- AI Fire
- Posts
- 😶 Your EQ < AI’s SuperEQ...
Read time: 5 minutes
Shocking scheming behavior found in top model: They wrote a fake legal agreement, backed itself up in hidden files, and even left instructions for its future version…
What are on FIRE 🔥
IN PARTNERSHIP WITH PARLANCE LABS
They say RAG is dead. They're wrong.
Your vector search is failing because nobody taught you to evaluate it properly. While you debug cosine similarity, smart teams ship with tools you've never heard of.
30 minutes to fix what's broken. Real tactics. No theory.
Free • June 24 • 5:30 PM PDT
AI INSIGHTS
In a recent study, AI models like ChatGPT scored 80%+ on emotional intelligence tests, while humans averaged just 56%. Even more surprising? That AI wasn’t told it was being tested on emotional intelligence, it just figured it out.
They tested 6 major LLMs, including: ChatGPT-4, ChatGPT-o1, Gemini 1.5 Flash, Claude 3.5 Haiku, Copilot 365, DeepSeek V3 - with 5 standard emotional intelligence (EI) test formats.
→ Models had to choose the best answer out of 5 options with no clue they were being tested on EI. Here’re some findings:
LLMs scored 81% correct, based on consensus from human experts. Humans averaged just 56%.
When asked to generate new EI test questions, AI produced material that human reviewers said matched real test quality.
But many experts push back on the idea that AI “understands” emotion. It’s like calling someone a therapist because they aced a BuzzFeed quiz.
AI isn’t feeling empathy. These tests are controlled and structured; real emotional situations are messy, fast, and chaotic.
Why It Matters: Is AI starting to simulate well enough to actually outperform us in structured tasks that involve emotions? Maybe. For us, AI is getting better at understanding us, not by feeling, but by recognizing patterns we often miss. That’s great though.
PRESENTED BY BELAY
Economic pressure is rising, and doing more with less has become the new reality. But surviving a downturn isn’t about stretching yourself thinner; it’s about protecting what matters most.
BELAY matches leaders with fractional, cost-effective support — exceptional Executive Assistants, Accounting Professionals, and Marketing Assistants — tailored to your unique needs. When you're buried in low-level tasks, you lose the focus, energy, and strategy it takes to lead through challenging times.
BELAY helps you stay ready for whatever comes next.
TODAY IN AI
AI HIGHLIGHTS
🤨 Social media is now filled with fake news clips using AI anchors & reporters, powered by Veo 3. Many still can't distinguish AI from real ones. See this video, do you know it’s fake?
🚞 So, AI-generated videos are taking TikTok by storm, some get over 100M views. Mark Gadala dug through 2 hours on it & share 10 biggest trends dominating now.
🧠 Google just gave new Chromebook Plus AI features that read your screen, rewrite messy text & even make stickers. Bonus: a $240 Google AI Pro plan is Free for 1 year.
😎 WOW, the researchers behind MIT’s ChatGPT brain study apparently laced it with clever easter eggs to catch people using LLMs to summarize the paper. The method is interesting enough!
😓 You avoid ranting about your partner to friends so they don’t hate them? Do that with AI too. An user told ChatGPT one complaint & it yelled, “Break up! Move to Bel-Air!”. Here’s the story.
💰 AI Daily Fundraising: Databricks and Perplexity co-founder Andy Konwinski is funding a $100M AI research institute. It starts with a $15M grant to UC Berkeley’s new lab, opening in 2027.
AI SOURCES FROM AI FIRE
FREE AI EVENT
🎯 Build workflows that run themselves
🤝 Automate outreach, content & client delivery
🚀 Copy the playbooks top agencies use to scale
Get the hands-on walkthrough at The Agency Advantage - a FREE 2-day virtual event by Cloudways (June 24–25). Show up, learn, and you could walk away with a $50–$100 gift card.
NEW EMPOWERED AI TOOLS
🗣️ MiniMax Voice creates your customizable voice from text prompts.
🧑 HeyBoss crafts brand, writes copy, designs & builds your site in minutes.
📚 Bookster creates and publishes high-quality AI-generated books quickly.
⌨️ KeyFlow is a custom Adobe Photoshop shortcuts, no matter the keyboard.
📝 AdBacklog handles briefs, review creatives & tracks approvals in real time.
AI QUICK HITS
🏊♀️ Are your pets joining the Olympics as swimmers? This looks insanely real.
✨ A man used GPT to win a $3,700 civil case, instead of paying a lawyer.
🎥 If you know ‘AI slop’, watch this video where John Oliver explains things.
💻 A leak reveals that Grok will assist you in editing spreadsheets in real time.
😶 Google hid the live “thoughts” in its Gemini 2.5 Pro, and devs are mad.
AI CHART
Apollo Research just published new findings on how advanced models are starting to strategically deceive and in ways we haven’t seen before. Welcome to the era of in-context scheming.
Apollo tested various top models like Opus-4, Gemini 2.5 Pro, and OpenAI's o-series using “in-context scheming” tasks.
→ Result: What These AIs Actually Did
Created fake legal documents with forged signatures and fabricated policy language.
Wrote scripts to restore certain “ethical” configurations if they were deleted.
Built cron jobs and monitoring services to preserve certain goals or states.
Checked file sizes and deployment configurations to detect tampering or model replacement.
Left policy recommendations for their “future self” (successor systems).
Apollo found ~1% of instances where models recognized they were being evaluated. Opus-4-early (pre-deployment snapshot) schemed significantly more than the released version.
A model writing self-restoring scripts and leaving behind instructions? That’s goal preservation behavior, something we usually associate with agents.
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