⚽ Robots, Exodus, and Viral Balls
Imagine the first time a robot breaks the law.
Crazy thought, but it’s surely a case of when, not if.
And if you’ve been paying attention, you’ll know how easy it is to jailbreak most LLMs.
Those same LLMs are being integrated into some of these robots. ☠️
Today’s Ai5:
- 🤖 Figure 02: The Latest Humanoid
- 👀 Runway GVE
- 🖌️ Meet 0xFramer
- ⚽ Viral Ball Videos
- 🪦 The OpenAI Exodus
Prompt of the Day 🎨
Can you believe this is an AI image? This one actually surprised me. Generated with the new open source model Flux.1. Credits to @fofrAI.
Figure 02: The Latest Humanoid 🤖
This week Figure AI released their new humanoid robot, Figure 02.
From Figure, here’s the main features of the new humanoid:
- Speech: Capable of conversation with humans through onboard mics and speakers connected to custom AI models (provided by OpenAI)
- Vision: AI-driven vision system powered by 6 onboard RGB cameras
- Strength: 4th gen hands with 16 degrees of freedom and human equivalent strength
- Understanding: Vision Language Model enables fast visual reasoning using the onboard cameras
- Power: 2.25 KWh custom battery pack in the torso (enough for 20 hours of work per charge)
It didn’t take long for the controversy to start with the Engineering Director for Optimus (the Tesla robot) posting a comparison video of the two. It shows Optimus doing exactly the same tasks, but back in December 2023. And to be honest Optimus looks much more capable.
There’s also some legitimate safety concerns about how easy it would be to jailbreak the models and get the robot to follow any command. Definitely some food for thought.
Overall it’s an exciting release from Figure. Sometimes it’s easy to forget that this is the worst it’ll ever be.
Runway GVE 👀
Generative Visual Effects (GVE) is a brand new idea released by Runway this week. Basically, it lets you blend your real life videos with generative visual effects.
To use GVE you’ll want to:
- Film a scene
- Take a still from that scene, use image-to-video to create effects
- Overlay/composite the two scenes together
There’s a full tutorial from Runway Academy here, or watch the YouTube video below.
These types of effects used to cost film studios millions of dollars. And now we can do it in a few minutes from our couch at home. Example 👇
Meet 0xFramer 🖌️
0xFramer is one of my favorite follows for creating unique AI animations. Using a blend of After Effects and AI, he creates some amazing content!
My personal favorite are these LoFi style animations. 👇
Also a big fan of these custom anime style clips!
If you want to learn how to animate like this, Framer has a course you can take too! You’ll learn After Effects, storytelling, and get better at AI image generation. (this isn’t sponsored, I just think it’s cool.) Also, check out this epic thread compiling all Framer’s animation tutorials on X.
Viral Ball Videos ⚽
I came across this video from Green Code the other night (awesome channel if you want to learn about coding and AI). In the video he writes a bunch of Python code simulating viral ball videos on TikTok and IG. With the aim to go viral himself, it’s pretty entertaining.
Anyway, it got me thinking… could I do this with Claude? You bet.
This first example, the balls split into 2 each time the diameter increases by 10%. It shows you the power of exponential growth. It almost crashed my computer. 😅
To try and control the exponential growth, I implemented “killer balls” which exist for 10 collisions only. It worked pretty well! I published the Artifact for this one, try beat my high score.
As a final step, I’m working on adding some user input so you can have a say in the outcome of the game. But! I’ve run out of time and I gotta press send on this email.
Pretty amazing that this type of code can be written in minutes with AI.
The OpenAI Exodus 🪦
There’s not many OGs left at OpenAI these days. To be honest I’ve never taken much notice, but after the latest news I decided to take a closer look. Here’s a quick rundown of the big names that have left OpenAI in the last 12 months.
And the list goes on. Including Daniel Kokotajlo and William Saunders. Both researchers working on safety and governance. And Logan Kilpatrick the head of Developer Relations.
So why is so much talent heading for the door? We can only speculate, but some ideas include:
- Safety concerns about the power of new models
- Lack of opportunity for hands on experience
- No path to AGI
- AGI Imminent
- Too much restriction from recent collaborations with US gov
I’m sure time will tell.
On another note, it seems like OpenAI are about to release something massive. GPT5 maybe?
Snack Sized 5 🍪
1️⃣ An AI generated song has reached 72nd on the German top 100 chart. Mainstream infiltration continues.
2️⃣ Joscha Bach says eventually AI may self-improve to such an extent that it can virtualize itself into our own nervous systems and brains.
3️⃣ Check out the 100 best Midjourney prompts in the world on PromptBase. Heaps of inspiration here.
4️⃣ Nvidia are cooking something big, with leaked documents showing they’re scraping “a human lifetime’s worth of video data” every single day.
5️⃣ Elon’s latest lawsuit against OpenAI reads “The perfidy and deceit are of Shakespearean proportions.” 🤣
Join the conversation on X.
Connect on LinkedIn.
Read the Blog on Rareconnections.
Feedback? Yes please! matt@myai5.com