⚙️ Zuckerberg wants your data

Author

The Deep View
May 08, 2025

Good morning. I just came across a Stanford study that delves into the risks of AI companion apps for users under 18. It's a timely find, and we’ll explore similar implications later in the newsletter.

— The Deep View Crew

In today’s newsletter:

🩺 AI for Good: Real-time signing avatars

Source: ChatGPT 4o Image Generation

Around 97% of the world’s population don’t know how to speak sign language, which means that deaf people can comfortably communicate with only the remaining 3% of the world. While cochlear implants, introduced in the 1960s, have helped deaf people process sound, they’re often ineffective in noisy environments like train stations or classrooms.

British startup Silence Speaks is taking a different approach to make communication more accessible for the deaf and hard of hearing. They use AI sign language avatars that can translate spoken word in real time into sign language. Founder Pavan Madduru, formerly an AI solutions architect at Vodafone, was inspired after traveling with a deaf friend and witnessing firsthand the communication barriers in public spaces like train stations and classrooms.

  • The avatars – customizable from realistic humans to superheroes – offer a solution for a language where interpreters are scarce, especially in the U.K., where only 150,000 people use British Sign Language (BSL).

  • Over the next year, they plan to bring real-time on-device translation while expanding to other sign languages like American Sign Language (ASL).

Silence Speaks is built by and for the deaf community; deaf engineers and sign language users are deeply involved in the development process as the company wants to ensure that its solutions reflect real-world needs and cultural authenticity.

Today, Silence Speaks is working with Transport for London to integrate its AI sign language avatars into train announcements, helping make travel across England more inclusive. Passengers can scan QR codes for train announcements in sign language. Real-time spoken word translation remains a longer-term goal with the team expecting it to be achievable within the next year.

Why it matters: If successful, Silence Speaks would change how over 400 million people globally communicate and navigate the world around them.

🚀 Faster mobile app releases with 100% parallel automated QA

Manual testing on personal devices is too slow and too limited. It forces teams to cut releases a week early just to test before submitting them to app stores. And without broad device coverage, issues slip through.

QA Wolf gets mobile apps to 80% automated test coverage in weeks. They create and maintain your test suite in Appium (no vendor lock-in) — and provide unlimited, 100% parallel test runs with zero flakes.

✅ QA cycles reduced from hours to less than 15 minutes

✅ Tests run on real iOS devices and Android emulators

✅ Flake-free runs, no false positives

✅ Human-verified bug reports

No more manual E2E testing. No more slow QA cycles. No more bugs reaching production.

Rated 4.8/5 ⭐ on G2

📊 Mistral’s enterprise foundation model

Source: Mistral

French AI startup Mistral AI has introduced Medium 3, a new enterprise-focused AI model that it claims delivers near-frontier performance at a fraction of competitors’ costs. The model is offered via API at $0.40 per million input tokens and $2 per million output tokens, an 8x cost reduction compared to models like Anthropic’s  Claude Sonnet 3.7, which achieving “at or above 90%” of its performance across standard benchmarks, even outperforming newer open models like Meta’s Llama 4 “Maverick”.

The details:

  • Designed for enterprise applications, Medium 3 excels in coding and STEM tasks and supports multimodal inputs.

  • Mistral also launched Le Chat Enterprise, an AI assistant powered by Medium 3, tailored for enterprise-scale productivity and privacy.

  • Medium 3 can be self-hosted on private or public cloud infrastructure with as few as four high-end GPUs, lowering the barrier for private deployments. 

  • Integrates with Google Drive and SharePoint, with features like document libraries and custom agent creation

Unlike Mistral’s earlier open-source releases, Medium 3 is proprietary (not Open-source, yet) and independent experts haven’t yet vetted the impressive claims. Key details like the model’s size and training data remain undisclosed. 

Notably, Mistral hinted that a much larger model, which it intends to “open” up, is coming in the next few weeks. The bold price cut – by a startup that’s raised over $1.2B – could pressure its foundation model competitors to match suit.

Maximize AI For Your Small Business

As a small business owner (or employee), you know the value of finding every little edge or advantage when it comes to getting things done – and AI just might be the greatest hack of all. But with the constant influx of new news, information, and tools, it can get overwhelming… which is why Salesforce has pulled together this free guide to help you out.

In the guide, you’ll find everything you need to make the most of AI for your small business. Whether you’re looking for a leg up with the best strategies or simply want to see how other small businesses are putting AI to work - this guide has you covered.

⚡️Meta’s monetization of AI companionship

Source: ChatGPT 4o Image Generation

Mark Zuckerberg envisions a future where AI chatbots become your friends – a proposal he touts as a cure for the “loneliness epidemic.” In a recent interview, the Meta CEO pointed out that “the average American... has fewer than three friends” and that “the average person has demand for meaningfully more… like 15 friends or something”.

Zuckerberg’s grand vision of AI companionship comes on the heels of Meta’s pivot from its last big social experiment, the “metaverse”, where in 2022, Facebook became Meta Platforms, betting that virtual worlds would become the next frontier of interaction. Meta staked billions on VR, and bought Oculus for $2 billion back in 2014. Yet Reality Labs has now racked up $60 billion in losses, and Zuckerberg’s January promise that 2025 would be the metaverse’s “pivotal year” vanished from Meta’s latest call, replaced by five AI talking points.

Zuckerberg’s recent roadshow, where he appeared on various podcasts, developer conferences, and interviews with other tech leaders, made his position clear: most people crave more connection than they currently have, and they’re not getting it from other people. “I think people are going to want a system that knows them well – that understands them the way their feed algorithms do,” - Mark Zuckerberg, during an onstage interview with Stripe co-founder John Collison at Stripe’s annual conference. That system now exists: Meta AI, launched May 1, 2025, on iOS, Android, and the web, powered by Meta’s Llama 3 model.

What’s next: Meta has been aggressively pushing out new AI features across Instagram, Facebook, and WhatsApp. 

  • They rolled out a new Meta AI Assistant for Facebook Messenger and WhatsApp, “an assistant that gets to know your preferences”. Meta AI positions itself as a direct ChatGPT competitor.  At first glance, it offers what you’d expect from an AI assistant: you can type or talk to it, ask questions, generate images in real-time, and even get web results. But Meta’s twist is to make the experience social. 

  • The app includes a “Discover” feed that displays conversations other users have had with the chatbot – posts that friends opt to share, turning private AI musings into content for others to like and comment on. 

The bigger picture: It’s a compelling sell, particularly after a pandemic that underscored how many people feel alone (The U.S. Surgeon General even declared a “loneliness epidemic” last year). Little wonder Meta isn’t alone in this arena – Snap has given its millions of teen users a chatbot “friend” by default, and countless startups offer AI companions for everything from romance to therapy. 

Between the lines: This isn’t altruism. Instead, it looks like a business model designed to monetize emotional need. The more time users spend confiding in their chatbot, the more data Meta gathers – emotional patterns, preferences, vulnerabilities – all of which can feed its advertising engine. Just like social media engagement, emotional engagement becomes a metric. And as with apps like Replika, a premium tier is almost inevitable: want deeper conversations, personalized memories, or role-play features? That’ll cost extra. Loneliness isn’t just a problem to solve — it’s a growth market.

As Meta rolls out these increasingly realistic AI personas the risks for teenage users surface. Startups like Character.ai followed this path, raising hundreds of millions in VC funding to give users “personalized AI for every moment of your day”. The result: lawsuits after young users have seen mental health declines.

Emotionally responsive chatbots without clear age filters or usage boundaries opens the door to unintended consequences. Lawmakers have called for stronger safeguards, warning that AI systems designed to simulate friendship and emotional support shouldn’t be left to evolve in the wild, especially on platforms widely used by children. Meta has said it's working on moderation tools, but for many, it raises a broader question: should AI companions be marketed to everyone, or do some groups need extra protection?

Meta’s AI friendship vision, if realized, could entrench isolation rather than cure it, all while harvesting our emotions for profit. Social media was supposed to connect us, yet here we are, being steered toward simulated companionship. Not a triumph of innovation, but a concession that we can’t organize ourselves to care for each other, leaving it to Big Tech to fill the void. Simulated companionship isn’t connection; it’s a business model. And we’ll all pay for it, one confession at a time.

Which image is real?

Login or Subscribe to participate in polls.

🤔 Your thought process:

Selected Image 1 (Left):

  • “AI would have tucked in his shirt”

  • “I looked at the people in the background of the first image and noticed one silhouetted person whose arm just sort of disappeared in an unlikely tapered off point. That felt like enough for a snap judgement.”

Selected Image 2 (Right):

  • “I was looking on my phone rather than desktop (yes, I’m blaming the gear, ), plus I couldn’t see the second element casting the shadows in the left hand bowl and thought that was an AI botch-up.”

  • “This one ended my streak of 19 correct guesses! Nice.”

💭 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.