema — fluid retention in feet and ankles that signals worsening heart failure.

How it works: The smart speaker-sized device uses foot recognition technology similar to facial recognition, capturing 1,800 images per minute from multiple angles to calculate fluid volume in feet and lower legs.

• Alerts care teams when fluid levels exceed preset thresholds

• Works automatically without patient input, even through thin socks

• Can operate without Wi-Fi for elderly patients without internet access

• Scans only up to 50cm from the floor to protect privacy

The results: Among patients who used the scanner for at least two weeks, the device correctly predicted hospitalizations with alerts arriving 8-19 days before admission. At the study's end, 82% of participants chose to keep the device.

Dr. Philip Keeling, the study's senior author and consultant cardiologist at Torbay and South Devon NHS Foundation Trust, said the scanner acts "like a virtual nurse" amid nursing shortages. Keeling added that only half of heart failure patients currently receive early nurse reviews before hospitalization.

Why it matters: More than one million people in the UK live with heart failure. The device offers passive monitoring that requires no daily patient input, potentially reducing the burden on overstretched heart failure nursing teams.

The research was presented at the British Cardiovascular Society annual conference in Manchester.

Instantly Turn One Video Into 30 clips with Goldcast’s Content Lab

Your team is sitting on a goldmine of untapped video content. 

Goldcast’s Content Lab unlocks this treasure. This AI video repurposing solution does the heavy lifting for you, taking your long-form content and transforming it into a variety of snackable clips, social posts, blogs, and GTM collateral. 

Whether it’s a podcast, a webinar, or a video, Content Lab can easily break down one video into 30+ ready-to-publish assets for all your marketing channels. 

There’s a reason it’s trusted by marketing teams at OpenAI, HootSuite, Workday – not only does it make their content repurposing efforts a breeze, but it’s also free. Can’t beat that.

Try Content Lab free today and make your content work harder than you do.

💤 A solution to AI hallucinations

Large language models like ChatGPT, Gemini and Llama continue to generate false information at alarming rates. These mistakes, known as hallucinations, are built into how these systems function and happen without warning.

One high-profile case involved ChatGPT falsely accusing law professor Jonathan Turley of sexual misconduct, citing a fabricated Washington Post article about a trip to Alaska that never happened. OpenAI blocked future responses about Turley, but the core problem persists.

What's happening: LLMs work by predicting the next word in a sentence based on statistical patterns from training data. They don't understand meaning or verify facts — when they lack information, they fill gaps with plausible-sounding fabrications.

The problem is worsening as AI companies increasingly feed models with synthetic data generated by earlier AI systems. This creates what researchers call "model collapse," where errors compound across generations, leading to degraded performance and reliability.

The solution: Neurosymbolic AI is gaining attention as a potential fix. This approach combines machine learning with logic-based rules, allowing AI to extract patterns during training and then apply formal reasoning rather than just statistical guessing. Companies like Google's DeepMind are already using neurosymbolic methods in specialized applications like AlphaFold for protein structure prediction and AlphaGeometry for mathematical problems.

Why it matters: Hallucinations aren't just technical flaws — they erode trust and create real-world risks in medicine, law and finance. Recent studies show OpenAI's latest reasoning model o3 hallucinates 33% of the time, double the rate of its predecessor.

Neurosymbolic AI offers a path toward more reliable, explainable AI systems that could restore confidence in the technology.

Free event: Join Scott Galloway and Microsoft’s Chief Scientist, Jaime Teevan for a discussion on AI and the future of work

On June 12, Scott Galloway and Microsoft’s Chief Scientist, Jaime Teevan, are going to discuss the question on everyone’s minds: What does AI mean for the future of our careers?

They’ll address questions like:

• Are most of us going to lose our jobs to AI?

• How will AI challenge and change the nature of knowledge work? 

• What will the role of the human be in 10-20 years?

• How do we avoid cognitively atrophying when we’re not doing the work ourselves?

It’s free to attend, but registration is required. Join us on June 12 from 12 - 1pm ET for this insight-filled event.

• Ohio State sets new rules for students using AI tools

• Meta’s CTO says 2025 will be a big year for AR and VR

• Anthropic launches Claude models built for U.S. government use

• A film festival just debuted movies made with AI

• Trump deploys 2,000 National Guard members after Los Angeles immigration protests

• Meta in talks for Scale AI investment that could top $10 billion

• Netflix: Machine Learning Engineer - Promo Media & Marketing

• Broadcom Software: Senior AI Engineer

• Glean: Connects across your work apps to find the exact doc, message, or answer you need

• Andi: An AI browser that answers questions visually — like search, but clean and conversational

• Suno: Make any song you can imagine with just a prompt

🚂 AI giants are steamrolling the “wrapper” startups

Major AI platforms are aggressively launching native features that directly compete with and eliminate the need for third-party wrapper companies and integrations, creating an existential crisis for startups built on platform APIs. This consolidation represents a fundamental shift in AI ecosystem dynamics, where platforms are moving beyond providing foundation models to offering complete, integrated solutions that bypass the third-party developer ecosystem entirely.

OpenAI quietly built a meeting notetaker directly into ChatGPT. The new “Record Mode” announced on June 4th – an AI that can record, transcribe and summarize meetings in real time – ships as part of ChatGPT’s desktop app for paying team users—directly competing with established players like Otter.ai, Fireflies.ai, and Granola which just raised $43m at a $250m valuation.

Anthropic cut API access to Windsurf on June 3, 2025, with less than five days' notice. The move came as OpenAI announced plans to acquire Windsurf for $3 billion, with Anthropic's Chief Science Officer stating it would be "odd for us to be selling Claude to OpenAI." Users experienced immediate capacity issues, forcing many to migrate to competing platforms.

These examples reflect a broader pattern where AI platforms are timing feature launches strategically during high-profile events while simultaneously restricting access to potential competitors.

Zoom out:

• Google has integrated AI capabilities across its Workspace suite, launching NotebookLM Plus in January 2025 and expanding Gemini integration throughout its productivity tools.

• Microsoft's Copilot expansion reached $500 million in revenue by 2024, directly competing with AI coding startups. The company launched a centralized Agent Store with 70+ pre-built agents and multi-agent orchestration capabilities, providing native alternatives to workflow automation and business process tools previously handled by third-party providers.

• Meta launched standalone AI applications and generative ad tools that directly displaces third-party ad optimization and content creation startups which we covered here.

• Amazon committed $230 million to AI startup investments while developing competing native services—a classic "embrace, extend, extinguish" strategy that provides credits to startups while building internal alternatives.

This dynamic isn’t entirely new in tech. In fact, there’s even a slang term for it – “Sherlocking” – a term that describes how Apple would often unveil its own version of a popular third-party app, effectively killing the original. A classic example came when Apple launched AirTag in 2021, directly undercutting Tile, the startup that pioneered Bluetooth item trackers. Apple’s AirTag reproduced Tile’s core tracking functionality and baked it into the iPhone’s native “Find My” app with the added boost of ultra-wideband precision and Apple’s massive user network.

Amazon's private-label strategy provides another template—the company systematically used third-party seller data to develop competing products, forcing merchants to pay nearly 50% of revenues in fees while battling Amazon's own house brands.

Go deeper: OpenAI CEO Sam Altman made the platform strategy explicit during two different 20VC podcast interviews: "There are two strategies to build on AI right now. There's one strategy which is to assume the model is not going to get better and then you build all these little things on top of it. There's another strategy which is build assuming that OpenAI is going to stay on the same trajectory and the models are going to keep getting better at the same pace."

Altman's warning was direct: "When we just do our fundamental job because we have a mission, we're going to steamroll you. It's not personal; it's our mission."

Successful survival strategies have emerged among startups that focus on specialized vertical applications, proprietary data advantages, or multi-model approaches that reduce platform dependency. Companies building "thick wrappers" with substantial additional functionality show greater resilience than simple API integrators.

The plight of these “wrapper” startups exposes a hard truth about building on someone else’s technology: if your business is essentially a feature of a larger platform, it can vanish the minute the platform decides to subsume it. The fragility of wrapper startups has become painfully clear. Once those models got better or the big player moved in, the wrapper’s advantage evaporated. In hindsight, it seems naive that investors poured millions into companies whose secret sauce was often just prompt engineering or UI polish on top of someone else’s AI.

While the FTC investigates and the EU implements the Digital Markets Act, platforms are moving faster than regulators can comprehend. When Microsoft invests billions in OpenAI while building competing AI models, when Google funds startups through accelerator programs while developing native alternatives, and when Amazon provides startup credits while creating internal competitors, we're witnessing coordinated market manipulation that makes historical monopolization attempts look amateur. 

By the time meaningful restrictions arrive, the competitive landscape will be permanently altered.

🤔 Your thought process:

Selected Image 1 (Left):

• “The black spots were just too weird on the AI image.”

• “The black spots on the fake bananas just looked weird in a way I have never seen. Almost sandblasted. I did account that it could plausibly be a species of banana I'm not familiar with, but that seemed an unusual choice for an AI to make, given the much larger availability of standard bananas for a data set.”

Selected Image 2 (Right):

• “The last two days I'm 0 for 2! The bananas in the other picture seemed to be oddly distorted to me. I was sure they were fake. Maybe tomorrow I'm break my losing streak.”

• “Well dang. This one ruined my 6 win streak... I was looking at the other image, seemed there was a lot of uniformity/similarities in the shape and characteristics- it seemed too perfect. This image had variations in shape and size of the ends... guess i need to be more familiar with the fruit and agriculture world.... :D”

💭 A poll before you go

Thanks for reading today’s edition of The Deep View!

We’ll see you in the next one.

P.S. Enjoyed reading? Take The Deep View with you on the go! We’ve got exclusive, in-depth interviews for you on The Deep View: Conversations podcast every Tuesday morning. Subscribe here!

If you want to get in front of an audience of 450,000+ developers, business leaders and tech enthusiasts, get in touch with us here.