284
That introduction...
(mander.xyz)
A place for majestic STEMLORD peacocking, as well as memes about the realities of working in a lab.
Rules
This is a science community. We use the Dawkins definition of meme.
Took me a second.
But man, I don't write academic papers anymore, but I have to write a lot of reports and such for my work and I've tried to use different LLM's to help and almost always the biggest help is just in making me go "Man, this sucks, it should be more like this." and then I proceed to just write the whole thing with the slight advantage of knowing what a badly written version looks like.
That's basically Classifier-free guidance for LLMs! It basically takes an additional prompt and says "not this. Don't do this. In fact, never come near this shit in general. Ew." And pushes the output closer to the original prompt by using the "not this" as a reference to avoid.
So real
My favorite was when it kept summarizing too much. I then told it to include all of my points, which it mostly just ignored. I finally figured out it was keeping under its own cap of 5000 words in a response.
I've had the reverse issue, where I wanted to input a large amount of text for ChatGPT to work with. Tried to do a workaround where part of my prompt was that I was going to give it more information in parts. No matter how I phrased things it would always try to start working with whatever I gave it with the first prompt so I just gave up and did it myself.