Jerry Li

i kinda like a thing called ai. also kinda like a thing called ml.

USC Student Researcher @ ALLeGRo Lab

[Jerry Li @jeli_04]

@johntzwei Congrats Dr. Wei 🎓🎉

[Jerry Li @jeli_04]

@ameya_godbole1 @johntzwei @robinomial 🙏 you were a super clutch mentor

[Jerry Li @jeli_04]

@rronak_ @MichaelElabd @QuantumArjun Are there plans on publishing any research?

[Jerry Li @jeli_04]

[3/3] I’m also grateful for the wonderful mentoring by @johntzwei, @ameya_godbole1 and @robinomial! I learned a lot from this project and am really thankful for their guidance throughout. 🎉

[Jerry Li @jeli_04]

[1/3] Excited to finally share what I’ve been working on the past few months! Spiking intentionally contaminates training data to measure test set contamination. We show it can estimate contamination and adjust test scores for a more truthful evaluation.

[Johnny Tian-Zheng Wei @johntzwei]

🧵[1/5] Works on test set contamination focus on detection, but we show *correction* of inflated test scores is possible. https://t.co/7D6lr63d40
Our proposal is to spike the training data and insert some test examples at known rates. The spiked examples are used to calibrate...

[Jerry Li @jeli_04]

[2/3] @johntzwei and I believe spiking opens a new direction for model evaluation for devs/labs. In the age of internet-scale training data, building robust models across a multitude of domains and environments starts with accurately measuring what our current models can truly do

[Johnny Tian-Zheng Wei @johntzwei]

🧵[1/5] Works on test set contamination focus on detection, but we show *correction* of inflated test scores is possible. https://t.co/7D6lr63d40
Our proposal is to spike the training data and insert some test examples at known rates. The spiked examples are used to calibrate...

[Jerry Li @jeli_04]

@R2Cdev_ How is grok losing to 4o 😭

[Jerry Li @jeli_04]

Great to see Thinking Machines taking a slightly different route instead of just trying to compete with the other big players in the LLM space (Meta).

And what’s even better is a technical report with details on their architecture! Was bearish before on TM but more excited now.

[Thinking Machines @thinkymachines]

People talk, listen, watch, think, and collaborate at the same time, in real time. We've designed an AI that works with people the same way.

We share our approach, early results, and a quick look at our model in action.

https://t.co/AFJZ5kH7Ku

[Thinking Machines @thinkymachines]

People talk, listen, watch, think, and collaborate at the same time, in real time. We've designed an AI that works with people the same way.

We share our approach, early results, and a quick look at our model in action.

https://t.co/AFJZ5kH7Ku

[Jerry Li @jeli_04]

They weren’t lying 💀

[Alex Albert @alexalbert__]

We released Claude Opus 4.6 just two months ago. Today we're sharing some info on our new model, Claude Mythos Preview.

[Jerry Li @jeli_04]

@SenSanders Unreal 😂

[Jerry Li @jeli_04]

@tunguz Static weights won’t enable dynamic thinking. However for this benchmark the context given to the LLM is key.

[Jerry Li @jeli_04]

@fchollet But doesn’t that just mean a better way of evaluating is giving context of the docs as well? Hence coding agents

[Zhikai Zhang @Zhikai273]

🎾Introducing LATENT: Learning Athletic Humanoid Tennis Skills from Imperfect Human Motion Data

Dynamic movements, agile whole-body coordination, and rapid reactions. A step toward athletic humanoid sports skills.

Project: https://t.co/MFy2NIOsrn
Code: https://t.co/A7B5H8PIBh

[Jerry Li @jeli_04]

@awnihannun Contextual token compaction is pretty good but in the long run it’ll always be capped by whatever the models capabilities are. More people should probably be focused on figuring how to do local updates within the model weights without being overly expensive like full backprop

[hardmaru @hardmaru]

Instead of forcing models to hold everything in an active context window, we can use hypernetworks to instantly compile documents and tasks directly into the model's weights. A step towards giving language models durable memory and fast adaptation.

Blog: https://t.co/iHoifpsLMu

[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.

[OpenAI @OpenAI]

We’ve developed a new way to train small AI models with internal mechanisms that are easier for humans to understand.

Language models like the ones behind ChatGPT have complex, sometimes surprising structures, and we don’t yet fully understand how they work.

This approach helps us begin to close that gap.
https://t.co/g4zOcdezPU

[Jerry Li @jeli_04]

No benchmark can convince me that Gemini is good