Therapy bots & more (April 2, 2025 edition)
David Pawlan
Co-Founder
Hey friends,
We're only a day past April Fools', but AI didn’t get the memo—because this week’s drops still sound like pranks… except they're real. Let’s dive into the news making us say “wait, that’s actually live?”
🤖 Agents That Build Agents (Yup, That’s a Thing Now)
NYC-based Emergence just revealed an AI platform that creates... other AI agents. You give it a prompt, and boom—it spawns mini-agents that can research, analyze, write, and even self-improve. One agent can break into teams, assign sub-tasks, and spin up support agents as needed.
🧠 Takeaway:
If you thought managing humans was tough, wait until you're overseeing five little bots arguing about who gets to write the report.
🎥 Runway Gen-4 Fixes AI Video’s Biggest Flaw
The new Gen-4 model actually maintains character and scene consistency across video shots—solving the “melts-into-mush” problem AI video had. You can now generate visual stories that feel coherent, not just clever.
🎬 Bonus:
Creators are already generating music videos and short films with it. AI video is no longer a party trick—it’s aiming for studio workflows.
🧠 Therapy Bots Just Passed Their First Clinical Trial
A Dartmouth study showed AI chatbot “Therabot” reduced depression symptoms by 51%, anxiety by 31%, and even created real emotional trust with users. That’s comparable to top-tier human therapy.
💬 It wasn’t a gimmick
Users kept coming back even when they didn’t have to. AI isn’t replacing therapists… yet. But it may soon be your first line of care.
🖼️ ChatGPT’s Image Generator Goes Free (With Limits)
OpenAI’s GPT-4o-powered image generation tool is now available to free users. It’s gone viral thanks to anime-style visuals, but beware—demand is so high it’s slowing everything else down.
💸 The scale’s wild:
OpenAI now has 20M paying users and hit a $40B funding round. But there’s drama—SoftBank’s money depends on OpenAI hitting its “for-profit” milestones by year’s end.
🧑🚀 Meta’s MoCha: Realistic Talking Characters
Meta just launched a model that combines video and audio to generate animated characters who speak and move naturally—hands, face, body and all.
🧪 Use case?
Interactive learning, digital actors, or AI stand-ins for influencers (👀). Next step: giving them opinions and letting them start YouTube channels.
🧭 Google's Gemini Robotics Wants to Move IRL
Google DeepMind revealed "Gemini Robotics"—extending their language model into physical space by fine-tuning it for robotic control. If Gemini 1 talked, Gemini Robotics walks.
🤖 Reminder:
These aren’t Boston Dynamics bots—yet. But they’re pointing toward language-first control of future machines.
🌐 OpenAI Academy Goes Public
OpenAI launched a free AI education platform with lessons on everything from GPT basics to alignment theory. The goal? Make AI literacy global.
💡 Insight:
Whether you’re a high schooler or policy advisor, it’s worth a browse. And yes, there’s even a course designed just for seniors.
Big Picture?
This week felt like a shift from "cool demos" to "systems you might actually use." Agents are working together, video generation is cinematic, and therapy chatbots just got real.
The underlying theme: AI is turning from novelty to infrastructure—and it’s doing it fast.
Until next time,
David
P.S. The agent that wrote this is very proud of itself. Just kidding… or am I?