Thread 106181860

>context window is still 32k even for plus users
>Even $200/m pro users get a context window of 128k

What a fucking disappointment, it's like 5% better at hard math but what's the fucking use if it can't read or remember for shit?

>>106181860 (OP)
its shit
because maths

sorry anon, but ive explained enough times whats up with the context size and why its an unsurmountable problem

>>106181918
Why does google give out 1 million token windows for gemini like it's candy then?

I have both paid, I use GPT when I need short burst, accurate technical shit but gemini is far superior for long conversations and has way better memory

Half the time when I put in a pdf longer than 15 pages in chatgpt it fucking lies about the end half simply because it can't read or remember the contents

>>106182231
All of Google's stuff is just better, but they dont' seem to get any attention about it.

>>106181860 (OP)
Does it still hallucinate? The making-shit-up problem is one of the critical issues, along with things like the cost of training.

>wahhhh i cant vibe code anything with it because it forgets after 3 prompts
based openai destroying the vibe coders

>>106182599
They say it has a lesser chance of hallucination but the fact that the context window is still so tiny means that when given tasks that require lots of information or guidelines be followed it will still hallucinate if the context window gets filled

So yea, less hallucination up until saturation, could have been a breakthrough but it's really not all that impressive now

>>106182334

I just can't trust a company that pulls of shit like this with a straight face.

i know it was going to be shit when hype man sam bragged about some email he got to hype up GPT-5 on the theo von podcast

>>106181860 (OP)
It's way too verbose now and does an Internet search for basic shit

>>106182653
They rolled that back and appologized.

>>106181860 (OP)
i am a plus subscriber and this shit isn't on my fucking account
gonna fucking switch to jewgle if this shit aint up

>>106182334
>All of Google's stuff is just better

>>106182693

It's the mentality that festers within the company that implemented it, had it passed through various levels of approval, created an environment where people were afraid to point out it was racist and ridiculous and then released it to production.

There are lots of companies out there with similar products that don't have this kind of bad mark on them. No need to use Google.

Imagine an all-White US-American AI lab. It would annihilate the competition.

scam altman at it again

>>106181860 (OP)
Bruh, gemini 2.5 with the 1m context is genuinely so nice when using LLMs for what they were designed to do, finding patterns in large amounts of text.

Bros, this new model is super censored. hey won't let me go back to the old models. Is it over?

>>106183269
yup, livestream said as of today all pre gpt5 models are deprecated. What they're doing is having an internal under the hood "context switch" that might use a model more like 4o or like o4 depending on your prompt. supposedly. schizo theory is that this is just a way to hide throttling and funneling users to a less intensive model.

>>106182693
Yeah, they said "Okay, yes, we did hard-code whitey out of our AI, but we didn't realize that would mean BIPOC nazi soldiers, so sorry for being racist to Black People (pbut) #BlackLivesMatter"

>>106183812
Of course the Zuck hired the guy responsible

>>106182693
Kike rats want whites to die for Israel now. They pretend it was all just a joke.