The deviation between modern ANNs and their biological inspiration has grown too large. We train networks using global error signals that physical neurons never see.
What happens if we force an artificial network to play by biological rules?
Meet PAULA. 🧵
@MiTiBennett If you're going to melt your Mac, then use Mac Fan Control and BatFi on 50% charge in passthrough. Will save you a couple of years in battery capacity
Complexity and interpretability rarely coincide: complex systems are usually too opaque to question, while transparent ones are too simple interesting dynamics to emerge.
What if we treated LLM agents as the basic computational unit to build agentic substrates? Paper thread 🧵
Happy to share my new open source tool for exploring connected neurons in the #MICrONS dataset (thanks @AllenInstitute, again) It's in the browser, online & free. No login or install required. 1/
I finally pulled an old M5Stick C2 by @M5Stack out of the drawer for a weekend experiment. Wanted to see if I could use Web Bluetooth to bridge the gap between hardware and web apps. The idea was to turn the device into a physical "lens" for hidden content on my personal website
Excited to share our big step towards decoding the brain’s thoughts!
When we were first injecting the microbubbles for the scan, I was surprised by the coldness of the saline flowing through my veins - it was my first IV.
And before you ask: no, the bubbles don’t tickle. (1/n)
@ben_j_todd if you do this for long enough, you'll filter out most of the bad decisions and get a better understanding of your problem. It's only annoying because it has friction and a need to not fall for the agreeableness of LLMs
@unconvAI Will there be another opportunity to submit a proposal? I'm working on a model where dynamical systems are the backbone of computation, with validation done, and even a working prototype of embodied goal-directed closed-loop agent. Source code available at https://t.co/akgutUXKWE
@finalsparkai Digital twins are super important for NeuroAI and drug research! I'm working on https://t.co/58cRZhWoE7 - C. elegans digital twin embodied closed-loop simulation with bio plausible SNN model (PAULA). One instance of it is live at https://t.co/k32fPwUxRG. Happy to contribute!
It went beyond any expectations. In a single day, it built the firmware, testing framework, harness that can be reused for other AI agents to work with Cardputer and even a virtual display that I can use to view what's on my Cardputer's screen using a phone or desktop 🫨
Let's see how good @OpenAI 's Codex is at developing @M5Stack ROM for Cardputer from scratch, end to end, autonomously. Turning Cardputer into an always-on device for a self-evolving life memory system.
Now I have an always-on recorder that lets me capture things that I say out loud but don't write down, store them, and upload to Exocortex, my personal memory system that processes audio and puts alongside all other data in a structured way. No more forgetting to press "Record"!
Let's see how good @OpenAI 's Codex is at developing @M5Stack ROM for Cardputer from scratch, end to end, autonomously. Turning Cardputer into an always-on device for a self-evolving life memory system.