Pedro P. Sanchez (@snchzpedro_) 's Twitter Profile
Pedro P. Sanchez

@snchzpedro_

Building @sinkove_. PhD from @EdinburghUni. Previously @ MSR, Canon Medical, GE Healthcare.

ID: 1094223152639602688

linkhttps://www.ppsanchez.com/ calendar_today09-02-2019 13:15:04

617 Tweet

1,1K Followers

684 Following

Pedro P. Sanchez (@snchzpedro_) 's Twitter Profile Photo

I often talk with ChatGPT about my startup, Sinkove. It loaded everything in memory. If I say the prompt "write a 30-second pitch of my startup", it does it better than me. Crazy.

Pedro P. Sanchez (@snchzpedro_) 's Twitter Profile Photo

Really enjoy the new Sana diffusion model from NVIDIA for two reasons: - efficient compression from the VAE - using decoder-only LLMs for text encoding (it’s time to ditch T5 and Clip text encoders).

Raman Dutt (@ramandutt4) 's Twitter Profile Photo

This was a joint work with Pedro P. Sanchez , Yongcheng (Vincent) Yao, Steven McDonagh, Sotirios (Sotos) Tsaftaris, and Timothy Hospedales ! Everything open-sourced! Paper - arxiv.org/abs/2505.10496 Github - github.com/Raman1121/CheX… Models - huggingface.co/collections/ra… Data - huggingface.co/datasets/raman…

Pedro P. Sanchez (@snchzpedro_) 's Twitter Profile Photo

Moving between Github Copilot and Cursor today. In terms of AI capabilities, it seems very similar. But cursor's speed makes the experience much better. I will try OpenAI's Codex once it comes to the Pro subscription.

Pedro P. Sanchez (@snchzpedro_) 's Twitter Profile Photo

I wrote a quick blog post with Sinkove's vision. We don’t crash planes to test wings. So why risk human bodies? sinkove.com/resources/post… #AI #Simulation #MedTech #radiology

Pedro P. Sanchez (@snchzpedro_) 's Twitter Profile Photo

⭐ New benchmark paper! Can you really trust your synthetic radiology data? Evaluating along multiple dimensions is key: fidelity, privacy and utility! Sinkove contributed to CheXGenBench, a work led by Raman Dutt to benchmark 11 generative AI models architecture on chest