Something fishy going on

[tender crest]

When chatgpt 3.5 was first introduced I remember using it to code indicators on tradingview using the pinescript language. It has basically no issue following my instructions and making even complex indicator. I have indicators on indicators that were solely made by chatgpt at the time, some really good that were praised by community even.

Forward about a year to now, 3.5 can't even finish the most basic of pinescript codes, not only that, but 4.5 is barely any better. Its as if they purposely made the models dumber because they were to powerful.

I have not only noticed this when it came to coding, but just in general, the models are a lot dumber than what they were when released.

Canceled my membership because the main reason I had pro was to help me with coding, but im better off writing the code alone at this point... what a downgrade. Hopefully devin will be better.

[red steppe]

GPT-4o is dumber than original GPT-4, it struggles to keep following prompt instructions and to understand some instructions that in the past was followed with no issue.

I suspect they have some way to config how smart their models are probably to control how much computation they consume. So when they release a new model they give it more resources to showcase its capacity then months later they dumb it to reduce costs.

It's sad they marketed Plus plan as we'd receive all features and models first, now they keep making the service worse for us. I guess Pro users just use o1-mini and never use 4o.