- AI Fire
- Posts
- 💰 Meta Can NEVER Buy AGI Talent
Read time: 5 minutes
An AI agent just walked and outperformed junior analysts at McKinsey & Goldman in 89.1% of tasks. A tiny new brain-like model that crushed top AIs on IQ benchmarks.
What are on FIRE 🔥
IN PARTNERSHIP WITH CLICKUP
Most AI tools start from scratch every time. ClickUp Brain already knows the answers.
It has full context of all your work—docs, tasks, chats, files, and more. No uploading. No explaining. No repetitive prompting.
ClickUp Brain creates tasks for your projects, writes updates in your voice, and answers questions with your team's institutional knowledge built in.
It's not just another AI tool. It's the first AI that actually understands your workflow because it lives where your work happens.
Join 150,000+ teams and save 1 day per week.
AI INSIGHTS
A new AI agent just entered the chat for Excel and it’s already claiming to beat first-year McKinsey and Goldman analysts. Meet Shortcut, a spreadsheet-native AI assistant that builds models, fills out data, and one-shots tasks usually handed to junior bankers.
And yes, it’s blowing up on X. One demo showed it auto-populating a DCF model using SEC filings, no human input needed.
Shortcut is a full-blown Excel AI agent, not just a formula helper. No need to manually research, format, or double-check formulas. The bot does it all in one pass.
Here’s what Shortcut did in its demo tests:
Was benchmarked against junior analysts in private equity, banking, consulting, and product roles.
Won 89.1% of the time when scored by managers from those same firms.
Also reportedly beat ChatGPT Agent in 90% of head-to-head tests.
→ These results are unverified (and the criteria for "winning" is a little fuzzy). But the hype is real.
Finance influencers on X are calling it a “ChatGPT moment for Excel.” Echoes Dario Amodei’s warning: half of junior office jobs may vanish due to AI.
Why It Matters: Let’s be real, if a robot can finish your 4-hour Excel task in 20 seconds, you’re at least going to try it. It’s full-task automation. Replace data. Build models. Write analysis. Export. Done. While factories were automated by machines, spreadsheets are now getting “one-shotted” by bots.
PRESENTED BY GROUPON
Groupon brings you thousands of local steals—from family-friendly museums and outdoor adventures to cozy wine tastings and hottest concerts—discounted up to 70%. Browse once and you’ll be planning three weekends ahead, confident that you’re squeezing every drop out of fun out of the season, and not your wallet
TODAY IN AI
AI HIGHLIGHTS
🤝 The new sub-agents feature in Claude lets you build a team of agents to carry out multiple tasks in parallel. See how people are building and using sub-agents here.
👨💻 JSON prompting is the most underrated AI skill of 2025. It turns your GPT, Gemini, Claude into consistent agents, no hallucinations. Here’s the guide (copy-paste ones)
📄 Google Search’s AI Mode gets an update to ‘see’ your homework. You could upload PDFs or use the camera to set AI Mode off on a search for answers. Here it is.
🎓 ChatGPT got a new ‘Study Mode’ that turns it into your personalized tutor with adaptive lessons. You can take an academic shortcut if you ignore this opt. Try it here.
🧵 Runway just dropped a new in-context video model called Aleph, for multi-task visual generation. People are already creating wild stuff. Here’re 13 insane examples.
💼 Meta’s still tried to poach half of Thinking Machines Lab, and failed! Zuckerberg personally slid into DMs. Some offers were $1+ billion. Still, not a single person joined.
💰 AI Daily Fundraising: The NSF is investing $100 million in AI research, partnering with the Simons Foundation, NIST, Capital One, Intel, to advance AI in materials, and fundamental technologies.
AI SOURCES FROM AI FIRE
NEW EMPOWERED AI TOOLS
🎬 Kasi AI turns long videos into relevant, viral clips your audience likes
🖼️ Cashzine removes backgrounds & enhances photos in seconds
🌌 Immersity for Mobile turns images into immersive videos instantly
⚙️ Layout.dev turns your simple ideas into working prototypes easily
AI QUICK HITS
💸 Z.ai released an AI model cheaper than DeepSeek, free to download.
💼 Meta is going to let job candidates use AI during coding tests.
🤖 Google revealed an internal site “AI Savvy Google” for non-tech staffs.
💼 Anthropic published research on automated agents that audit models.
🎧 Spotify hints at a more chatty voice AI interface in the future.
AI CHART
A new 27M-parameter AI model from Sapient Intelligence just embarrassed Claude and OpenAI on reasoning tasks, by acting more like a brain than a blender.
Meet the Hierarchical Reasoning Model (HRM): a lean, fast, and shockingly effective new AI that uses a dual-module setup to separate thinking from doing.
HRM uses a two-part brain-like structure:
Planner: thinks slow, strategizes (like chess).
Worker: acts fast, executes (like facial recognition).
Together, they form a hierarchy, planning and solving in a single forward pass. And it smokes larger models on reasoning tasks. On the ARC-AGI benchmark (basically an IQ test for AI):
Claude 3.7: 21.2%
o3-mini-high: 34.5%
HRM: 40.3%
Sudoku-Extreme: HRM solved 55%, Claude and OpenAI scored 0%
30x30 mazes: HRM found the optimal path 74.5% of the time. Others? Again, 0%
But HRM is also tiny in comparison, just 27M parameters. GPT-1, the O.G. GPT model, had 117 million parameters… that’s 4x MORE than HRM.
Sure, it’s still early. The model isn’t general-purpose (yet). But it might signal a future where advanced reasoning doesn’t require trillion-token pretraining or 8 A100s.
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