.jpg)
.jpg)
Welcome, humans.
Meta just dropped some major AI news Tuesday that's got everyone buzzing (more on that bombshell below).
Meanwhile, OpenAI spent the day essentially putting its overenthusiastic ChatGPT-4o in timeout. The update that made the model my biggest fan (and yours, and everyone's) got yanked faster than free donuts disappear from the office kitchen.
This 4o rollback saga has been wild to watch unfold. It took X users approximately 0.2 seconds to uncover the differences in the system prompt that caused all the trouble. Consider this your friendly reminder that even tiny tweaks in wording can completely change how these AI models interpret what you want.
I've learned this lesson the hard way, too—last week I asked Claude to “write something fun” and somehow ended up with a delightful 2,000-word history of carnival games. AI, am I right?
Before we move on to the day’s big news, check out the website a crafty dev built to play chess with 1 million of your closest friends—simultaneously.
Here’s what you need to know about AI today:
- Meta AI launched its first standalone AI app.
- Duolingo announced it will become an "AI-first" company.
- The White House is investigating “woke AI.”
- WhatsApp launched privacy-focused AI tools.
Meta AI: Your Digital Life + Its App = The Most Personal Assistant Yet?
Hey Siri, meet your new competition! Meta just launched its first standalone Meta AI app, merging its Ray-Ban smartglasses companion with a social-first AI assistant that's gunning for ChatGPT's crown.
Remember those Ray-Ban Meta smartglasses? If you own a pair (like I do), you might have noticed your Meta View app acting strange today. That's because Meta just transformed it into their new AI hub—absorbing your glasses settings while adding a whole new layer of AI functionality. — If you've been on Instagram since 2012, Meta probably knows your aesthetic preferences better than your significant other.
What makes Meta AI different?
- Powered by Llama 4: Built on Meta's freshly released models, including the still-cooking Behemoth (yes, that's really its name).
- Knows your digital life: Taps into your Meta profile and engagement history for more personalized responses.
- Social AI: Features a "Discover feed" where you can see how others are using AI—turning it into something more Instagram-like.
- Voice-first approach: Emphasizes hands-free interaction with a "Ready to talk" toggle.
- Ray-Ban integration: Start a conversation on your glasses and continue on your phone (but not vice versa...yet).
For Mark Zuckerberg, this app is clearly step one in his master plan to dominate AI by year's end. During Meta's fourth-quarter earnings call, Zuckerberg predicted his AI assistant would reach "more than 1 billion people" this year.
Bold words from Zuck, considering ChatGPT has already secured more than 5.5B monthly visits. But remember when Instagram copied Snapchat's Stories and then completely dominated it? Meta has a history of successful "borrowing."
The good: If you've been on Meta's platforms for years, this AI might actually "get" you better than competitors.
The not-so-good: Meta is jumping into an already crowded pool, and the idea of Meta digging deeper into your data might set off privacy alarm bells.
Reality check: Early tests suggest the personalization isn't dramatically better than competitors (yet).
Our take: Can Meta really overtake ChatGPT? In the short term, probably not. But Meta has something most AI companies don't: billions of active users across existing platforms. From a data perspective, this is likely the most intrusive AI approach yet. Whether that leads to AI bliss or privacy hell, however, remains to be seen.
If even a small percentage adopts the standalone app, Meta AI could reach a scale that took others years to achieve. Is that enough? Zuck clearly thinks so.
Read our full deep dive here.
FROM OUR PARTNERS
Hostinger Horizons
Dario Amodei, CEO of Anthropic, says AI will generate 90% of all code within six months. Within twelve - maybe all of it.
Stay ahead of the curve. Launch your app in minutes with Hostinger Horizons.
All it takes is a prompt - you no longer need a big team or a massive budget. Describe your idea and instantly receive the first version.
Want to change something? Just chat with the AI. Then, publish your creation with one click. No coding, no hassle.
Hostinger Horizons has everything to make your vision happen, including hosting, domain names, professional email, and 24/7 live support.
Prompt Tip of the Day
This “historical time-freeze” prompt transforms your AI chat into a legitimate time machine. We've been testing it with historical figures from Aristotle to Einstein, and it's crazy how quickly the AI adapts to historical mindsets without leaking future knowledge.
Just copy-paste this formula:
"You are [historical figure] in the year [YYYY]. I'm a journalist from 2025. Answer my questions about [topic] strictly from your current-year perspective—no hindsight, no spoilers."
We just had Marie Curie explain radioactivity circa 1898, and she perfectly captured the early excitement without mentioning any of the dangers scientists wouldn't discover for decades.
Our favorite insight: Adding “I'll be grading your historical accuracy” at the end makes the AI triple-check itself before mentioning anything that wouldn't exist yet. The difference is night and day!
Treats To Try.
- *RAD Intel helps Fortune 500 brands like Hasbro read the room in real-time. Pre-IPO at $0.60/share with 16X valuation growth. Invest by May 8th.
- Dzine helps you create and edit AI images with layer-based tools and style controls—free trial with 30 daily credits.
- Luna.ai finds leads and writes personalized B2B outreach emails—free to start, paid plans available.
- zTalk.ai helps you chat with your data across different platforms, letting you ask questions and get summarized insights from your documents, spreadsheets, and presentations.
- Poised coaches you through presentations and meetings by analyzing your communication in real-time—free tier, then $12/month.
- ChatwithPDF enables natural conversations with your PDF documents, extracting the exact information you need—free tier, then $9.99/month.
- Amazon Bedrock provides access to foundation models from various AI companies with enterprise-grade security—usage-based pricing.
- Exa Websets helps you find and enrich perfect lists of leads, candidates, or anything on the web through semantic search—free trial, then paid.
See our top 51 AI Tools for Business here!
*This is sponsored content.
Around the Horn.
- Duolingo announced it will become an "AI-first" company, gradually replacing contract workers with AI, integrating AI into hiring and performance reviews, and only approving new headcount if teams cannot further automate their work.
- The Trump administration launched an investigation into tech companies' AI diversity efforts, targeting previous initiatives to reduce algorithmic bias and reframing the discourse around "woke AI" in technology development.
- Cisco and ServiceNow partnered to simplify and secure AI adoption for businesses at scale.
- Tempus introduced Loop, an oncology platform combining real-world patient data, biological models, and CRISPR screens to accelerate target discovery and validation.
- WhatsApp launched privacy-focused AI tools that aim to add features without breaking end-to-end encryption.
- ICYMI: Alibaba launched Qwen3, an open-source large language model series with eight variations, featuring hybrid reasoning capabilities and multilingual support, challenging both Chinese and US AI industry leaders—try it here.
FROM OUR PARTNERS
🚀 Transform Raw Documents into Structured Data in Minutes
Give Unstructured a try—just drag in any document to get structured JSON with no set up. Instantly parse PDFs, images, emails, and more. And when you’re ready, you can scale up and process thousands of documents just as easily.
Try Unstructured for free today!
Midweek Wisdom
- SemiAnalysis did a deep dive on Microsoft’s “data center pause”, which to date has included abandoning 2GW+ of potential power due to right size their capacity for the demand.
- New research from UCLA and Meta AI slashes language model response times from 30 seconds to just 3, introducing a revolutionary framework (d1) that turbocharges diffusion models' reasoning abilities through clever reinforcement learning, potentially reshaping enterprise AI workflows.
- Check out this hilarious AI experiment where researchers unleashed AI agents into a simulated company, revealing a comedic disaster of robotic incompetence.
- Researchers developed ALLMAssert, a novel approach using large language models to automatically generate multiple test assertions for software testing, demonstrating significant improvements in generation and bug detection.
- European researchers conducted a comprehensive analysis of AI's evolving role in media, revealing how artificial intelligence is fundamentally reshaping content creation, distribution, and marketing while navigating complex ethical and legal landscapes.
A Cat's Commentary.
