Thread 106190402

It's pretty obvious now what kind of trick OpenAI is pulling with GPT-5

>remove ability to select models from chatgpt
>GPT-"5" automatically routes your prompt to whatever model it thinks can handle it
>it will most likely choose the cheapest model
>80 message limit /3 hours as well

It's all a big fucking scam. OpenAI was bleeding money and all GPT-5 is, is just a fucking router disguised as a new model. I wouldn't be surprised if it's routing my prompts to a GPT-3 model by how dogshit it is.

/ai/ board would be really really really fucking nice desu

It's just chatgpt 4o but trained to score better at benchmarks.
They literally had to change their release date because Alibaba released their LLM that would have scored better than GPT-5.

We run out of data to train it to make it better, so they started to use AI-generated data to improve benchmarks (at the cost of making it worse for real-world use)

>>106190402 (OP)
Wouldn’t shock me if they’re leaning hard on older models to save cash.

>>106190402 (OP)
Add ?model=gpt-4-1-mini to force choose model.
Since it only works once, if you want to continue your chat using this model, keep appending that.

>>106190402 (OP)
I am listening to an educated pajeet at work praising the latest releases.
And when he reached to the new OpenAI models being open source he explained it by "customer pressure to enter the open source space".
Yeah, right.

so how is grok more profitable? I think gpt4 was better than grok for general problem solving. also grok has a simple pricing model (non api)

>>106190402 (OP)
But is GPT-5 mini better than GPT-4o?

>>106190402 (OP)
no complaints

>>106192697
You literally can't know if they are actually routing you to a mini model, they could be routing you to fucking gpt-1 and you'd never be able to actually tell other than it being shitty

>>106191714
Grok is GPT4 stupid, Elon literally just API ChatGPT.

>>106190426
>implying 4chan will ever add new boards
>implying 4chan isn't just a public honeypot at this point

>>106190426
and banning anime posts outside if /a/ aswell

>>106193533
Grok feels far worse than ChatGPT. It does this thing where it randomly and unnecessarily recites the date and time of when it’s responding for some asinine reason and if you try to continue a conversation with it, it will repeat things it said before verbatim as if I didn’t read them the first time. It will also tell you can do something like “give me your Twitter handle and I’ll tell you who you follow” and then you give it your Twitter handle and it says “sorry I can’t find out who you follow from your Twitter handle”

>>106193553
People have been saying that forever, but since then we've gotten worksafe alts, containment boards like /tr/, /mlp/, /lgbt/, /qst/, 6 different video game ones, a few meme boards, &c.

>>106193599
oh it must be based on gemini

>>106190402 (OP)
I'm so utterly disappointed. This summer was insanely productive with 4o with free generous limit rates.

>>106190448
All of this is complete bullshit (as in, made-up).
Your most intelligent users interacting with it *are* the training data.

>>106192697
No, I've used it extensively for months now and I can immediately confirm that GPT-5 is noticeably lobotomized as far as effort and depth are concerned. Even their replacement "add more detail" is a fake option meant to give you the illusion of choice that's been taken away from you.

>>106193387
this
but I *can* tell because I have months of reviewable chatlogs where I was getting full-page-length replies for every query