Artificial Intelligence thread

9dashline

Captain
Registered Member
Alice has 4 sisters and a brother. How many sisters does Alice's brother have?

Qwen72b got this wrong, and of the open weight nonreasoning models only llama 405 got it right

The fact that QwQ got it right at just 32b and even the quantized versions that fit on my laptop is insane
Played around with system prompt, told it to only give short answers, and quality dropped, seems like it has to be verbose and those extra tokens arent just all for show
 

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
Played around with system prompt, told it to only give short answers, and quality dropped, seems like it has to be verbose and those extra tokens arent just all for show
of course, what did you expect?

I'm really not sure how many additional tokens o1-preview generates but it charges me a fortune for every longish question I ask. It's unbelievable.

100x cheaper than o1 makes it probably even cheaper than gpt-4o, which btw, is $2.5 per million token, but I assume QwQ will generate more tokens.
 

9dashline

Captain
Registered Member
of course, what did you expect?

I'm really not sure how many additional tokens o1-preview generates but it charges me a fortune for every longish question I ask. It's unbelievable.

100x cheaper than o1 makes it probably even cheaper than gpt-4o, which btw, is $2.5 per million token, but I assume QwQ will generate more tokens.
OpenAI charges $60 per million token for o1

I just saw on hyperbolic.xyz its only 20 cents per million token... a third of deepinfra's 60 cents per mil and same BF16 and context length...

So 300x cheaper than O1 in nominal terms. Of course o1 is stronger than QwQ and probably a little less verbose but still....
 

9dashline

Captain
Registered Member
I guess they're going to complain about China's cheap labour advantage even when employing AI workers
The copium is wild, half the news articles always brings up the T square event which was a CIA failed coup.... so I guess the 1.6 billion paid off a lot of "reporters"...

Like does China or rest of the world bring up the Waco Texas thing everytime ClosedAI , Anthropic, or Meta releases a new AI model?

Its sore loser behavior, that they tried to kill China AI/tech with the EUV underhanded thuggery and yet when China still manages to catch up, they seeth yet again....

This is why China must win the tech war, and the Race to AGI. Literally the rest of humanity and all other life on earth depends upon it now.
 

iewgnem

Junior Member
Registered Member
The copium is wild, half the news articles always brings up the T square event which was a CIA failed coup.... so I guess the 1.6 billion paid off a lot of "reporters"...

Like does China or rest of the world bring up the Waco Texas thing everytime ClosedAI , Anthropic, or Meta releases a new AI model?

Its sore loser behavior, that they tried to kill China AI/tech with the EUV underhanded thuggery and yet when China still manages to catch up, they seeth yet again....

This is why China must win the tech war, and the Race to AGI. Literally the rest of humanity and all other life on earth depends upon it now.
People play dirty when they're desperate, people become desperate when they're losing. Open sorce model running 300x cheaper than OpenAI 's model basically mudered them.

At end of the day China has way cheaper energy from renewables and nobody uses only a single chip for training, EUV might not be optimal in the cost equation even if its availiable. The root problem with American C suites is they dont understand technology is about talent, not just money
 

9dashline

Captain
Registered Member
People play dirty when they're desperate, people become desperate when they're losing. Open sorce model running 300x cheaper than OpenAI 's model basically mudered them.

At end of the day China has way cheaper energy from renewables and nobody uses only a single chip for training, EUV might not be optimal in the cost equation even if its availiable. The root problem with American C suites is they dont understand technology is about talent, not just money
Yup, seems like DeepSeek kicked off the AI API inferencing price war 2.0 , this time reaching Anglo shores as opposed to strictly internal/domestic unlike last time

Please, Log in or Register to view URLs content!
 
Top