- AI Fire
- Posts
- 💬 Chatbot Talk Affects Your Mental Health?
💬 Chatbot Talk Affects Your Mental Health?
TikTok’s Algorithm Will Be Open-Source?
Jane Lee
March 24, 2025
Plus: TikTok’s Algorithm Will Be Open-Source?
Read time: 5 minutes
The world’s top AI models just took a new AGI test—most barely scored 1%, while human panels averaged 60% accuracy.
What are on FIRE 🔥
IN PARTNERSHIP WITH BELAY
Accomplish More. Juggle Less.
When you love what you do, it can be easy to take on more — more tasks, more deadlines, more hours – but before you know it, you don’t have time to do what you loved in the beginning. Don’t just do more – do more of what you do best.
BELAY’s flexible staffing solutions leverage industry experience with AI systems to increase productivity without sacrificing quality. You can accomplish more and juggle less with our exceptional U.S.-based Virtual Assistants, Accounting Professionals, and Marketing Assistants. Learn how with our free ebook, Delegate to Elevate, and leave the more to BELAY.
AI INSIGHTS
🧠 A New AGI Test Just Beat Most of Today’s Smartest AI Models
The Arc Prize Foundation, co-founded by AI researcher François Chollet, just dropped a new test for general intelligence in AI: ARC-AGI-2. And guess what?
Even the best AI models today barely scored 1%.
Key Takeaways:
So what’s ARC-AGI-2?
It’s not another benchmark that rewards brute force or memorization. This test challenges AI to solve puzzle-like reasoning problems — spotting patterns in colored squares and generating new ones on the fly.
The twist? It’s not just about solving the problem — it’s about how efficiently you solve it. So, raw compute doesn’t save you here.
Here’s how the top models performed:
OpenAI’s o1-pro & DeepSeek R1: 1% to 1.3%
GPT-4.5, Claude 3.7 Sonnet, Gemini 2.0 Flash: ~1%
OpenAI’s o3, which crushed the older ARC-AGI-1 with 75.7%, scored just 4% on ARC-AGI-2 — using $200 per task 😳
In comparison, human panels averaged 60% accuracy
Why does this matter?
Because ARC-AGI-2 was designed to fix the flaws of ARC-AGI-1 - where models could get high scores by throwing massive computing power at the problem.
ARC-AGI-2 shows that today’s top AI models, while powerful, still lack the ability to adapt and reason efficiently. They struggle when faced with new problems that require flexible thinking rather than memorized solutions or brute-force computation. This test highlights a clear gap between task-specific performance and true general intelligence — the kind needed to handle real-world complexity.
The Arc Prize Foundation is now challenging developers to close that gap. Their new contest sets a tough goal: reach 85% accuracy on ARC-AGI-2 using just $0.42 of compute per task. It’s a bold benchmark aimed at pushing AI beyond raw power into smarter, more efficient reasoning — a step closer to true AGI.
PRESENTED BY PROTON
Keep Your Family Safe Online
It's not easy to keep your family safe online. That's why Proton created Pass Family, a password manager subscription for up to six people managed by one or two admins. Whether you're helping your parents or your children, Pass Family makes it easy to create, store, autofill, and share passwords securely. Vaults allow you to organize logins and make them accessible to all or selected family members. Store logins, bank cards, and secure notes with end-to-end encryption so nobody else can access them. As an AI Fire reader, you can unlock an exclusive 60% discount on Proton Pass today!
TODAY IN AI
AI HIGHLIGHTS
📱 Perplexity wants to rebuild TikTok in the U.S. - but smarter and more transparent. With open-source algorithms and U.S.-based servers, it aims to make content more trustworthy, powered by its own AI tech.
⌚ Apple plans to launch AI Apple Watch with a built-in camera by 2027. It’ll sit near the Digital Crown and power a new AI system called Visual Intelligence—so your watch can “see” and give real-time info.
🧪 Nvidia is opening a new quantum computing center called NVAQC. CEO Jensen Huang once said quantum tech was 20 years away—then joked at GTC, “This is the first event in history where a company CEO invites all of the guests to explain why he was wrong.”
🧠 Tencent just dropped Hunyuan T1, a powerful reasoning model that rivals DeepSeek R1 - using a new hybrid Transformer-Mamba design for better speed and efficiency at the same price.
🧬 VantAI just launched Neo-1 - the first AI model that can both decode life’s protein structures and design new ones. Backed by big pharma, it even outperforms AlphaFold 3 in versatility.
🤝 OpenAI and Meta are in talks with Reliance to expand AI in India. Reliance is planning a giant 3GW data center in Gujarat—possibly the world’s largest—to power next-gen AI models.
💰 AI Daily Fundraising: n8n just raised $60M after 5x revenue growth—doubling in the last two months. With 3,000+ enterprise customers, it’s now leading the way in AI-powered workflow automation.
AI SOURCES FROM AI FIRE
NEW EMPOWERED AI TOOLS
⚡ Zapier MCP lets your AI assistant connect with 6,000+ apps—no API work needed.
🧠 Claude’s Think Tool helps Claude pause and reflect before answering for smarter results.
🌟 StarVector turns images or text into clean, editable SVG code instantly.
📝 Knowt turns your notes into quizzes, flashcards, and study guides in seconds.
✅ Beacons is an AI-powered link aggregator, link in bio, and social media marketing tool.
AI QUICK HITS
🎥 Google Rolls Out Gemini’s Real-Time AI Video Tools (Link)
🧬 AI Diagnoses Major Cancer With 99% Accuracy (Link)
🛡️ Microsoft Launches Security AI Agents for Human Support (Link)
🗣️ OpenAI Upgrades Voice Assistant for Smoother Conversations (Link)
💡 Ant Mixes Chinese and U.S. Chips to Cut AI Costs (Link)
AI CHART
New MIT & OpenAI Study Says Yes — But It's Complicated
A team from MIT Media Lab and OpenAI just ran one of the most detailed studies yet on how AI chatbots impact your emotional well-being - and the results are… eye-opening.
1. Voice Chat Helps - In Moderation
Chatting with voice (vs. text) actually helped people feel less lonely and less dependent on the chatbot... BUT only if they didn’t overuse it.
People who used it heavily every day? Their loneliness and dependence got worse — especially with neutral voice.
2. Personal Conversations Have Pros & Cons
When people opened up emotionally (personal convos), they felt slightly more lonely short-term. But also had less dependence on the AI and fewer signs of unhealthy use.
Non-personal, task-focused chats? → More emotional dependence, especially for heavy users.
3. Too Much Chat = Bad News
The more you use it, the worse it gets — across the board.
Heavy usage led to:
More loneliness
More emotional attachment
Fewer interactions with real people
Higher chance of relying on the AI too much
4. It’s Also About You - People who already have high emotional attachment tendencies, or view AI as a “friend,” were more likely to experience negative effects — like loneliness or emotional reliance.
5. Design Choices Matter
The way AI talks to you (text vs voice, friendly vs neutral) changes how it impacts your well-being.
But it’s not just the tech — how you use it matters just as much.
Why This Matters
This is one of the first real-world, controlled studies to show that AI chatbots are not emotionally neutral.
Used briefly and thoughtfully? They might help.
Used excessively or as emotional crutches? They can quietly pull people away from human connection.
It’s a reminder that as AI becomes more social, we need to think about mental health, not just features.
Let the AI talk — but maybe not all day, every day.
AI JOBS
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply