I have just finished my first chat with Deepseek (the big one online with reasoning enabled). After some wild random questions I decided to play a game: I started a new chat and made Deepseek guess who I am based on more and more clues I gave over time. That was really impressive. Even with lesser known physicists it mostly took the system 3 guesses to happily shout my name out. After that I grabbed a beer and played the other way around. This is truly amazing. Can't wait to use those systems when they can fluently be used by voice. This really is the next step.
@@dwiii1635 Wow! Get back in your bunker 😆 I don't tell personal stuff on online services, neither on US platforms nor other ones, be it here in Europe or whereever.
I focus more on the human language side of AI, and I've found DeepSeek so much better at synthesizing complex layers of information in terms that are easy to understand. Definitely a fan so far! And you're right about the loading speed; DeepSeek definitely takes a little more "thinking time" than some of the alternatives. Thanks for making & sharing this video :)
Great video. I like the way you did your comparison of the two products. I'm also keeping my ChatGPT subscription a little longer but I agree with your conclusions on where AI is hopefully going.
Hello Thu, amazing video! i enjoyed how you actually showed examples and tested it in a very logical and easy to follow way. Looking forward to see more of your content 😊
I think the DeepSeek wins the chart test in the end because openAI just assumed that I understands what the word "log scale" means but Deepseek tried to make it more visual for me to understand what logarithmic scale is
How would someone be involved with graphs who's ever vaguely involved in data science and not know log scale? besides, you can look up log scale and get a definition quite quickly
@@daxeckenberg I am relatively new to Gen AI and I have no experience using any programming languages and I had never hear what Log scale meant. But when I looked at the image I understood how the values on left looked compressed and deepseek told us the same thing that's why I felt the reply was better. but you are right I can look up the definition
In a few years, all the public models will reach a high level of parity. We are still in the "golden age". This is like the NES of AI. GPT 01 might be the SNES. Eventually we'll get the PS3 and 4 of AI. And by then it will be very reliable and available everywhere.
Hello fellow Beta boulderer. Great to see you again with a great video. I really enjoyed your comparison of DeepSeek vs ChatGPT side by side. Great job!! Hi from Denmark!
Always appreciate your video. Thank you for putting this together. I would also add that what sets DeepSeek apart is the display of its reasoning/logic. I think this is extremely helpful for humans since it shows how the answer was derived - much like showing one’s work in solving equations and science experiments. The black box that ChatGPT operates within is more dangerous in the longrun because it makes humans lazier and more dependent on the AI to solve problems (even if the answers are wrong sometime). I don’t want a future where people water plants with “electrolytes.” 😂
yes ! I have also found DeepSeek-R1 less capable than GPT-o1 for coding tasks ! Like most of our work ...it is it a tradeoff between our time and our money !?!
I use copilot .its initially dum . Its on purpose . Its divided . Hello . Simple . A bit like intro . Or pre . Do not go into detail wrong step . 1/2.2 is gamma , logic gate binary code , human readable code . This was the context . Next step , while staying in context ask all your question . You can go academia mode but warn it first . And require it to stay in context till a safeword is used . Copilot is underestimated . If you just want an out of context answer ai is not for you . Ai are like 9 year old highest educated doctor . Human readable code . Ai see it as useless .
Tough question to answer without context, but I think whatever tech job in the future will be more and more about collaborating with AI. Whatever you study, I'd suggest being open-minded, learning fast and being good at what you do!
At least for quite some time we need to be very cautious with using the output of AI. It's still going to take experts in the particular knowledge space to validate that an AI answer is accurate and not just a hallucination. I would think of AI as a enthusiastic junior analyst. where 90% of the time it's going to be mostly right, but it's still going to have edge cases where only with experience. will you realize it's errors. I imagine this might get addressed in the future, but in the near term AI still needs to be mentored so to speak
I want you to know that o1 is a match to the full 600 billion parameter. The one you are using is just distilled version. Fair test use r1 model Edit: she is using the r1
It appears she used the deep-seek web interface which would be the full r1 model. earlier in the video she talks about running it locally but what she ran in the demo was what appears to be public large model interface.
I wouldn't be too quick to cancel your chatGPT subscription. The o3 model should be out soon and might be better than DeepSeek. And I suspect openAI will put in more protection against other companies harvesting training data from their frontier models, so it may become harder for copycat companies to replicate the newest models
Keep your friends close and your enemies closer. Avoiding deep seek just because it's Chinese backed isn't the best approach. This isn't to say to be reckless but avoiding something just because it's from a nation state that may not be the most friendly isn't a good approach long term. If you don't know their capabilities, you won't know their strengths and weaknesses and will therefore not be able to capitalize on that knowledge.
I was curious how these compared for Data Nerds! Thanks for doing the in-depth analysis, Thu 🙌
I have just finished my first chat with Deepseek (the big one online with reasoning enabled). After some wild random questions I decided to play a game: I started a new chat and made Deepseek guess who I am based on more and more clues I gave over time. That was really impressive. Even with lesser known physicists it mostly took the system 3 guesses to happily shout my name out. After that I grabbed a beer and played the other way around. This is truly amazing. Can't wait to use those systems when they can fluently be used by voice. This really is the next step.
Welp, looks like the CCP knows who you are now, lol. What else you tell it? Your social? Lol.
@@dwiii1635 Wow! Get back in your bunker 😆 I don't tell personal stuff on online services, neither on US platforms nor other ones, be it here in Europe or whereever.
@@dwiii1635 bruh, its the same with chatGPT, they already know everything about you
Hi Thu Vu. Thank you for talking about DeepSeek vs OpenAI-o1 and testing them for data science tasks. Keep it up!
Hey, thank you for watching and commenting! ❤
Hats off for such fair comparison and letting us know how to use these AI tools in secure manner
Hi everyone, it's been a while! 🤯 Have you tried DeepSeek-r1 yet? What's your impression?
I focus more on the human language side of AI, and I've found DeepSeek so much better at synthesizing complex layers of information in terms that are easy to understand. Definitely a fan so far! And you're right about the loading speed; DeepSeek definitely takes a little more "thinking time" than some of the alternatives. Thanks for making & sharing this video :)
@@kaylee.m_ai Indeed, it's so good at unpacking a question into small coherent steps, it's mindblowing! Thanks for sharing :)
O1 also thinks through it but "Open"Ai hides its reasoning steps
Great tests!
Hello Thu, fantastic video! Looking forward to watching all of them.
Great video. I like the way you did your comparison of the two products. I'm also keeping my ChatGPT subscription a little longer but I agree with your conclusions on where AI is hopefully going.
Hello Thu, amazing video! i enjoyed how you actually showed examples and tested it in a very logical and easy to follow way. Looking forward to see more of your content 😊
O1 also thinks but "Open"Ai hides its reasoning steps
Great comparison!
Thanks for a great video!
I think the DeepSeek wins the chart test in the end because openAI just assumed that I understands what the word "log scale" means but Deepseek tried to make it more visual for me to understand what logarithmic scale is
How would someone be involved with graphs who's ever vaguely involved in data science and not know log scale?
besides, you can look up log scale and get a definition quite quickly
@@daxeckenberg I am relatively new to Gen AI and I have no experience using any programming languages and I had never hear what Log scale meant. But when I looked at the image I understood how the values on left looked compressed and deepseek told us the same thing that's why I felt the reply was better. but you are right I can look up the definition
Hi Thu Vu thank you for your content it's very interesting as always.
In a few years, all the public models will reach a high level of parity. We are still in the "golden age". This is like the NES of AI. GPT 01 might be the SNES. Eventually we'll get the PS3 and 4 of AI. And by then it will be very reliable and available everywhere.
Thank you Thu Vu for the detail and concise video on both AI models. But there is now a 3rd one which out performs DeepSeek called Qwen2.5! 🤯
Great content!
Thinking longer ... couldn't it be connectivity? Thanks for the video
point to note, if you are using gpt for a while then it kinda know your preferences already and tailor it's responses accordingly..
i was about to send you a message, it has been long time since we see your content.
thank you,
Hello fellow Beta boulderer. Great to see you again with a great video. I really enjoyed your comparison of DeepSeek vs ChatGPT side by side. Great job!! Hi from Denmark!
The web site version of deep seek is not the whole model, its a distilled version also. So your compared a crippled deep seek with full chatgpt.
Always appreciate your video. Thank you for putting this together.
I would also add that what sets DeepSeek apart is the display of its reasoning/logic. I think this is extremely helpful for humans since it shows how the answer was derived - much like showing one’s work in solving equations and science experiments. The black box that ChatGPT operates within is more dangerous in the longrun because it makes humans lazier and more dependent on the AI to solve problems (even if the answers are wrong sometime). I don’t want a future where people water plants with “electrolytes.” 😂
yes ! I have also found DeepSeek-R1 less capable than GPT-o1 for coding tasks ! Like most of our work ...it is it a tradeoff between our time and our money !?!
OpenAIs o3 is coming soon.
I use copilot .its initially dum . Its on purpose . Its divided . Hello . Simple . A bit like intro . Or pre . Do not go into detail wrong step . 1/2.2 is gamma , logic gate binary code , human readable code . This was the context . Next step , while staying in context ask all your question . You can go academia mode but warn it first . And require it to stay in context till a safeword is used . Copilot is underestimated . If you just want an out of context answer ai is not for you . Ai are like 9 year old highest educated doctor . Human readable code . Ai see it as useless .
Is it still worth the effort and time to study for Masters Degree in Data Science?🙏
Tough question to answer without context, but I think whatever tech job in the future will be more and more about collaborating with AI. Whatever you study, I'd suggest being open-minded, learning fast and being good at what you do!
At least for quite some time we need to be very cautious with using the output of AI. It's still going to take experts in the particular knowledge space to validate that an AI answer is accurate and not just a hallucination.
I would think of AI as a enthusiastic junior analyst. where 90% of the time it's going to be mostly right, but it's still going to have edge cases where only with experience. will you realize it's errors.
I imagine this might get addressed in the future, but in the near term AI still needs to be mentored so to speak
I think 01 is a bit better.. but deepseek is just nuances off
To me, it is obvious that GPT-01 has far better answers.
🥰
I want you to know that o1 is a match to the full 600 billion parameter. The one you are using is just distilled version. Fair test use r1 model
Edit: she is using the r1
It appears she used the deep-seek web interface which would be the full r1 model.
earlier in the video she talks about running it locally but what she ran in the demo was what appears to be public large model interface.
I wouldn't be too quick to cancel your chatGPT subscription. The o3 model should be out soon and might be better than DeepSeek. And I suspect openAI will put in more protection against other companies harvesting training data from their frontier models, so it may become harder for copycat companies to replicate the newest models
#cancele_closeai
As ‘quick’ as you you
I will move all my projects out of openai
the pi/pi ? Those math will face mathematician soon . They re very bad computer wise
Why does this feel so much like an advertisement?
I am not sure. Maybe she is funded by ClosedAI
Don't use DeepSeek. Its Chinese...and you know what they are doing with all data used for the model.
Keep your friends close and your enemies closer.
Avoiding deep seek just because it's Chinese backed isn't the best approach.
This isn't to say to be reckless but avoiding something just because it's from a nation state that may not be the most friendly isn't a good approach long term.
If you don't know their capabilities, you won't know their strengths and weaknesses and will therefore not be able to capitalize on that knowledge.
Uh Microsoft Azure is hosting it as of today and already many others. Framing this is a China vs US is a bad take. Open source vs.closed
Yea... Is ok when American Companies take your data. Tool...
It’s #1 app on App Store in 140+ countries… people love free services 😂😂😂