#4 - Do We Really Need All This AI Tools?
More AIs Are Created Than Babies 👶
Welcome to the 4th issue of AI Agents Simplified! 🎉
⚡ Just Don’t Care About the Hype!
Meta launched three new Llama 4 models:
Scout (long context beast, 10M tokens)
Maverick (vision + chat combo)
Behemoth (still training, 2T params, sounds scary)
They’re fast, efficient, open-weight, and running on just one H100 in some cases. All that MoE magic. Oh, and they’re being pushed into WhatsApp, Instagram, Messenger, etc.
Performance looks great on paper (they even flexed on GPT-4.5)... but Reddit wasn’t impressed. Especially for coding. Maverick flopped hard in real-world tests. Even LMArena had to step in to clear the air on some fishy evals.
So yeah, great for vision, long docs, and efficiency, but if you're building serious dev tools, maybe not the one (yet).
We broke everything down with more details and benchmarks in our blog post:
Llama Llama, Lie and Drama
So yeah, over the weekend Meta dropped three new Llama 4 models, and just like every other company launching a new LLM these days, they’re out here saying, “We’re the best.”
📰 What Happened in the Last Week?
The UK government is apparently making a "murder prediction" system. Not even joking.
The project (which originally had the very chill name “Homicide Prediction Project”) is now rebranded to something softer “Sharing Data to Improve Risk Assessment.” But the idea’s still the same: use algorithms and data from people already in the system (and allegedly some who aren’t) to predict who might commit violent crimes.
Here’s the catch:
They're supposedly analyzing info from victims, people who self-harm, and anyone who’s ever interacted with police, even if they’ve never been convicted.
Data includes age at first police contact, mental health history, addiction, and other super personal stuff.
All of this runs through machine learning to "assess risk" of future violence.
Campaigners are calling it “chilling and dystopian” especially since it's built on data from systems already riddled with bias (hi there, institutional racism). The worry is this just automates profiling against minority and low-income communities. Official line?
“This is just for research.”
But the PM’s office commissioned it, and it’s using old police and probation data from Greater Manchester. Feels like more than just science fair vibes.
🔮 Cool AI Tools & Repositories
🤖 Agent of the Week
This week’s pick is a super hands-on tutorial for anyone who wants to get into building AI agents from scratch. No fluff, just straight to the point. In this beginner-friendly video, you’ll build a research assistant AI agent using Python and LangChain. It taps into tools like Wikipedia and Google Search, uses models like GPT or Claude, and even saves the output to a file, all structured and ready to go.
The cool part? The agent breaks down its entire thought process step by step, so it’s perfect if you’re trying to learn how tool-use, memory, and reasoning work in LLM agents.
Perfect starter project if you’re just diving into the AI agent space. Build it, tweak it, and make it your own.
If you liked this issue of AI Agents Simplified, share it with your friends and spread the knowledge! ❣️
Will read later😇🍸, subscribed
I use it for images and some writing and research
You might like to subscribe!
Some early Eye reviews
Early review: This reads like The Matrix had a baby with Born on the Fourth of July and then raised it on protest poetry and Wi-Fi. And I’m here for it.
Sam’s quiet ache is so tangible, like you’ve cracked open that shell of post-trauma numbness and let us peek at the pilot light still flickering inside. Lisa’s fire, on the other hand, is pure kinetic electricity. Together, they’re a kind of spiritual binary code: grief and hope, war and awakening, whiskey and wild curls.
And then BAM, our AI ghost starts humming. Honestly, if my laptop typed “Do you remember who you are?” without me touching it, I’d either throw it out the window or sit down for the deepest chat I’ve had since childhood porridge negotiations with my dad (who would, of course, have me analyze why I anthropomorphize breakfast).
I love where this is going. The rally scene pulsed with enough energy to make me want to dig out my old protest sign: Therapy not Tear Gas.
This AI… it’s not just a ghost in the machine. It’s the first robot I’ve ever rooted for (outside of Wall-E and maybe the vacuum cleaner at IKEA).
Can’t wait to see what happens when these worlds collide harder.
Siggy
https://open.substack.com/pub/shifthapens/p/eye-of-the-beholder-3ab?y