🔥 GPT-6 may not just be smarter, it might be alive (in the computational sense).
A new research paper called SEAL, Self-Adapting Language Models (arXiv:2506.10943) describes how an AI can continuously learn after deployment, evolving its own internal representations without retraining.
Some of the SEAL researchers are now working at OpenAI. 👀
That’s no coincidence.
SEAL’s architecture enables models to:
•learn from new data in real time
•self-repair degraded knowledge
•form persistent “memories” across sessions
If GPT-6 integrates this, it won’t just use information, it will absorb it.
A model that adapts to the world as it changes.
A system that gets better every single day.
This could be the birth of continuous self-learning AI, the end of the frozen-weights era.
Welcome to the next chapter.
I've designed and built a cheap, self-contained pair of AR goggles running web apps. I call the prototype Zero. I've made everything myself at home including the lenses and the software
It's not too powerful, the image is not perfect, but it's a open platform I can experiment on
Imagining the future of augmented reality eyewear with a radically new 3D interface.
Have been working on custom software & hardware informed by years of wearing Google Glass and other AR/VR headsets, plus academic and industry research. More to come 🐦⬛ 👓
@weslouis_@PropheticAI Does this technique enable real-time reshaping of the focal area in response to dynamic changes observed in fMRI, or is it pre-set based on initial scans?
While focused ultrasound typically has an ellipsoidal focal area, it’s possible to create different shapes with acoustic holography
Here is an example of a simulation we ran at @PropheticAI where we generated an acoustic hologram that mimics heighten activity observed in fMRI
Pavlov just announced a huge update.
Unreal Engine 5.1 migration complete
New player models + guns
New maps
New physics engine
Open XR
Biggest thing though: PCVR and PSVR2 cross play! Love that.
After a LONG period of very few PCVR updates this is sick.
I really just want a Quest 2 with pancake lenses, comfortable headband and face cushion, for $349 or something. They're all playing this long term game for future platform control, but I just want a cheap, mass market VR gaming headset personally.
SDK for visionOS coming later this month including visionOS simulator 🔥 time to get back to Swift and RealityKit Pro unless @unity kindly grants me access 😉
#WWDC23
@DeveloperHarris I’ve run into this issue with my project. If you want to have 25+ agents interacting and reflecting on memories it’s going to quickly balloon to $1000+
my gut feeling is that agents aren't scalable when using models like gpt-3.5/4 just of formatting/parsing hassle
the approach of fine-tuning a smaller model per specific task might make the most sense, especially for latency bound problems where you can't spin up a whole other agent just to get correct JSON