gangqing hu

[gangqing hu @hugangqing]

Scientific figures interpreted by ChatGPT: strengths in plot recognition and limits in color perception https://t.co/prhdtXE7pw

[Dong Xu @DongXu10710609]

Iterative Prompt Refinement for Mining Gene Relationships from ChatGPT https://t.co/FUy6QdnrhX

[gangqing hu @hugangqing]

Also not good at telling positional relationships among visual elements from complex figures. Nor counting of visual elements. "infusing chatbots with domain-specific expertise and image-input-specific prompts will be pivotal in enhancing the quality of their responses."

[Steven Ge @StevenXGe]

Can GPT-4 Vision understand scientific figures like this?
An early test from @hugangqing's lab found it can "proficiently explain different plot types and apply biological knowledge to enrich interpretations." Struggles with color-coded groups.
https://t.co/kO7zBqXg9r #gpt4v

[Stephen Turner 🦋 @stephenturner.us @strnr]

Bioinformatics Illustrations Decoded by ChatGPT https://t.co/Ur4aBOPW09

[bioRxiv Bioinfo @biorxiv_bioinfo]

Bioinformatics Illustrations Decoded by ChatGPT: The Good, The Bad, and The Ugly https://t.co/al14dGq9Hg #biorxiv_bioinfo

[bioRxiv @biorxivpreprint]

Bioinformatics Illustrations Decoded by ChatGPT: The Good, The Bad, and The Ugly https://t.co/V6a0OVXya6 #bioRxiv

[Alexander Schubert @alexschbrt]

Have the feeling that computational biology roles will soon be oversaturated.

Helping a company with recruiting for a few scientific roles and we receive > 10x applicants for computational biologists than for any other position.

[gangqing hu @hugangqing]

@alexschbrt The entering bar to bioinformatics and computational biology is getting low than ever before, especially for those who knows biology. You know, chatbot is one of the the driving forces that make coding more accessible to ALL.

[Dong Xu @DongXu10710609]

S-PLM: Structure-aware Protein Language Model via Contrastive Learning between Sequence and Structure https://t.co/a4KED49KpH

[Steven Ge @StevenXGe]

A new blog post with my own grant proposals as examples.
A good proposal needs to convince reviewers to invest in your research. As investors, funders are often risk-averse, favoring well-defined, tightly focused projects rather than ambitious ones clouded with uncertainties.
https://t.co/r0vKebfjmF

[gangqing hu @hugangqing]

@StevenXGe How about using GPT to rate the three versions?

[gangqing hu @hugangqing]

#ChatGPT has an amazing feature: multi-role-playing, meaning that it can role play different experts in one chat session. Isn’t it cool to use it to tackle interdisciplinary problem? We tested this feature on chatbot applications in sports rehab.

[medRxiv @medrxivpreprint]

Interdisciplinary Inquiry via PanelGPT: Application to Explore Chatbot Application in Sports Rehabilitation https://t.co/xp7cABYiIb #medRxiv

[Steven Ge @StevenXGe]

The latest "upgrade" to GPT-4 significantly degrades its performance, dropping from ~90% to ~5% in some tests. You can use the old (3/14) version in the "Playground" at the OpenAI website, or through API. I am switching https://t.co/7dsucYepH1 app to this GPT-4 model.

[medRxiv @medrxivpreprint]

Interdisciplinary Inquiry via PanelGPT: Application to Explore Chatbot Application in Sports Rehabilitation https://t.co/xp7cABYiIb #medRxiv

[Steven Ge @StevenXGe]

Introducing https://t.co/T4xuk6r3YA, an app (like Code Interpreter) that can greatly simplify data analysis workflows. Based on the tree-of-thoughts prompting strategy, it breaks down problems into smaller tasks, each solved through trial and error. Video: https://t.co/g71jat7lMR

Other features:
- Supports both R & Python
- Upload multiple files of any type
- Download report & code for reproducibility
- Use GPT-4!
- Have ChatGPT write your prompts

Freely available, https://t.co/7uS9mKC5KD uses both R and Python with 16,000 packages installed. It can access the internet and be run locally (later) for better data security. It is a general purpose platform for analyzing data of any type. All you have to do is ask!

Accessible to paid subscribers, Code Interpreter writes and runs Python code to analyze data. But the sandbox only has 343 Python packages pre-installed and lacks internet access.

Example reports:
- Classifier: https://t.co/10KyjWOisq
- Bioinformatics: https://t.co/gJcfhTiOyv
- Finance: https://t.co/Pb6fnhzonn

[gangqing hu @hugangqing]

@ETulowetzke @StevenXGe Not sure. That will be more on privacy and security things.

[Steven Ge @StevenXGe]

Code Interpreter for Bioinformatics: Are We There Yet?
By @hugangqing, me & others
We tested this new #ChatGPT feature:
- Great for manipulating and visualizing data
- Cannot download annotation or install packages
- Concerns on storage, security, etc.
https://t.co/QtLNOxoJmC

[Steven Ge @StevenXGe]

Funny that 25 students paid $$ to come to my class. After I yelled at them for 3.5 months, some still do not know what P=0.35 means.
Yet, a free workshop attracts 600. About 220 joined the Zoom. If they learned anything, does that mean I should do more free online lectures?

[Steven Ge @StevenXGe]

Thumbs up for @j_perkel's article on #ChatGPT coding in @Nature!👏 It's a wealth of info! Highly recommended!
It took Jeff months to interview many experts and practitioners (me included). My blogs often take 15 minutes to create.😁
#MustRead #AI https://t.co/ZoIrWXYUPW