As I wait for the SigLIP2 OpenCLIP and timm checkpoints to upload, a few comments. Like v1, these are very strong image encoders, but even better. They should become a go to ViT encoder for many tasks. Yet, even when we had ViT encoders from SigLIP, DFN CLIP, DataComp CLIP,

Fascinating, surprising results

Looking for a paper I remember but can't track down: optimal stopping for deep learning for chest X-ray radiology, multilabel classification with N labels, stops M times (M<N) for M checkpoints. Anyone know it?

🚨Should We Still Pretrain Encoders with Masked Language Modeling? We have recently seen massively trained causal decoders take the lead in embedding benchmarks, surpassing encoders w/ bidirectional attention. We revisit whether Bert-style encoders are a thing of the past? (1/N)

My NYU Center for Data Science colleague, Carlos Fernandez-Granda, released the 700-page textbook «Probability and Statistics for Data Science», where he condenses 10 years of teaching experience at @NYUniversity. 200 exercises, 102 notebooks, 115 videos! 🥳🥳🥳 ps4ds.net

I did not know that, for any random variable x, | mean(x) - median(x) | <= stdev(x) Direct proof:

Hey Sam Altman OpenAI – small idea with big impact: If a prompt comes from an audio transcription, add a system-level note for the model. It would help interpret typical voice-related quirks: misheard names, sentence fragments, course corrections.

you don't need a framework to train models pretraining? for loop SFT? for loop RL? two for loops inside a for loop

this screenshot from GPT-5 livestream has to be among the worst chart crimes of the century

Really not a good sign if Gary Marcus is that gracious

i saw this guy’s video on simulating gravity in C++ and that pushed me into learning OpenGL (failure lol). now he’s back with another one simulating black holes. he incorporated ray tracing, Einstein’s field equation, Schwarzschild solution to the field equation, space-time,

Incredibly important, often overlooked

Both

Every tool is "superhuman", in the specific use cases relative to it. Otherwise we would not use it.