Armin Bazarjani

PhD Student @USC

[Just Jack @7Veritas4]

Hard to believe Spencer Pratt lost the election despite having no experience, being highly unpopular and running in a 4:1 democratic stronghold.

Makes no sense.

[Neuron @NeuroCellPress]

What can brains teach machines & machines teach brains?
We launch the first of a three-part special issue series on neuroscience & AI, featuring reviews on embodied intelligence and the schema spectrum, plus Q&As w/ leaders in the field. https://t.co/KhMnhEGf12

[hardmaru @hardmaru]

For over a decade, we’ve accepted that end-to-end backprop is the only way to train deep networks. But holding the entire network in memory all at once is why AI training is hitting a resource wall.

We found a new way to break the network into blocks and train them independently. The trick? Treating the network’s forward pass like a diffusion model denoising a signal.

This reinterpretation slashes the memory needed to train deep models. In our #ICLR2026 paper (https://t.co/PK5h0mqQSo), we matched end-to-end performance across ViTs, DiTs, and LLMs. We did this while training just one isolated block at a time.

[Drew Schreiner @SchreinerDrew]

Where, exactly, does learning happen in the brain?

Out now @Nature , we identify a synaptic locus of birdsong learning and show that the circuit can be tuned to make birds learn faster, but at a cost.🧵 #neuroscience
https://t.co/mS6EJUPVa2
Open link: https://t.co/wiJ16guRj1

[james hawkins @james406]

AI just saved me ~4 hours of driving on my vacation

i was planning a trip to the Grand Canyon and noticed the driving route was incredibly inefficient

i spun up a few claude code agents and used agentic ai to find this faster route

it will take me just a few minutes to drive, compared to 4 hours

if anyone has contacts at the US government lmk - happy to relay this to them

[Ali Behrouz @behrouz_ali]

We keep scaling model parameters by increasing width and stacking more layers, but what if the truly missing axes for continual learning are compression and stacking the learning process?

Excited to share the full version of Nested Learning, a new paradigm for continual learning and machine learning in general.

Paper: https://t.co/75T93mvwKm

[Yashar Ali 🐘 @yashar]

BREAKING - IRAN UPDATE

HRANA’s latest numbers.

Reminder: HRANA death confirmations take time as they go through multiple layers of confirmation.

Total confirmed fatalities: 3,919
Deaths under investigation: 8,949

Total: 12,868

Protesters: 3,685
Children under 18: 25
Military and regime forces: 178
Non-protesting civilians: 31

Severely injured: 2,109

Arrested: 24,669

Broadcast forced confessions: 145 cases

[kev joon @never_oppressed]

Reza Pahlavi’s daughter is posting messages on IG about how people in Iran are killing themselves because they don’t think Trump is going to attack. Incredible stuff.

[Yashar Ali 🐘 @yashar]

UPDATE on IRAN

HRANA’s latest numbers.

This is absolutely horrific.

Reminder: HRANA death confirmations take time as they go through multiple layers of confirmation.

Confirmed fatalities: 3,308 people
Deaths under investigation: 4,382 cases
Protesters: 3,097
Children under 18: 22
Military and regime forces: 166
Non-protesting civilians: 23

Severely injured: 2,107 people

Arrested individuals: 24,266 people

Broadcast forced confessions: 132 cases

[Nature Human Behaviour @NatureHumBehav]

In this study, Piray shows a problem of low statistical power in many studies that use Bayesian model selection with computational modelling in psychology and neuroscience.
https://t.co/eCkStA1JIw

[Danijar Hafner @danijarh]

Excited to introduce Dreamer 4, an agent that learns to solve complex control tasks entirely inside of its scalable world model! 🌎🤖

Dreamer 4 pushes the frontier of world model accuracy, speed, and learning complex tasks from offline datasets.

co-led with @wilson1yan

[Annabelle Singer @DrACSinger]

Check out our new paper in @Nature “Goal specific hippocampal inhibition gates learning”: https://t.co/k5eG9SZYo8
By @NuriJeong, Xiao Zheng, Abby Paulson, @StephmPrince and colleagues.

[Payam Piray @payampiray]

New paper with @nathanieldaw in Nature Communications: an RL model that builds a successor map compositionally: it plans as well as the best models, and links components of the map used for planning to neural codes in the medial entorhinal cortex.
https://t.co/EjiCMi7qwa

[Armin Bazarjani @arminbazza]

@Ikuperwajs @evanrussek @TrendsCognSci @marcelomattar @weijima01 @cocosci_lab Looks interesting!

[Armin Bazarjani @arminbazza]

@chingfang17 @KanakaRajanPhD Looks cool!

[Marlos C. Machado @MarlosCMachado]

📢 I'm happy to share the preprint: _Reward-Aware Proto-Representations in Reinforcement Learning_ ‼️

My PhD student, Hon Tik Tse, led this work, and my MSc student, Siddarth Chandrasekar, assisted us.

https://t.co/RQ65fuTCda

Basically, it's the SR with rewards. See below 👇

[Alex Kendall @alexgkendall]

Here's my favourite scenario from autonomous driving in Tokyo last week! 🇯🇵

[Fight for a Union @FightForAUnion]

THIS. #HandsOff

[François Chollet @fchollet]

Much of the field obsesses over end-to-end learning. But strong generalization requires compositionality: building modular, reusable abstractions, and reassembling them on the fly when faced with novelty.

The models of the future won't be just pipes, they will be Lego castles.

[Nature Neuroscience @NatureNeuro]

A model of compositional state spaces in the #hippocampus shows latent learning and rapid generalisation, and predicts the emergence of place responses in replay – which is discovered empirically in an existing dataset

@BakermansJJW

https://t.co/RKtOUu5xf7