Just downloaded Deepseek, but when I clicked on the Terms of Use and Privacy Policy, they both went to a 404 not found page. I hope people aren’t just blindly signing on. #deepseek
How strange to root for the success of a Chinese entrepreneur to save the world from dictatorship and scam bubble of Trump and the weird techbros. Did Nazi this deep.
I love the DeepSeek timing. So perfect ! When I saw the SoftBank guy and 500B claims I knew OpenAi was the new Wework 💥💣. I can hear Altman screaming. They opened “stargate” and an alien came up destroying their world ! Godzilla style !
that one guy going "no wait don't fall for it it's a ccp psyop" only to get his ass community-noted where it's pointed out that his dad owns openai shares
By most benchmarks, it's about as good as the previous release of chatGPT, so not better.
But close enough to be very impressive as it needs a fraction of the hardware to run, and is Open Source, so users can run it on a local server, instead of needing to pay ChatGPT for every interaction.
my views on ai as well as china are mixed but I do love it when china beats america at its own game and when techbros want to end their own lives so hell yeah
They're so good
Btw for the Taiwan question it started writing a long answer about how it's complicated, posing both views on the situation of Taiwan, just to replace it with that
Update I hate Deepseek
It has this function where it'll act like it has an actual thought process with a text before it's actual answer, "The user is asking me about something, I have to verify my informations about said thing and provide a concise answer"
But it's very frustrating because you can see it trying to gaslight you in real time
I was asking it question about John Cena's Bing Chiling video, and it insisted that the video I was talking about was an edited version, that the real on only said Bing Chiling once
Way cheaper. GPT 4o is $2.50 per million tokens and DeepSeek V3 is $0.14 per million tokens. Output is also $10.00 per million tokens for GPT 4o while it’s $0.28 per million tokens for DeepSeek V3. 29.8x cheaper. And it has comparable benchmarks. Was $5.5 million to train the base model.
To put into perspective, GPT 4o for training alone was more than $100 million. GPT 3 cost $4.6 million to train and DeepSeek was able to match the o1 model (the base model) of ChatGPT for 3% of the cost. Puts into perspective how needlessly expensive OpenAI is.
It’s mostly about LLMs. I personally don’t believe that LLMs should be available to the average consumer and that they won’t actually result in the concept of AGI that these corporations desire. I think we need to get rid of all these non enterprise LLMs that bleed money and resources.
I lol'd so hard when I heard about the sell-off. U.S. firms have no intention of making the world better, or Deepseek would've been celebrated. It's all about investment, profit, and market share.
By most benchmarks, it's about as good as the previous release of chatGPT, so not better.
But close enough to be very impressive as it needs a fraction of the hardware to run, and is Open Source, so users can run it on a local server, instead of needing to pay ChatGPT for every interaction.
It's not as good as the OpenAI o3 model, but better or as good as their o1 model and significantly cheaper. Like worlds cheaper. That is the reason why it's huge. Also considering its open source and a Mixture of Experts, is also a big deal. MoE is considerably more efficient than a regular LLM.
Comments
However, people naively installing apps on their phone aren't that savvy.
The current issue explained.
https://youtube.com/shorts/DifzbqnoQfo?si=UA4ZEUDSibE1pKo7
10/10 no notes
Remind me how this is a good thing?
But close enough to be very impressive as it needs a fraction of the hardware to run, and is Open Source, so users can run it on a local server, instead of needing to pay ChatGPT for every interaction.
Btw for the Taiwan question it started writing a long answer about how it's complicated, posing both views on the situation of Taiwan, just to replace it with that
It has this function where it'll act like it has an actual thought process with a text before it's actual answer, "The user is asking me about something, I have to verify my informations about said thing and provide a concise answer"
I was asking it question about John Cena's Bing Chiling video, and it insisted that the video I was talking about was an edited version, that the real on only said Bing Chiling once
And the censorship is definitely a lesson in the Streisand effect 😄
Either way, it's happening if we like it or not
It was instructing 😄
https://youtube.com/shorts/DifzbqnoQfo?si=UA4ZEUDSibE1pKo7
But close enough to be very impressive as it needs a fraction of the hardware to run, and is Open Source, so users can run it on a local server, instead of needing to pay ChatGPT for every interaction.