Winston Chen

A Ph.D. student of @MIMRTL

[Claude @claudeai]

Fable 5 is back.

[Winston Chen @Winston_AI4Med]

Fable 5 快到期了，这两天多玩了一下。

发现个小技巧：代码、ML 或论文任务里只要带生物医学类别标签，Fable 5 有时会格外谨慎。

我现在先把需求丢给 Opus，让它按 Fable 5 的口味重写 Prompt，并保持原意。

有时候 AI 最有用的地方，是帮另一个 AI 听懂问题。

#Fable5 #Opus #AI

[Winston Chen @Winston_AI4Med]

3/3

Tested in two workflows:

• Idea → codebase construction
• Review → manuscript → submission

Opus becomes the translator.
Fable 5 becomes the executor.

Sometimes the best use of one AI is helping another AI understand the task.

Enjoy the new model

#PromptEngineering

[Winston Chen @Winston_AI4Med]

Opus rewrites. Fable 5 executes.

1/3

Fable 5 is great, but biomedical datasets, medical class names, biology terms, or bio-med article-review content can sometimes trigger unnecessary friction.

My workaround:

Use Opus as a prompt translator.

#Fable5 #AI #LLM

[Winston Chen @Winston_AI4Med]

2/3

I tell Opus what I want to do, say the target model is Fable 5, and ask it to rewrite the prompt while preserving the original intent.

The task is still normal coding, ML engineering, or manuscript-review work.

Only the wording becomes easier for Fable 5 to process.

[Maciej A. Mazurowski @MazurowskiPhD]

Does Segment Anything Model (SAM) really work on medical images?

We tested it on 11 datasets and wrote a paper:

https://t.co/cYaNOSSpRY

In brief: Performance varies widely across different datasets from impressive (given zero-shot setup) to poor.

Details follow. 🧵1/14.

[Alan McMillan @alan_b_mcmillan]

Had a chance to give a talk today at the Practical Big Data Workshop about robustness in medical imaging AI! https://t.co/YYVLMSZycb #deeplearning #radiology #robustness

[Winston Chen @Winston_AI4Med]

@angryfermion Very cool stick!

[MIMRTL @mimrtl]

Thrilled to share a video by the @NLM_NIH (National Library of Medicine) highlighting our team's work toward advancing robustness for AI in medical imaging. Honored to contribute to this transformative field! 🏥🤖📚 #AI #NIH #Radiology @mimrtl

https://t.co/HaBeLtbYx8

[AK @_akhaliq]

BREAKING

OpenAI released a implementation of Consistency Models

consistency models, a new family of generative models that achieve high sample quality without adversarial training. They support fast one-step generation by design, while still allowing for few-step sampling to trade compute for sample quality

github: https://t.co/75WYZhxKQo

[Winston Chen @Winston_AI4Med]

People can be a discriminator instead of a generator with the help of LLMs.

[Gautam Kamath @thegautamkamath]

My wife complained I don't give her enough attention.

I sent her this picture.

I'm sleeping on the couch tonight.

[Kangwook Lee @Kangwook_Lee]

Compilers made low-level programming obsolete and enabled high-level programming, but now large language models are revolutionizing coding by writing code from natural language. Will LLMs become the ultimate compiler 😮? #GPT #ChatGPT #LLM #programming

[UW Radiology @UWiscRadiology]

#Congratulations to @Prof_TimStick, Dr. Fred Lee, @TyBradshaw11, @alan_b_mcmillan, and Dr. Daiki Tamada for being awarded patents in the last month!

[Winston Chen @Winston_AI4Med]

@m_torresvelazq @UWMadison_BME @UWMadEngr @UWMadison @uprm Congratulations!

[Alan McMillan @alan_b_mcmillan]

Congrats to @MIMRTL Undergraduate Research Scholars (Camilla, Chengnan, Paridhi) as they presented their work at the 24th UW Undergraduate Symposium on April 29th, 2022! #undergraduateresearch #deeplearning #radiology #academictwitter

[Tyler Bradshaw, PhD @TyBradshaw11]

Great panel discussion today from industry representatives, moderated by @JoyitaDutta at the @SNM_MI #AI Summit.

[Radiology: Artificial Intelligence @Radiology_AI]

#Superresolution on 3D brain MRI sequences with 45% scan time reduction https://t.co/fxFRsa9irx @UCSFimaging @RadRudie @GregZ_MD #brainMRI #ML #MachineLearning

[MIMRTL @mimrtl]

Last week, @mimrtl team members headed up to the bowling alley for some fun and friendly competition! #bowling #UWMadison @UWiscRadiology @UWMadison_BME

[Winston Chen @Winston_AI4Med]

@arjunreddy2613 I am a PhD student in USA. I am working on multi-modality but abou medical images using Transformer. Am I qualified?