How ChatGPT was secretly designed to suck at real work

[@yashghatti]

The reason it still sucks at serious tasks is because it has sort of hit the limit of the rich data it was originally trained on. Now it's at the risk of becoming a "Habsburg AI" - AI trained on AI-generated data. Even with its newest models today, it still generates code that has errors in it, loses the context pretty quickly, and sometimes just takes a very roundabout approach to things that a professional SWE would never do

[@spencerbentley8852]

Hmmm, I can sort of see where you get this idea from, but it falls down when you realise that the API can do anything you'd like of it, and you can even get ChatGPT to create the python code to do so. If you want to build a multi-step prompt system, it's maybe 10 mins of work with an API key, the OAI API documentation and a laptop with Python for Windows installed and a couple of libraries.

[@sachanoob]

I am a writer and I use chatgpt extensively

[@realhuman2545]

We can all tell when something has been written by chatgpt. As soon as i can tell i just get the ick from whatever I'm reading. It doesn't feel real at all.

[@daveinpublic]

I used chatgpt to make this reply

[@StefanReich]

@@daveinpublic "Please write a comment reply that states that I used ChatGPT to write the reply" "Certainly!"

[@cryptogenik]

270 Subscribers?

[@aquapixel0]

May I ask what kind of microphone you’re using, because the sound is so good? Manufacture and model

[@VladHoltz]

Thank you! I am using a Sennheiser MKH416.