The last line proves the user asking was making it use this type of language by prompting it
Afxal
Forerunner
Point 1> Line 1The last line proves the user asking was making it use this type of language by prompting it
Just a joke mate
blr_p
Oracle
Was looking for a way to assess how good this deepseek thing is and this graphic is the only post to demonstrate that.
How bad is open AI doing here? Add up its scores and it's still ahead.
But this cnbc video doesn't show you this. Instead they want to show you how meta's llama is losing to deepseek.
Maybe llama needs some work but openAI is still holding.
They mention openai here as a sidenote claiming it was beat but there is no nice graphic like with llama. I wonder why
So why are people here already writing its obit. So soon. Why?
America undermined? Dead giveaway this is hyped up BS
I've noticed this trend in US media going back decades where they hyperventilate about any competition.
Sputnik was a good example. The result was a few years later the US dedicated national resources towards the moon program and came out on top.
Same story going on here. More funds required for American AI. That is the only credible takeaway from this video.
Pearl harbor was attacked with 6 Japanese aircraft carriers.
Total US carriers at the time was seven divided between pacific and Atlantic. The battle of midway showed who was better with less. And the final result well known.
For the last century contenders have shown up to challenge the US and always lost to the US in the end. Primary reason being whenever the US gets its ass kicked they introspect and come back stronger. This is the only reason they have persevered as long.
Never bet against the US!
Last edited:
It is so great to see open source models but the vram required to run the highest end model is Still this should put openai under so much pressure.
Still this should put openai under so much pressure.It is so great to see open source models but the vram required to run the highest end model is
Much slower way to run - 6-8t/s - but this is one way to do it.
Kaching999
Forerunner
This video provides and insight into why Deepseek is damaging Nvidia and also how to download and run the model locally(no the entire model but a cut down version):
jayanttyagi
Forerunner
Who is Winnie the Pooh? ChatGPT and DeepSeek’s new AI chatbot beg to differ
ChatGPT said Pooh is a playful taunt of President Xi Jinping. DeepSeek talked of China "providing a wholesome cyberspace for its citizens."
madhukannan
Galvanizer
Emperor
Juggernaut
CasualGamer91
Discoverer
nullpc
Forerunner
It's a reasoning model, this is expected. What you've posted is what the model is "thinking". The actual answer is below
</think>
NoobP
Discoverer
Well, it is definitely a Chinese company
Attachments
You should be able to prompt it in such a way that it outputs only the answer and not the entire thought process.
Kevin Lane
Galvanizer
That is the same with all LLMs. There are parameters which can be passed to control that behavior.