I taught myself to code just to build a prototype of a couple of products. ChatGPT seems to be good enough for figuring out a lot of functions and digging through documentation. I know that some of the code might be off and likely doesn't follow best practices, but my goal is just to get the prototype up and running.
I've learned enough that I can put it all together. Learning any more will mean going down the full-time dev route, which, while enticing, is something I don't have enough time for.
For me, chatGPT is simply a good replacement for {stackexchange, reddit, wikipedia, github, google,bing, etc.}. All information is accessible in one place and it's easy to ask about anything and get a structured answer. My topics of interest are advanced math, philosophy of physics, and modern semiconductors technology.
GPT is amazing as a stackexchange, et. el. replacement. Just earlier today I used GPT to answer a really simple question which I Google failed to answer. I just wanted to programmatically change a setting in a Kubernetes deployment, but I really didn't want to take the time to figure out the insane JSON nesting.
This is so worth a subscription fee.
Please write a kubectl command to correctly patch the imagePullPolicy to be equal to Always.
kubectl patch deployment <deployment-name> --patch '{"spec":{"template":{"spec":{"containers":[{"name":"<container-name>","imagePullPolicy":"Always"}]}}}}'
My main use cases are: d&d prep, help re-writing important communications, some programming stuff. I'm sure I would use it even more if I got faster responses and I was paying for it.
If you're comfortable sharing more I'd love to hear it. I think this hits the sweet spot of my nostalgia for playing D&D and this being the first application that doesn't feel like an existential threat in some way :)
It is great for names:
-Provide me with some names for a Frost Giant Clan.
-Provide me with some names for a elf thieves guild.
Last session I had a corrupted unicorn that the players need to subdue to cleanse her. So I asked chatGPT for possible rituals that could be used to cleanse the corruption.
I also use it to come up examples of cons that npc's might try and pull and most recently a bunch of taunts that pixie's would sling at the group as they crossed the Whispering Woods.
It's great for doing easy work and for bouncing ideas off of to shore up storylines or to spark creativity.
And the biggest use case, flavor box text.
- Describe a musty library with water damage.
- Describe the players entering a peacefully glade that lets them know they can relax.
i'm learning some foreign languages, and like to read newspaper articles in the target language. chatgpt is great at translating, and, what is really killer, i can ask follow-up questions to explain acronyms, references to people / institutions, etc. really just absolutely amazing. can also ask it to break down compound words (quite handy for german!)
also great for programming/script questions. e.g., "write a git-push hook that errors out if a commit contains 'DO NOT PUSH'". i could figure that stuff out on my own, but it's way faster to have ChatGPT get me started. or for unity questions -- somehow chatgpt knows way more about unity than unity's own docs. :-)
finally, i just find it useful for random questions. e.g., "recommend a good, cheap kitchen knife". "i have X, Y, Z ingredients, what's something tasty i can make?"
it's honestly amazing.
i do hit the content filter quite a lot tho (w/ innocuous stuff, i'm not even trying to do anything spicy!) -- i defnly wish premium would have a mode to turn that off, or at least turn it way down.
if they persist in this, i'd assume over time some rival w/o the filter will eat their lunch. (unless such filters get legislated -- which is certainly a not-unlikely possibility.)
I wonder if it would make sense to re-train an AI like ChatGPT on a large internal code base and then ask it to write new modules in the style of existing modules. If the code it produces is only 30% correct, it would still save a ridiculous amount of time.
Where are you seeing pricing information? $20/month still feels pretty steep for someone not using it constantly. I was really hoping they'd have a pay-per-query or something like it similar to DALL-E
Each dalle query is hugely expensive at ~$0.076 . They must have a 1000x markup on the incremental costs. Stable Diffusion is free, locally installable, and you can generate 1000 custom images in an hour.
I'm looking forward to LLMs you can host on your local machine.
I'm using it for tech research and onboarding. It's just so convenient to ask "how to request github API with Go req library" and get a working answer even if it's not perfect.
Novelty questions are awesome too. I'm studying philosophy as a hobby and it's surprisingly accurate at answering humanitarian questions which were always such a struggle for other assistants like Google search etc.
I'm definitely paying 20$ though 42$ would be a tough call.
If you are willing to pay for this service, why? What are your use cases?