Meta's new VR glasses will bring personalized AI capabilities to the next level
But nobody wants their whole day uploaded to the cloud!!
For the #metahackathon I built an AI model that runs locally and fast (numbers below). It’s a non-autoregressive system one model.
Introducing Remembery: a 360M model that runs on device
Meta gave me $150 api credits for this, meta really stepped up their ai game recently since I last used it. #muse @alexandr_wang
Look at Alexander wang’s post
https://t.co/B98tyxdOKF
This just proves how useful remembery is. Local personal memory that never leaves your device.
Even Zuckerberg posted on it
Meta's new VR glasses will bring personalized AI capabilities to the next level
But nobody wants their whole day uploaded to the cloud!!
For the #metahackathon I built an AI model that runs locally and fast (numbers below). It’s a non-autoregressive system one model.
Introducing Remembery: a 360M model that runs on device
Meta gave me $150 api credits for this, meta really stepped up their ai game recently since I last used it. #muse @alexandr_wang
Why don't we at TypeSafe call Jev a decision model?
If you think of Jev's questions as making *decisions*, I believe you'll use it at the wrong layer of abstraction and your reliability will suffer.
I wrote all about this on our site 🧵👇
Meta's new VR glasses will bring personalized AI capabilities to the next level
But nobody wants their whole day uploaded to the cloud!!
For the #metahackathon I built an AI model that runs locally and fast (numbers below). It’s a non-autoregressive system one model.
Introducing Remembery: a 360M model that runs on device
Meta gave me $150 api credits for this, meta really stepped up their ai game recently since I last used it. #muse @alexandr_wang
Meta's new VR glasses will bring personalized AI capabilities to the next level
But nobody wants their whole day uploaded to the cloud!!
For the #metahackathon I built an AI model that runs locally and fast (numbers below). It’s a non-autoregressive system one model.
Introducing Remembery: a 360M model that runs on device
Meta gave me $150 api credits for this, meta really stepped up their ai game recently since I last used it. #muse @alexandr_wang
1/2
it scores an 80% on recall while a cloud LLM that is 100x the size only scores 81%.
Also, the cloud LLM makes a wrong memory on 5 in 500, while Remembery only does on 1 in 500.