GPT-5 is worse than GPT-4

[thales20]

It didnt add anything useful; its not faster (quite the contrary of fast); its processing got worse.

Model router just made it 300x slower. Doesnt matter if I click "I want a faster answer" it will still be garbage speed.

How can AI engineers whose get paid millions and millions annualy do such a horrific job?

Was this the model that would break all the benchmark scales?

 

[Uehdbwidbfngj]

D

It didnt add anything useful; its not faster (quite the contrary of fast); its processing got worse.

Model router just made it 300x slower. Doesnt matter if I click "I want a faster answer" it will still be garbage speed.

How can AI engineers whose get paid millions and millions annualy do such a horrific job?

Was this the model that would break all the benchmark scales?

Dnr

 

[thales20]

D

Dnr

Uehdbwidbfngj​

Mistral​

JoinedDec 4, 2024Posts2,166Reputation2,684

 

[Uehdbwidbfngj]

Uehdbwidbfngj​

Mistral​

JoinedDec 4, 2024Posts2,166Reputation2,684

Problem?

 

[thales20]

Problem?

dnr

 

[laworg]

It didnt add anything useful; its not faster (quite the contrary of fast); its processing got worse.

Model router just made it 300x slower. Doesnt matter if I click "I want a faster answer" it will still be garbage speed.

How can AI engineers whose get paid millions and millions annualy do such a horrific job?

Was this the model that would break all the benchmark scales?

Ehh no his coding abilities are very friggin good u obviously dont have the 200 monthly one peasant

 

[NudeSelfiesTilliDie]

Yeah it king of popped the AI bubble. People realizing they were overhyping it lmao.

 

[thales20]

Ehh no his coding abilities are very friggin good u obviously dont have the 200 monthly one peasant

I'm talking about the pure GPT-5. If we compare the benchmarks between the 3 and 4 its just mind blowing we didnt get anything different

 

[CantStopTheMog]

It didnt add anything useful; its not faster (quite the contrary of fast); its processing got worse.

Model router just made it 300x slower. Doesnt matter if I click "I want a faster answer" it will still be garbage speed.

How can AI engineers whose get paid millions and millions annualy do such a horrific job?

Was this the model that would break all the benchmark scales?

Nerd ramblings- check out my post

 

[thales20]

Nerd ramblings- check out my post

dnr

 

[laworg]

I'm talking about the pure GPT-5. If we compare the benchmarks between the 3 and 4 its just mind blowing we didnt get anything different

We did it was 600m compared to 5 billion or 1.5 i dont remember much they yapped for a Hour it is non debatable and a fact this one is more capable bcus gpt anwsers now and doesnt call u king papi anymore doesnt mean its bad maybe stop asking it if rice is a carb u donut

 

[thales20]

We did it was 600m compared to 5 billion or 1.5 i dont remember much they yapped for a Hour it is non debatable and a fact this one is more capable bcus gpt anwsers now and doesnt call u king papi anymore doesnt mean its bad maybe stop asking it if rice is a carb u donut

I honestly do not care about him being cold or whatever now. If we compare the benchmarks its almost the same shit. Maybe just 2% more and it doesnt pay the slowness we recieved.
What you are seeing are those funny trust me bro benchmarks.

Like its so horrific done that even the charts they made from the benchmark are retards:

You either doesnt use chat gpt or trust too much on the benchmarks THAT OPENAI DOES LMAO.

Either way claude mogs.

 

[laworg]

I honestly do not care about him being cold or whatever now. If we compare the benchmarks its almost the same shit. Maybe just 2% more and it doesnt pay the slowness we recieved.
What you are seeing are those funny trust me bro benchmarks.

Like its so horrific done that even the charts they made from the benchmark are retards:

View attachment 4025137

You either doesnt use chat gpt or trust too much on the benchmarks THAT OPENAI DOES LMAO.

Either way claude mogs.

claude is better from a anecdotal view but i saw the vid and also from my experience it now anwser better even if it isnt technical on easy stuff too it listens better

 

[got.daim]

It didnt add anything useful; its not faster (quite the contrary of fast); its processing got worse.

Model router just made it 300x slower. Doesnt matter if I click "I want a faster answer" it will still be garbage speed.

How can AI engineers whose get paid millions and millions annualy do such a horrific job?

Was this the model that would break all the benchmark scales?

its literally the exact same imo it gives the exact same answers and replies as gpt 4

 

[Saint Casanova]

AI is a glorified search engine. JFL at all the fearmongering

 

[pope]

Gpt 3 was a God at estimating body fat percentages

 

[gooner23]

definitely much better than gpt 4 people just a little outraged by the deliberate attitude change i keep telling people anyone that was using chatbot marketplaces like poe, the #1 bots were always roleplay bots.

whatever was in copilot and perplexity was definitely on par with gemini i think people were expecting something like absurd but we are running out of room for it only price/context matters nowadays since most people don't need a stronger model

 

[MouthBreathingElite]

Its better (knowledge wise) but wayyy slower

 

[gooner23]

I honestly do not care about him being cold or whatever now. If we compare the benchmarks its almost the same shit. Maybe just 2% more and it doesnt pay the slowness we recieved.
What you are seeing are those funny trust me bro benchmarks.

Like its so horrific done that even the charts they made from the benchmark are retards:

View attachment 4025137

You either doesnt use chat gpt or trust too much on the benchmarks THAT OPENAI DOES LMAO.

Either way claude mogs.

think people overrated claude a little too much absurdly expensive for slightly better performance and if multi step/agentic is the main indicator gpt 5 woops it in both performance and pricing

but i still think gemini > all models by a mile due to context and pricing and if you need an agent just use copilot on like the cheapest model and using gemini for the answer

 

[registerfasterusing]

It didnt add anything useful; its not faster (quite the contrary of fast); its processing got worse.

Model router just made it 300x slower. Doesnt matter if I click "I want a faster answer" it will still be garbage speed.

How can AI engineers whose get paid millions and millions annualy do such a horrific job?

Was this the model that would break all the benchmark scales?

Ive noticed its memory is worse too