Share this page!

Enter URL or ID to Unroll

You can paste full URL like: https://x.com/threadreaderapp/status/1644127596119195649
or just the ID like: 1644127596119195649

How to get URL link on X (Twitter) App

  1. On the Twitter thread, click on or icon on the bottom
  2. Click again on or Share Via icon
  3. Click on Copy Link to Tweet
  4. Paste it above and click "Unroll Thread"!
  5. More info at Twitter Help
neurosock🧠Brain Chips🦾 Profile picture
neurosock🧠Brain Chips🦾
@neurosock
#BrainChips monthly recap. I make #neuro papers easy to understand. To make #Neuralink possible. Neuro PhD. AI🤖ML👾Data Sci 📊 Monkeys🐵Future🚀Cyberpunk⚡🦾🌌
Dec 11 • 21 tweets • 6 min read
The new Titans architecture shows how badly Google wants to go back to recurrence.

Why? Recurrence is damn cheap.

This paper shows how the brain does it.

Striatum, working with cortex, HPC, BG, and thalamus, in recurrent loops and shortcuts.

Here is my toy model and notes: Image This paper proposes a new way to see the brain's core architecture and compares it to transformers.

The canonical view suggests goal-directed and habitual systems are separate and compete for control.
Oct 30 • 22 tweets • 6 min read
Dopamine ≠ reward but turns out, also not the learning molecule we thought.

If DA RPE is the emperor, this work SCREAMS it was running naked all the time.

This paper got quite some attention recently. Let me explain why it is so relevant.

Here is my toy model and notes: Image The canonical view is DA signals RPE, a "teaching signal" for learning.

This paper argues we were wrong all the time: DA is a "performance signal" that controls the vigor and direction of movement and it is only indirectly related to learning by controlling effort.
Oct 22 • 20 tweets • 6 min read
To plan the future, the PFC represents a step-by-step map of actions, and at every step, this plan moves to the past like a conveyor belt.

This proposes a super simple neural-subspace architecture for planning.

A toy environment can clarify it.

Here is my toy model and notes: Image The old view saw planning as a slow, sequential search, like exploring a maze path by path.

This paper proposes a fast, parallel inference, where the optimal plan emerges all at once.
Dec 17, 2024 • 13 tweets • 5 min read
Since 2023 I put together a month summary of what's going on in Brain Chips.

In 2024, beyond Neuralink, we saw devs. from Meta, Blackrock, ScienceCo, lots of research, and more.

It was a really good year for BCs.

Here the top-10 events this year and what to watch for in 2025: Image #10 Speech Neuro prostheses:

Although the are still Milwaukee to achieve in movement recovery, decoding from speech areas is a next frontier. Now we know even movement can be decided from these areas.

Language in AI and BCs is something to watch on 2025.