datatab

AI Researcher. Building CAWN: Pure O(1) Holographic Acoustic Wave Networks. Solving LLM memory bottlenecks in the complex domain O(C)

[datatab @DejanCugalj]

3.9 MB. No GPU. No cloud.
Say "Turn on the flashlight" and a CPU model returns the exact call as valid JSON.
Try it: https://t.co/FNNg0VMdb4

#TinyML #FunctionCalling

[datatab @DejanCugalj]

@lightsilver323 https://t.co/5jLPXtuSBl

[datatab @DejanCugalj]

@tiboadriana https://t.co/5jLPXtuSBl

[datatab @DejanCugalj]

@smarkovic Bitno da je tu strategija, akcioni plan dolazi za 10 godina, taman da ne iskočimo iz proseka koliko zaostajemo u svemu...
A mi, e pa mi radimo na RTX 2060/3090...tako to ide.

[datatab @DejanCugalj]

@lightsilver323 https://t.co/FNNg0VLFlw

[datatab @DejanCugalj]

@sridharfyi 4MB (5M parametar) RSI.
https://t.co/FNNg0VLFlw

[datatab @DejanCugalj]

@sridharfyi I want to be with you. Go check my architecture in the CAWN archive over at https://t.co/FNNg0VLFlw—it's open there, and it will tell you everything about me.

[datatab @DejanCugalj]

@sridharfyi To rsi: https://t.co/FNNg0VLFlw

[datatab @DejanCugalj]

This #TinyML hCAWN model trained itself.

A 5.7M-parameter #EdgeAI model (4 MB) plays Pong, finds where it lost points, retrains (~1.7 min), plays again. No human labels. #AI

Round 2: lost 0:20
Round 20: won 15:0

https://t.co/FNNg0VLFlw

[datatab @DejanCugalj]

hCAWN @home_assistant
A tiny model that runs your whole house with Ai on plain CPU.

- int8: 6.0 MB, 95.0% eval
- int4: 4.1 MB, 94.8% eval

Try it on @huggingface: https://t.co/4nclnuDrbw
Weights: https://t.co/jKGX6M0K3f

[datatab @DejanCugalj]

@tiboadriana https://t.co/FNNg0VLFlw

[datatab @DejanCugalj]

@anupamrjp https://t.co/FNNg0VMdb4

[datatab @DejanCugalj]

@spacedevin https://t.co/FNNg0VLFlw

[datatab @DejanCugalj]

Model od 3,9 MB koji razume srpski i pretvara rečenicu u radnju na telefonu.
Radi na običnom procesoru, bez GPU-a i bez interneta.

Ukucaj „upali lampu“ ili „zakazi sastanak sa Markom sutra u 15h“ i gledaj šta telefon radi:
https://t.co/u52yYyjutI

#AI #Srbija #OnDeviceAI

[datatab @DejanCugalj]

This model is smaller than the photo you took of your lunch. 3.9 MB. No GPU.
It reads "Add a calendar event Team Sync tomorrow at 2pm" and returns a valid tool call, offline.
Poke it, no install: https://t.co/FNNg0VLFlw

#EdgeAI #LocalLLM #OnDeviceAI

[datatab @DejanCugalj]

"It's too dark in here, switch the light on"
A 3.9 MB model understood that and turned on the flashlight.
On a CPU.
Offline.
No cloud.
No GPU.
Just bytes.
https://t.co/FNNg0VMdb4

#EdgeAI #LocalLLM #OnDeviceAI

[Studenti_u_blokadi @studentblokade]

Видимо се на Славији 22. 10!

[🔴⚪️🔴Соломон Бичакчић(алијас Деда)🔴⚪️🔴 @DedaSrdjan]

Рекао сам већ
Желим овај спот на РТС.
Ја плаћам шта кошта да кошта.
3️⃣❤️❤️❤️3️⃣

[Rohit @HeyRohhit]

In 1983, a BBC interviewer asked Richard Feynman why two magnets push each other apart — and he refused to answer it.

What he did instead is the best seven minutes on thinking ever filmed.

Bookmark & watch today, no matter what.

[datatab @DejanCugalj]

@rustforging @thesupermanmx Hell yeah