The only thing I’ve found them actually useful for is generating random lists for my D&D games.

When it comes down to needing some mundane descriptions, its great having an LLM brainstorm for you. “Give me 10 examples of weird things I might see in jars in a witch’s hut.” This works well because you can just cut the 5 you don’t like and use the other 5 to brainstorm your final list.

I feed it TOS, Service Agreements, etc and have it simplify and summarize them so i can have a general idea of what is in them without 10 minutes of reading.

Brainstorming ideas; it’s something to bounce ideas off and see what can be tweaked.

Single-player D&D. Can setup multiple players/characters and a DM and just play D&D by myself, which is rad.

Having conversations with fictional characters. Like Data from Star Trek.

None. I avoid them when ever possible which thus far is most of the time.

Among other things: Cooking. They’re really helpful in those situations where I have a bunch of ingredients lying around in my pantry but I lack concrete recipes that can make a proper meal out of them.

I mostly just use them to help write Python scripts.

Absolutely nothing, because they all give fucking useless results. Hallucinates, is confidently wrong, and isn’t even grammatically competent (depending on the model). Not even good for a draft, because I’d have to completely rewrite it anyway.

LLMs are only as good as the guys training it (who are mostly morons), and the raw data they train on (which is mostly unaudited random shit).

And that’s just regular language. Coding? Hah!

Me: Generate some code to [do a thing].
LLM: [Gives me code]
Me: [Some part] didnt work.
LLM: Try [this] instead.
Me: That didn’t work either.
LLM: Try [the first thing] again.
Me: … that still doesn’t work…
LLM: Oh, sorry. Try [the second thing again].
Me: …

Loop continues forever.

One time I found out about a built-in function that I didn’t know about (in LLM generated code that didn’t work), and read the manual for it, and rewrote the code from scratch to get it working. Literally the only useful thing it ever gave me was a single word (that it probably found on Superuser or StackExchange in the first place).

porn

Nothing. Remember that bartender from the original Star Wars?

Nothing because it sucks and isn’t worth the effort if you have an Internet connection and any knowledge on how to use a search engine.

I have a GUI to interface with locally-run image generators that I sometimes use when I need art for D&D.

Instead of reading a manual file for the badrillionth time, I ask it how a shell command should be formated. If it is easy, it gets it right away and I say “oh, yea, that’s right”. If it is hard, I still get a starting point and can correct it fairly quickly. I ask it for translations when learning a new language, which I’ve been doing lately. This it excel at. Even languages that conventional machine translation fails at. I asked chatgpt for Minecraft blocks with some specific set of desirable redstone properties that I didn’t want to dig through a wiki to find. This one had varying success. It is not aware of every odd redstone secret, but it can spit out something useful if you are lucky. I had a quick poem made for one of my rp characters. We had a 5 minute break and I wanted something that made sense for the next scene. Some quick directions to the LLM and a little shoveling paragraphs around and there you go.

I also have tried some light rp with the ai for entertainment. I tried merging harry potter and star trek once. It was mildly entertaining.

If you know how they are dumb and where they kinda work, you can get stuff done. Especially if the answers are easily verifiable. That about summs up how I use them.

I occasionally use LLMs to generate a set of characters for a TTRPG, if I don’t have the time to prepare and/or know we’ll play a very limited scenario in terms of who my PCs are able to meet. This is especially true for oneshots where I just don’t want to put too much work in.

I recently built a scenario for a cthulhu themed scenario, that was set in a 1920s Louisiana prison and planned for two to three sessions. I just had an LLM do a list of all the prisoners and guards on the PCs block, with a few notes on the NPCs look and character. This drastically reduced the time I had to put in preparing the scenario.

Sometimes I have it re-write emails for me if I feel like my tone is off. Once in a blue moon I’ll use an LLM to write a trivial AHK script. I also use AI art bots for throwaway character art for NPCs in my D&D campaign.

Beyond that, almost never.

I use Chatgpt 3.5 both personally and at work for tip of the tongue questions, especially when I can’t think of a word. Sometimes as a starting point when I have trouble finding the answer to a question in Google. It can sometimes find an old movie that I vaguely remember based on my very poor descriptions too.

For example: “what is the word for a sample of a species which is used to define the species” - tip of the tongue, holotype. “What is the block size for LTO-9 tape” - wasn’t getting a clear answer from forums and IBM documentation is kind of behind a wall, needed Chatgpt to realize there was no single block size for tape.

It’s excellent for difficult to search things that can be quickly verified once you have an answer (important step, as it will give you garbage rather than say it doesn’t know something).