I have paid version of GPT now. Kinda accidental, I wanted the API version and that seemed to be smth completely else. Didn't know. I have a lot of opinions / mixed feelings about it so far. Take the imagine generation, this image is kinda perfectly what I wanted. But it took so much iterations and it was forgetting what i've said lines before. Everytime adding text balloons why I said to remove it, then it removes it but generates a complete new image with something else I don't want with it. THan i fix that, and tadaaa, again a text balloon.

This is the question where it is about: is GPT in current state worth the money? I have no idea, I need some more time.

Version two.

Is that Gerry Adams or Colonel Sanders?

@donkulator would be fucked up if he has chosen some famous person i don't know., The changes are there ofc.

Depends on what you'll be using it for. Idk, I like it. Code generation is incomparably shittier than claude, but claude has those annoying limits. Chatgpt code is.. Well.. Workable. Meaning you can work on it to make it fit your bill. Ie not a complete nonsense.

Images.. Meeh. Exactly as you described. It's a decent challenge to make it generate smth according to your prompt. Sometimes it completely ignores parts of the prompt, esp. where you tell it NOT to do smth.

O1 is cool though. O1-mini is too. Takes longer to respond, but significantly less delerium

Is it worth paying for an image generator that everyone can tell is made by an image generator? I dunno, but I cringe everytime I see a generated image, so dunno if that's worth it

HOLY PONY: THIS WORKED: "Create a docker compose file for the docker file in this repository: https://github.com/[blabla]."

But it did publish the port to the whole world tho, if docker does that, it goes past your firewall. Thank god I know this shit myself :P

And the compose file is a little bit outdated, it has still the 'version'-tag on top which is depricated.

All of this money wasted to attack a superior language with some 90s relic

C was conceived between 1972 and 1973.

The original K&R book was published in 1978.

"90s relic"

Tell me you're a schmuck, without telling me you're a fucking schmuck.

@netikras I notice that api gpt is very fast but not as good as the web version. Also I reach token limits that I don't have on their website. They keep the good stuff for themselves huh

@Tounai language for boys wearing cat ears.

@Tounai whole AI and your pc works on 'that relic'.

@retoor Do I look like someone who wears cat ears?

@Tounai yes, and they look fine on you, no cap fr fr and shit or however Rust devs talk :P

I had some nice time with C# today. It can become my daily driver. More decent than python, faster development than C, has an okay-ish std, not as crazy as Rust, has better std than C++. I don't like C++'s incest with C on certain points. I rewrote partial my favorite python lib to C#. It includes an Sqlite database. Works nice with C#.

@retoor All that books in the back while we know that Rust devs don't read. They're ticktocking the whole day :P

@retoor that’s John Wick doing Rust with cat ears.

@Tounai no silly, it's you. And what a attitude to compare yourself with John Wick btw. This is the "we have John Wick at home. John wick at home:"

For image generation, try ideogram instead. Way more realistic and less generated-looking results.

When it comes to the API, the Assistant API is too slow for any real-time applications. Regular Chat Completion is faster, but not persistent so you need to provide the full context in every prompt = zillions of tokens = darn expensive.

IMHO this tech isn't mature enough for any sensible application just yet. But give it a couple of more years and it will be awesome - just like any new technology, such as cars, computers, batteries or the internet, which were all kind of useless at their early stages. GPT has been around for two years and has learnt to crawl and talk but still, sometimes sucks titties.

@TerriToniAX I noticed everything you say and totally agree. Especially about context. Speed is kinda enough tho. A few seconds is acceptable for users that really want smth. So far, I'm still very impressed by the tech, for me it's weird that people expected so much more. I get more surprised than dissapointed. In a few years if will blast our socks off.

@TerriToniAX I want to upload this using the api and ask questions about it: https://molodetz.nl/retoor/drstats/.... So far, only webgpt is possible to do that. Api gpt complains about too many context. Maybe I can submit a file with gpt api and then ask about the context, still have to try.

@retoor thins thing has my shirt but that’s all.

@Tounai and your beard and your ears.

@Tounai see my comments at bottom regarding accuracy if imagine generation here: https://devrant.com/rants/11646867/...

@retoor I wish my beard was so well kept

@Tounai me too.

@retoor

The file doesn't look ginormous to me. Which API did you try?

@TerriToniAX, GPT-4, the expensive one. I just can't paste the file as context data; I can do it if I upload. Yesterday, I trained a model on Kiki's data. It cost $10, for goodness' sake. That, while file uploads and asking about it is possible too.

@TerriToniAX, I create such statistics with it:

8. MongoDB:

Contributions: 10

Upvotes: 8

Post Length (Total/Average): 1541 total / 154 average.

Upvotes Per Post (Average): 0.8

Observation: A user with moderate activity and slightly under one upvote per post on average.