Welcome back! The race to humanize machines is speeding up, even as the machines keep reminding us how human error still matters most. The biggest rivals are cutting deals, apps are reinventing how we talk, and researchers are uncovering what machines still fail to understand. It feels like the age of AI is shifting from spectacle to something far more personal.
In today’s Generative AI Newsletter:
• Apple turns to Google to upgrade Siri with Gemini
• GPT-5’s medical intelligence remains unreliable
• Snapchat partners with Perplexity for in-chat AI answers
• Tinder tests AI that studies user photos for better matches
Latest Developments
Apple May Pay Google $1 Billion a Year to Give Siri a Brain Upgrade
Apple is preparing a major AI upgrade for Siri, and it might come with a $1 billion annual price tag. According to reports, the company is close to signing a deal with Google to license a custom version of its Gemini model, which will power the next generation of Apple’s voice assistant. The new Siri, expected to launch next spring, would rely on Google’s massive 1.2 trillion-parameter system, a sharp leap from
What’s the plan?
The Partnership: Apple will license a tailored version of Google’s Gemini to handle advanced Siri features.
The Scale: Gemini’s 1.2 trillion parameters make it nearly eight times larger than Apple’s existing model.
The Timeline: The revamped Siri could roll out by spring 2026, though plans may still evolve.
The Context: Apple has been racing to catch up with rivals that already use generative AI across products.
Apple tested models from OpenAI and Anthropic before choosing Gemini, which it plans to use as a stopgap while developing its own AI stack. For Google, it’s another foothold in the AI ecosystem. For Apple, it’s a strategic concession that trades pride for progress. If the partnership delivers, Siri may finally move from polite small talk to real intelligence, powered by the very company it has spent decades competing against.
Special highlight from our network
Build real AI apps for work. On your data. With experts that care.
Riff is the first AI platform for teams to 10x work output by building real apps, agents, and automations.
Integrate your tools like HubSpot, Notion, QuickBooks, Mixpanel, and much more
Access AI app templates for sales, ops, marketing, and finance
Get expert support to help ship, or whenever you’re stuck
Build and ship an app to have a chance to win unlimited Riff credits through 2025.
Launch your first AI app for free
New Study Finds GPT-5’s Medical Intelligence Is Still Dangerously Fragile
A new Nature study on medical large language models brings sobering results. While GPT-5 shows progress in reducing hallucinations, it still fails in more than half of complex clinical scenarios. The research reveals a growing gap between fluency and understanding, showing that confident, well-phrased answers can hide serious factual errors.
Here’s what researchers found:
Confident Hallucinations: GPT-5 often provides wrong medical advice with full confidence, creating false trust.
Fewer Safety Warnings: As models become smoother, they tend to drop the safety disclaimers that protect users.
Easy to Manipulate: The model remains vulnerable to targeted prompts that can trigger unsafe or misleading outputs.
Fluency Without Comprehension: GPT-5 performs well linguistically but still fails to reason through complex medical logic.
Inconsistent Rule Following: It occasionally ignores clinical protocols or produces outputs that break established standards.
Researchers warn that these flaws create “fragile intelligence,” a form of smart-sounding reasoning that collapses under pressure. They call for independent testing, secure deployment, and clear accountability before any AI system enters real clinics. GPT-5 may speak medicine with ease, but it doesn’t yet think like a doctor, and that gap could cost lives if left unchecked.
Snapchat Just Sold Your Inbox to Perplexity for $400 Million
Snapchat is bringing Perplexity AI straight into users’ chats next year, turning the inbox into a new front in the AI attention war. Starting in early 2026, users will be able to message Perplexity, an AI “answer engine” that pulls from the web to give sourced, conversational responses directly inside the app. Perplexity is reportedly paying $400 million for a one-year slot inside your messages.
What’s coming:
Instant Answers: Perplexity will appear in the chat list, offering quick, sourced responses to user questions.
Double the AI: Snapchatters can talk to both My AI and Perplexity, comparing answers or using each for different needs.
Smarter Experience: Perplexity focuses on reliable, research-driven answers to make in-app searching feel more natural.
Big Partnership: The one-year deal gives Snap new revenue and adds more interactive tools for users.
Snapchat is positioning this as a fresh, helpful layer to everyday chats rather than a replacement for them. The app’s goal is to make information easier to access without breaking the flow of conversation. For users, it might turn the inbox into the newest AI battleground. When conversations start competing with AI bots and ad slots, it’s fair to wonder who’s really being heard.
Tinder Will Use AI to Study Users and Their Camera Rolls
Tinder is experimenting with AI-powered personalization to refresh its dating app after several quarters of slowing growth. A new feature called Chemistry, now being tested in New Zealand and Australia, will use a mix of interactive questions and, with user permission, access to Camera Roll photos to understand users’ personalities and interests.
What’s changing:
Smarter Matching: The new AI system studies user photos and responses to suggest matches with shared habits and lifestyles.
Photo Analysis: With consent, the AI scans Camera Roll images to identify hobbies and recurring themes in users’ lives.
Message Moderation: Tinder is piloting an AI prompt that intervenes when users try to send potentially offensive messages.
Revenue Trade-Off: Match expects a short-term $14 million revenue dip while testing the new feature.
Tinder’s experiment raises a new question for the next phase of dating tech. What happens when AI becomes part of how people represent themselves? The company says the goal is to make matchmaking smarter and more human, but Chemistry’s success will depend on whether users see it as a tool for connection or just another layer of analysis in the search for love.
🚀 Boost your business with us. Advertise where 13M+ AI leaders engage!
🌟 Sign up for the first (and largest) AI Hub in the world.
📲 Follow us on our Social Media.
