Brian Cheung

This is my hat, there are many like it, but this one is mine.
Faculty @ UCSF
ex: @MIT_CSAIL 🧢 @berkeley_ai 🎓 Google B̶r̶a̶i̶n̶ DeepMind 🎩

[Brian Cheung @thisismyhat]

The Platonic Representation Hypothesis
https://t.co/eoz1GTBEiU

Surprising (?) results:

- Pure vision models align with pure text models as scale increases.
- This alignment correlates with better downstream performance.

Fun work with @minyoung_huh @TongzhouWang @phillip_isola

[Sophie Wang @SophieLWang]

"The Truth Lies Somewhere in the Middle (of the Generated Tokens)"

In autoregressive language models, mean pooling hidden states across generation yields better representations than any token alone.

project page: https://t.co/kXddYUir4k
w/ @phillip_isola and @thisismyhat

[Badr AlKhamissi @bkhmsi]

🎉 Re-Align is back for its 4th edition at ICLR 2026!

📣 We invite submissions on representational alignment, spanning ML, Neuroscience, CogSci, and related fields.

📝 Tracks: Short (≤5p), Long (≤10p), Challenge (blog)

⏰ Feb 5, 2026 for papers

🔗 https://t.co/BEtKUM9oQP

[Brian Cheung @thisismyhat]

There are self-play environments everywhere for those with the eyes to see.

[Yuxiang Wei @YuxiangWei9]

Software agents can self-improve via self-play RL

Introducing Self-play SWE-RL (SSR): training a single LLM agent to self-play between bug-injection and bug-repair, grounded in real-world repositories, no human-labeled issues or tests. 🧵

[Kush Tiwary @ktiwarylab]

👁️🌋 Our new @ScienceAdvances paper: We replayed the Cambrian explosion of vision by evolving AI agents inside a physics engine to understand the principles that shape visual intelligence.

We believe that this is a promising way to do AI for science and build new forms of AI by computationally mimicking biological design principles of evolution and learning.

Website: https://t.co/xd1Hkt8bvs and paper links at the end of this thread 👇

[Brian Cheung @thisismyhat]

@phillip_isola An old model's trash is a future model's treasure!

[Brian Cheung @thisismyhat]

There's no such thing as bad data, only bad learning.

[Physical Intelligence @physical_int]

We discovered an emergent property of VLAs like π0/π0.5/π0.6: as we scale up pre-training, the model learns to align human videos and robot data!

This gives us a simple way to leverage human videos. Once π0.5 knows how to control robots, it can naturally learn from human video.

[Brian Cheung @thisismyhat]

@KyleStachowicz @chris_j_paxton Then you've learned to not buy from then again. 🙂

[Brian Cheung @thisismyhat]

With a sufficient model of the world, any data can be useful data

[Physical Intelligence @physical_int]

This also shows up in the representations learned by the model. We plot the model’s representations of human and robot images. As pre-training is scaled up, the representation of humans and robots become more aligned: to a scaled-up model, human videos "look" like robot demos.

[Brian Cheung @thisismyhat]

@chris_j_paxton Data is just information. There's always something to learn from data, maybe not what you need right now. But can become important later.

[Brian Cheung @thisismyhat]

@JieWang_ZJUI The human data was originally out of domain and unusable becomes useful after a certain scale of model capability https://t.co/wIa62hyNG5

[Physical Intelligence @physical_int]

This also shows up in the representations learned by the model. We plot the model’s representations of human and robot images. As pre-training is scaled up, the representation of humans and robots become more aligned: to a scaled-up model, human videos "look" like robot demos.

[Phillip Isola @phillip_isola]

Impromptu NeurIPS meetup: "representational convergence by the beach." We will meet at ballroom 20c (near lunch) 2pm Fri and walk over to Marina. Will chat about platonic reps, fractured reps, or anything else about where all these models are heading.

Anyone is welcome to join!

[Brian Cheung @thisismyhat]

In all this craziness, let's take a moment and enjoy the three accounts @ilyasut follows.

[Phillip Isola @phillip_isola]

This paper is really interesting to me -- it shows substantially stronger representational convergence than previously measured!

In the PRH we found ~0.2 mknn alignment between vision and text models. This new paper reaches ~0.4. Challenge: find a setting where it reaches ~1.0.

[Brian Cheung @thisismyhat]

The one time you get quoted about your research philosophy and then get described as an engineer 🤣@VentureBeat

[Sakana AI @SakanaAILabs]

Sakana AI’s CTO says he’s ‘absolutely sick’ of transformers, the tech that powers every major AI model

“You should only do the research that wouldn’t happen if you weren’t doing it.” (@thisismyhat) 🧠

@YesThisIsLion
https://t.co/cGdHONcqDV

[Aritra @ariG23498]

Read and reproduced this paper in a free tier colab notebook. 🔥

[Phillip Isola @phillip_isola]

Over the past year, my lab has been working on fleshing out theory/applications of the Platonic Representation Hypothesis.

Today I want to share two new works on this topic:

Eliciting higher alignment: https://t.co/KY4fjNeCBd
Unpaired rep learning: https://t.co/vJTMoyJj5J

1/9

[Brian Cheung @thisismyhat]

@jasonfurman Prediction markets don't just predict, they can influence outcomes (e.g. 'the fix is in', reflexivity, etc)

[Brian Cheung @thisismyhat]

So going back to the original takeaway, much like what @sama, @ericjang11, and @noampomsky have said in the past:

You can just ask for things.

Something interesting might happen.
Paper: https://t.co/4bgbJagRH3
Code: https://t.co/VylkC7oOg8
Website: https://t.co/GG6F8fLIIm

3/3

[Brian Cheung @thisismyhat]

A takeaway I learned from LLMs:

You can just ask for things.

What if you asked a language model to imagine senses it never experienced?

@SophieLWang , @phillip_isola and I asked language models to "Imagine seeing..." and "Imagine hearing...".

1/3

[Brian Cheung @thisismyhat]

It turns out, a simple cue like asking the model to ‘see’ or ‘hear’ can push a purely text-trained language model
towards the representations of purely image-trained or purely-audio trained encoders.

2/3