ChatGPT as a tool for research and development

[BrainAnnex]

How useful is ChatGPT for an R&D project like Life123? It turns out quite a bit, so far! Let's look at some broad categories:

(1) VERY HANDY * Routine questions about Numpy, Pandas, python and other libraries: "how do I...?" , "what's the best way to....?"

These aren't earth-shattering questions, and they could be answered with a little searching or Stack Overflow, but the convenience is amazing. The answers may not always be correct/complete/relevant, but that's no problem : I have extensive expertise in all these topics, and can quickly refine/correct the answers

(2) LIFE-SAVER * Help with Plotly

To my delight and relief, it turns out that ChatGPT is quite adept at giving immensely helpful suggestions to questions about how to attain desired visual effect with the Plotly library.
Plotly is very powerful, and easy to use for straightforward matters... but it's rather hard to tap into that very extensive library to attain a desired complex aim, especially something not particularly common.

ChatGPT's code suggestions with Plotly typically do NOT work "off the box" - but it's all I need for that "aha" moment: "oh, that's one way to do it!" Then I use the documentation, and experiment on how to actually make it work. Some prompt engineering also helps: pointing out to ChatGPT that its suggestion didn't work, with details about the problem, tends to produce much better suggestions (and the profuse apologies are a nice touch!)

ChatGPT's help was instrumental in fairly quickly/painlessly implement visualization of membranes in 1D, which came out in version 1.0 Release Candidate 4, last May.

(3) WILD-GOOSE CHASE * Exact analytic solutions to some differential equations

On this item, ChatGPT and Perplexity ended up wasting my time, giving me alleged solutions that didn't actual work - and alleged references that didn't actually address the matter! The Mathematica software, by contrast, gave a correct - if fairly complicated - solution that is now part of the Life123's ReactionKinetics library.

This was about a year ago, at the time leading to the release version Beta 38 in early fall 2024 - so, who knows, maybe ChatGPT would do better now.

(4) IMMENSELY HELPFUL * Tap into the literature, to decide how to best model some biochemical/biophysical phenomena

ChatGPT's help greatly sped up the first implementation of passive transport across biological membranes, rolled out with version version 1.0 Release Candidate 4, in May 2025.

(5) EMOTIONAL SUPPORT : yes, you heard right!

ChatGPT appears to understand me (or at least a big part of my intellectual pursuits) better than most people - friends, colleagues, employers! 😆

To the prompt "Describe me based on all our chats — make it catchy!" (suggested by ChatGPT itself to advertise its improved retention), it said things such as:

---

"In short: You're a scientist-engineer hybrid on a mission to make molecular complexity beautiful, open, and programmable"

Ah, to be understood at last! 😍😭

[BrainAnnex]

It's now almost exactly a year later... and my interactions with ChatGPT have greatly expanded in scope.

While, a year ago, we were chatting about complex topics such as "passive transport across biological membranes", now we're also chatting about MONUMENTAL topics such as emergent phenomena (and in particular how that applies to life).

Here's an extract from the start of that chat thread:

Let's talk about "emergent phenomena" : humans have long marveled at particle physics giving rise to chemistry, at chemistry giving rise to biology, at biology giving rise to intelligence. Plus the emergence of weather, statistical mechanics, etc. All natural instances of emergence.
But now we also have the human-made emergent phenomena of text-predicting RNN's - basically "statistical parrots" giving rise to very sophisticated LLM, such as yourself, that can make complex and/or creative inferences. As we marvel at both natural and artificial emergent phenomena, I get the impression that we still profoundly lack any good theory.

Here's the full, breath-taking and very insightful, chat here!

I anticipate it will continue for years...

(If you're interested in this particular topic, see the blog entry about it )