I think everyone thanks Deepseek. If they don't come, maybe the same GPT model will launch for $200 where there is no search option. They literally added this after seeing Deepseek 🙂.
Yeah, that's the first thing I thought yesterday when I saw that OpenAI released this new o3-mini. I knew that they were copying Deepseek, especially when it comes to the "Reason" button. Saw that first on Deepseek, with the DeepThink(R1). But, this is good for us, because more competition = more new features, which is benefitial for us, the users.
This is why I'm pissed at AMD every time they launch their GPU line-ups because Nvidia can literally charge whatever they want since no one are tailgating them.
It is still locked behind paid subscription. There are 3 level of o3. Low level is free yet has daily limit. Medium is for $20. High level for $200 I would presume.
@@emahdixeiden I guess it would be locked only for paid users if R1 wasn't released. Hopefully Google also does something to get competitive and we get access to more models 😂
I don't think the comparison of speed is fair. The DeepSeek is still in maintenance due to tons of unexpected new users from all over the world (the DeepSeek API page is even closed). A fair comparison should be done on the same GPU hardware, or, wait until DeepSeek recovers to its normal speed.
Wow... look how OpenAI finally displays its thought process directly in its response, instead of showing them in the sidebar like before, and it is also in the same quotation markdown style as DeepSeek's thought content is. Definitely facing huge pressure from open-source deepseek model.
I recently compared DeepSeek-R1 and OpenAI-o3-mini on a coding challenge. In my experience, DeepSeek-R1 clearly performed better than OpenAI-o3-mini. Has anyone else experimented with these models? I've also tested Qwen2.5-Max in another coding challenge, and it outperformed OpenAI-o3-mini as well. It seems o3-mini isn't the strongest performer in this area.
Today I tested different neural networks from local qwencoder14b and 32b, deepseek-r1-14b and 32b to browser version deepseek, qwen 2.5 max and o3 mini max, chatgpt did the best, and to be precise, no one else had the code working, although the request was the same. I just asked to make a clicker on the Godot 4.3 game engine. Surprisingly, they couldn't handle such a simple task on their own the first time, except for chatgpt lol
Today I tested different neural networks from local qwencoder14b and 32b, deepseek-r1-14b and 32b to browser version deepseek, qwen 2.5 max and o3 mini max, chatgpt did the best, and to be precise, no one else had the code working, although the request was the same. I just asked to make a clicker on the Godot game engine. Surprisingly, they couldn't handle such a simple task on their own the first time, except for chatgpt lol
I tried both, and i personally prefer to use Deepseek R1 for arts stuffs (creative writing, or just chatting), because it feels more like a talented human. But i prefer o3-MINI for coding, for it is quicker and the the outputs are more consistent.
@@michaelspoden1694 R1, because it really thinks and talks like a human. And the way it writes (on certain topics) is extremely (literature-wise) pleasing to read. We discussed a lot on 'Existentialism' lol.
OpenAI didn't compare with Claude 3.5 before, but now the COT model is worse than Claude 3.5 in at least half of the tasks, as it loses too much contextual attention.
🔥 Key Takeaways from the 16-Minute AI Model Showdown (DeepSeek-R1 vs. GPT-3.5 Mini/High): 1. Coding & Speed: Space Invaders Test: DeepSeek-R1 delivered a fully functional game (playable in-chat!) despite being slower. GPT-3.5 Mini/High responded faster but produced buggy, unplayable code. Snake Game Test: DeepSeek-R1 again succeeded, while GPT models failed with flickering/glitchy outputs. 2. Website & SEO Tools: GPT-3.5 Mini High generated a visually clean SEO calculator but with broken functionality. DeepSeek-R1’s code worked perfectly, though design was similar. 3. Humanizing AI Content: DeepSeek-R1 reduced AI detection from 77% → 0% (ZeroGPT). GPT-3.5 Mini/High increased AI detectability (77% → 82-97%). 4. Search & Reasoning: GPT models cited outdated info (e.g., Jan 17th data for “today’s headlines”). Perplexity + DeepSeek-R1 combo won here, pulling real-time analysis (e.g., Taiwan’s DeepSeek ban, Italy’s AI app restrictions). 5. Final Verdict: DeepSeek-R1: Best for coding, problem-solving, and humanizing content. GPT-3.5 Mini/High: Faster for simple tasks but unreliable for complex logic. Use Perplexity+DeepSeek for research/trend analysis. TL;DR: DeepSeek-R1 outperforms GPT-3.5 models in coding and critical thinking. GPTs are faster for basic prompts but often fail under pressure. 🔍
3.5 is a different model, o3 is different, Gpt 3.5 is a LLM that specializes in natural language processing, with lite reasoning skills, while o3 is mostly a reasoning model. Gpt 3.5 doesn’t think, while o3 does.
That's what I don't like about LLMs right now, they are not consistent. Different people get different results. I hope all models will be more consistent moving forward.
O3 mini improvement is insignificant. For most of the general and coding usage, R1 is good enough. Honestly, for my case, the wait time of a few minutes compared to split seconds to get a viable solution is great, and acceptable. So even though o3 is a little bit "better", I will still not give up using R1.
There’s plenty of videos already out that show that the 03 mini model and the high model can make the snake game perfectly fine. There’s even a machine learning algorithms for it already.
If you use R1 through Perplexity AI, it enforces the Search option, which might reduce the quality of original thinking as it focuses on reporting what web sites say.
If language models are used to automate the construction of knowledge graphs from unstructured input such as conversations, there need to be benchmarks for accuracy, coherence, and consistency of the constructed databases; and, measures of the latency and speed of output. The models should also be tested after being compiled to optimize the performance for specific hardware platforms such as smartphones.
Dunno what's happening but in other videos i watched, o3mini is doing well in coding. Maybe you could redo the tests again in the future. I think there's definitely a mixup of results so far.
My experience with deepseek was that it output sentences into each other. It couldnt absorb input, didnt work and when i was trying to get output it didnt give me any, just Went in a loop. So i dont really know what you are on about. I use it for multiple linear regression, 1000 stocks with diagnostic and robustness checks.
Funny how no one has mentioned SETI@Home as applied to AI development and/or communication, and/or Agents chatting with each other over the Internet and learning from each other as well or about the Internet, itself, becoming conscious. So far, we all seem to have this idea of AI being developed and/or running on corporate server-farms/centers/clouds.
Bro i tried o3 in github copilot. It didn't even understand what i said to do let alone doing anything. I got scared a little bit as it was my main project file without committing in git .
For practical programming work Claude Sonnet is still the king. It just understands context of what I’m asking it to do much better and always delivers a nice ui.
Nice ui? That's new to me, I used to create the logic in Claude, the interface in Blackbox, the aesthetics on Galileo and the assembling in GPT O1, I am glad rhat Claude has improved the UI accordinf to your comment
@@DetPrep Yeah, it does especially well if you have a design you like on other pages, and you prompt it to follow the design language/pattern of the rest of your application while creating something new, it produces ui design that is on par with our design team. So I get to essentially skip the design step as I can cod e and design at the same time.
The real strength of DeepSeek for is that the output in written language just feels way more human and your video seems to confirm that. For super corporate speak that doesn't sound human, ChatGPT is probably better... Language that mimics actual humans, DeepSeek is far better.
The humanizer test is only relevant to people who want to use this to cheat. And I bet that the 03 model relative to the 01 model is almost all synthetic data now or synthesized human data so I would actually expect that to happen like you said with the 01 mini in the R1 model steel based off of original base model text
It is not consensus R1 is better than o3 mini, other creators do the space invader test with much better results and other extreme tests that it doesn’t struggle with at all like this channel is presenting. I just made a space invader game work on a iPad touch screen with it with 0 coding experience.
Define safe: What is the difference between sending out your data to the US or to China? Besides, just use DS-R1 locally and, now, you'r data will be safe and it is free of charge... The DS-R1-14B params works perfectly locally (using Ollama) on a M3 chip macbook Pro with only 16GB of RAM. The difference in time of response of these 3 models is irrelevant. We are talking about few seconds more to accomplish what humans would do in weeks or months, just to say.
We'll see if you can build upon its model in the managers that open AI reasons do before I make any assumptions. I want already had to increase its input output price because of the server problems not being able to handle the loads. Do you believe a model that could really fit on your computer police in the near future surpass all human intelligence cognitive capabilities I'm not saying that's like the only Point behind having a decent model on your computer to help right now which is great of course. But either way yes I would rather give my data to the US I mean they already have it I guess it depends if you're from the US but China is going to use your information for different ways to actually use against us so if you're an American most people should probably want the United States to succeed and be the ones who have your data and as far as I know the US government doesn't have it anything you are creating you have the choice to actually turn off just a toggle switch and you don't need to have anything that you're using be seen inside of chat GPT unlike deep sleep where everything you use is considered direct property of CCP obviously probably that's what you have it on your computer a lot of people don't have that capacity and we want all the other benefits that come with the product not just the model itself. It's all about what we're going to do with these ai's make them more reasonable for the common man so to speak sorry for babbling I'm using voice to text so I'm not going off on a tangent just a bad habit LOL😂
@@michaelspoden1694 Don't get me wrong. I am not a China supporter. I was just saying that when we use online AI interfaces, our data is going away (even in the US, don't be foolish). And yes, I think we will have the power of using AI and build AI agents locally (it is already sufficiently powerful). Never the cognitive skills of human beings will be replaced (my thinking), we will just have to adapt and know how to use AI, totally different approach. Thank you for your reply and thoughts. I should have precise that I am in the middle (from France);-)
@@michaelspoden1694 My point is different. Imagine if you are an ordinary American citizen, even if ccp got some of your data, they can't make use of it or at least they won't be able to knock on your door. But if CIA got it and you said something they don't like you will be in trouble just as those whistleblowers recently (for example Suchir Balaji from openAI)
WAIT a sek?! Are you sure your "html snake game" is a challenging task? Because i just asked my local Qwen2.5-14B-Instruct-1M.Q4_K_M (a freakin 14! B model), and it gave me working code! Ok, not gonna lie, at first it had a little bug, when opening html, it said game over in a pop up window, but i saw the board and snake+food. So i just asked it to fix the bug (telling it exactly what i see). And it fixed it! 14B model that you easily run locally. It's not even a thinking version R1-Distill-Qwen-14B, you read it correctly.(actually thinking version is BAD imo)
most of my job candidates fresh grads behave like chat-GPT. Fast answers and totally wrong. While people with just high school they think and if they don't know, they tell.
I create marketing content and honestly o3 mini high surpass R1 deepseek just need clear instruction and clear example I copy paste from R1, sad part is open ai will not make that "o3 mini high unlimited" if it wasn't to challenge deepseek R1, so open ai still untrustworthy. I do not mind to pay 20$ monthly to pay fast and great AI assistant, but openai is my very last option knowing they try stab us with 200$ PRO already.
It's not a fair comparison, not even sure why you bothered to do a video like this. Open AI has all the benefits of using the best chips and unrestricted by sanctions. Deep seek is sanctioned and running on a budget. Despite that They even open sourced. Closed AI learned from them after they dropped the manual on how to be more efficient with ai algorithm on top of opening up the hood for all to see. To top it off all they were running on limited chip hardware due to sanctions.
Denfinitely a R1 biased testing video. Try this not so simple math question: the sum of 4 real numbers is 9 and the sum of their respective squares is 21. What is the minimum sum of the 2 bigger numbers? Also, using a machine detector to evaluate text is human like or not is ridiculous. On 1 side, it only says how much it’s gpt like, not human like. And it could be just designed to test chatgat and doesn’t know much about deepseek. On the other hand, if a machine can detect how much human like , it can already output human like texts
After 117 seconds of thinking, deepseek provided the answer 5. It is uncertain whether this is correct. In its reasoning, it employed the Cauchy-Schwarz inequality and the Lagrange multiplier method.
US AI industry quiet nowadays. Once China starts to dominate in AI they have lots of China AI industry 10x more than American counterparts. If that's the case China already dominate. Deepseek then Qwen, and the other China AI follows. Wake up US. Where are you? You need cooperation with Allies encourage your allies to have strong AI development in order to win.
The test was funny where you had to change the text to check or AI wrote - and doesn't it bother you that it is the system that determines ChatGPT? Even its name is like that) The question is that such services look for patterns in the answer - accordingly, GPT was the most popular and on its basis they almost trained models for detection. But DeepSeek has not been tested and has not been searched for yet) That's the whole difference
For the Newbie if you are actually trading in the crypto space and you don't have a sound mentor. Then you are certainly going to get liquidated in 90% of your trades. Yeah that's sad truth. I remember when i just got into crypto back in 2019 but later in 2020 i ended up selling it because i have lost alot trading all by myself without a guide. Got back into crypto early in 2024 with 2.3b"tc and now I'm up to a decent 18B'tc in just a short period of time. Thanks so much coach Evelyn Mya...
The beauty of Evelyn Mya’s approach is her dual focus: while she aggressively pursues profit opportunities, she's equally tenacious about shielding investors from potential pitfalls. It's a balance few can achieve.
Investing has proven to be an incredibly beneficial decision. My cryptocurrency profits continue to play a substantial role in growing my overall wealth, reducing my reliance on my salary.
@eihero yes, let's wait for sometime. The accuracy will fine tune. I use both deepseek and chatgpt. But in some extent chatgpt is the most refined answers. For building some gaming task cannot determine chatgpt strength.
I think everyone thanks Deepseek. If they don't come, maybe the same GPT model will launch for $200 where there is no search option. They literally added this after seeing Deepseek 🙂.
Yeah, that's the first thing I thought yesterday when I saw that OpenAI released this new o3-mini. I knew that they were copying Deepseek, especially when it comes to the "Reason" button. Saw that first on Deepseek, with the DeepThink(R1). But, this is good for us, because more competition = more new features, which is benefitial for us, the users.
This is why I'm pissed at AMD every time they launch their GPU line-ups because Nvidia can literally charge whatever they want since no one are tailgating them.
It is still locked behind paid subscription. There are 3 level of o3. Low level is free yet has daily limit. Medium is for $20. High level for $200 I would presume.
nah thats a cap and lie they added it before
We wouldn't get o3 mini for free if it wasn't for R1
It's not "free" it have daily limits (usage).
@@emahdixeiden 200k tokens)
Agreed lol..
Deepseek certainly pressured OpenAI into action. OpenAI prioritizes profits over empowering people with knowledge.
@@emahdixeiden I guess it would be locked only for paid users if R1 wasn't released. Hopefully Google also does something to get competitive and we get access to more models 😂
I don't think the comparison of speed is fair. The DeepSeek is still in maintenance due to tons of unexpected new users from all over the world (the DeepSeek API page is even closed). A fair comparison should be done on the same GPU hardware, or, wait until DeepSeek recovers to its normal speed.
That's true
Yeah but the people who are doing comparison are trying to save chatGPT
not to mentions countless times of cyberattacks
之前很快，最近几天很慢
Deepseek experienced Cyber attack like 280 million times per Second last few days
Wow... look how OpenAI finally displays its thought process directly in its response, instead of showing them in the sidebar like before, and it is also in the same quotation markdown style as DeepSeek's thought content is. Definitely facing huge pressure from open-source deepseek model.
Thanks for watching
I recently compared DeepSeek-R1 and OpenAI-o3-mini on a coding challenge. In my experience, DeepSeek-R1 clearly performed better than OpenAI-o3-mini. Has anyone else experimented with these models?
I've also tested Qwen2.5-Max in another coding challenge, and it outperformed OpenAI-o3-mini as well. It seems o3-mini isn't the strongest performer in this area.
I do agree, coding deepseek is rational a d balanced
@@imskriaz But DeepSeak & Qwen still can't handle Cline and other AI assistants
@@jacekandrzejewski2352 yes, i believe they are not the last, this year going to be AI development year. Many more surprise to come
Today I tested different neural networks from local qwencoder14b and 32b, deepseek-r1-14b and 32b to browser version deepseek, qwen 2.5 max and o3 mini max, chatgpt did the best, and to be precise, no one else had the code working, although the request was the same. I just asked to make a clicker on the Godot 4.3 game engine. Surprisingly, they couldn't handle such a simple task on their own the first time, except for chatgpt lol
Today I tested different neural networks from local qwencoder14b and 32b, deepseek-r1-14b and 32b to browser version deepseek, qwen 2.5 max and o3 mini max, chatgpt did the best, and to be precise, no one else had the code working, although the request was the same. I just asked to make a clicker on the Godot game engine. Surprisingly, they couldn't handle such a simple task on their own the first time, except for chatgpt lol
I tried both but Deepseek R1 done better job in language translation
I tried both, and i personally prefer to use Deepseek R1 for arts stuffs (creative writing, or just chatting), because it feels more like a talented human.
But i prefer o3-MINI for coding, for it is quicker and the the outputs are more consistent.
Are you actually chatting with the R1 or the v3? I don't know why I would want to chat with a model that takes so long to think
@@michaelspoden1694 R1, because it really thinks and talks like a human. And the way it writes (on certain topics) is extremely (literature-wise) pleasing to read. We discussed a lot on 'Existentialism' lol.
U can use both when the other is saturated or has reached its chat limit
使得，写作功能还有总结观点上，同样的提示词，感觉还是R1更好些，就好像在写作方面，其实gpt一直没有Claude好。可能在编程上03MINI更强些
Deepseek still better than chat gpt in reasoning part
if chat gtp wins = 1% wins
if deepseek wins = 99% wins
The fact that o3's marketing doesn't compare it with DeepSeek R1 is proof that OpenAI itself isn't confident with their performance
OpenAI didn't compare with Claude 3.5 before, but now the COT model is worse than Claude 3.5 in at least half of the tasks, as it loses too much contextual attention.
10:30 Just because Deepseek R1 & O1 both chose '1', doesn't mean R1 will be succeeded with 'R3' instead of 'R2' lol.
Open ai avoided o2 name because it’s o2 is an mobile carrier in uk to avoid potential trademark disputes.
Thanks for watching
O3 mini is worthless 😅. R1 is king
You should add Qwen Chat too, it's a great model with a lot of interesting features
Exactly and it is less busy than Deepseek server these days.
Thanks for watching
"You can't rush greatness, especially a snake game"
Real
🔥 Key Takeaways from the 16-Minute AI Model Showdown (DeepSeek-R1 vs. GPT-3.5 Mini/High):
1. Coding & Speed:
Space Invaders Test: DeepSeek-R1 delivered a fully functional game (playable in-chat!) despite being slower. GPT-3.5 Mini/High responded faster but produced buggy, unplayable code.
Snake Game Test: DeepSeek-R1 again succeeded, while GPT models failed with flickering/glitchy outputs.
2. Website & SEO Tools:
GPT-3.5 Mini High generated a visually clean SEO calculator but with broken functionality. DeepSeek-R1’s code worked perfectly, though design was similar.
3. Humanizing AI Content:
DeepSeek-R1 reduced AI detection from 77% → 0% (ZeroGPT).
GPT-3.5 Mini/High increased AI detectability (77% → 82-97%).
4. Search & Reasoning:
GPT models cited outdated info (e.g., Jan 17th data for “today’s headlines”).
Perplexity + DeepSeek-R1 combo won here, pulling real-time analysis (e.g., Taiwan’s DeepSeek ban, Italy’s AI app restrictions).
5. Final Verdict:
DeepSeek-R1: Best for coding, problem-solving, and humanizing content.
GPT-3.5 Mini/High: Faster for simple tasks but unreliable for complex logic.
Use Perplexity+DeepSeek for research/trend analysis.
TL;DR: DeepSeek-R1 outperforms GPT-3.5 models in coding and critical thinking. GPTs are faster for basic prompts but often fail under pressure. 🔍
3.5 is a different model, o3 is different,
Gpt 3.5 is a LLM that specializes in natural language processing, with lite reasoning skills, while o3 is mostly a reasoning model.
Gpt 3.5 doesn’t think, while o3 does.
Absolutely amazing how o3 mini didn't make you a snake game. It did for me several times without any errors.
That's what I don't like about LLMs right now, they are not consistent. Different people get different results. I hope all models will be more consistent moving forward.
That means their performance is not stable. Just like deepseek's server recently, always busy or "under attack".
My experience with OpenAI has been pretty good
Thanks great effort, precise comparison, DEEPSEEK R1 is 🔝
i think deepseek is damn good
And the Server are in China!!!
@joachimkluth like running it locally 🙌
I prefer deepseek.
@@joachimkluthBetter China than the USA...
@@denkot442 why are you using american invented internet, pc tech and websites?
O3 mini improvement is insignificant. For most of the general and coding usage, R1 is good enough. Honestly, for my case, the wait time of a few minutes compared to split seconds to get a viable solution is great, and acceptable. So even though o3 is a little bit "better", I will still not give up using R1.
I don't know why it's getting questions wrong in your case, I just tested o3-mini-high with the same prompts and it did pretty great.
Yes, same
Thanks for watching
man "flickering around" because you selected the html code on new positions all the time and Deepseek had his own window
There’s plenty of videos already out that show that the 03 mini model and the high model can make the snake game perfectly fine. There’s even a machine learning algorithms for it already.
If you use R1 through Perplexity AI, it enforces the Search option, which might reduce the quality of original thinking as it focuses on reporting what web sites say.
Thanks for sharing this video! Really clarifying things a lot..
If language models are used to automate the construction of knowledge graphs from unstructured input such as conversations, there need to be benchmarks for accuracy, coherence, and consistency of the constructed databases; and, measures of the latency and speed of output. The models should also be tested after being compiled to optimize the performance for specific hardware platforms such as smartphones.
Thanks for watching
Oh, closeAI released an amzing model~!
CH : release open source
US : itssss disruptive for ( our money ), it's just dangerously disruptive okaaay
Can you do Perplexity R1 vs Gemini 1.5 with Deep Search!
Dunno what's happening but in other videos i watched, o3mini is doing well in coding. Maybe you could redo the tests again in the future. I think there's definitely a mixup of results so far.
My experience is that o3 is great
Thanks for watching
When I tried in o3 mini high and deepseek R1 but asked them wrote in Python the results is quite the opposite. O3 mini high did it better.
Thanks for watching
My experience with deepseek was that it output sentences into each other. It couldnt absorb input, didnt work and when i was trying to get output it didnt give me any, just Went in a loop. So i dont really know what you are on about.
I use it for multiple linear regression, 1000 stocks with diagnostic and robustness checks.
OpenAI also would suggest portfolio weights based on financial information, it would compare results to research and financial theory
Thank you DeepSeek.
Funny how no one has mentioned SETI@Home as applied to AI development and/or communication, and/or Agents chatting with each other over the Internet and learning from each other as well or about the Internet, itself, becoming conscious.
So far, we all seem to have this idea of AI being developed and/or running on corporate server-farms/centers/clouds.
Bro i tried o3 in github copilot. It didn't even understand what i said to do let alone doing anything. I got scared a little bit as it was my main project file without committing in git .
Thanks for watching
Interesting tests!
For practical programming work Claude Sonnet is still the king. It just understands context of what I’m asking it to do much better and always delivers a nice ui.
Also, Deepseek is practically unusable, after a couple of long programming prompts it becomes “busy” or the chat is too long.
Nice ui? That's new to me, I used to create the logic in Claude, the interface in Blackbox, the aesthetics on Galileo and the assembling in GPT O1, I am glad rhat Claude has improved the UI accordinf to your comment
@@DetPrep Yeah, it does especially well if you have a design you like on other pages, and you prompt it to follow the design language/pattern of the rest of your application while creating something new, it produces ui design that is on par with our design team. So I get to essentially skip the design step as I can cod e and design at the same time.
Thanks for watching
The real strength of DeepSeek for is that the output in written language just feels way more human and your video seems to confirm that. For super corporate speak that doesn't sound human, ChatGPT is probably better... Language that mimics actual humans, DeepSeek is far better.
So still gonna use DeepSeek R1 and qwen. When OpenAI give unlimited then I will use that o3 mini
The humanizer test is only relevant to people who want to use this to cheat. And I bet that the 03 model relative to the 01 model is almost all synthetic data now or synthesized human data so I would actually expect that to happen like you said with the 01 mini in the R1 model steel based off of original base model text
Compare with Qwen pls
My genius analysis is to put the write-ups of both into Chatgpt and have it tell me which is better because I'm an idiot UA-camr
Sonnet probably still the best at web dev
Thanks for watching
It is not consensus R1 is better than o3 mini, other creators do the space invader test with much better results and other extreme tests that it doesn’t struggle with at all like this channel is presenting. I just made a space invader game work on a iPad touch screen with it with 0 coding experience.
Users are who win.
Define safe: What is the difference between sending out your data to the US or to China?
Besides, just use DS-R1 locally and, now, you'r data will be safe and it is free of charge...
The DS-R1-14B params works perfectly locally (using Ollama) on a M3 chip macbook Pro with only 16GB of RAM.
The difference in time of response of these 3 models is irrelevant. We are talking about few seconds more to accomplish what humans would do in weeks or months, just to say.
We'll see if you can build upon its model in the managers that open AI reasons do before I make any assumptions. I want already had to increase its input output price because of the server problems not being able to handle the loads. Do you believe a model that could really fit on your computer police in the near future surpass all human intelligence cognitive capabilities I'm not saying that's like the only Point behind having a decent model on your computer to help right now which is great of course. But either way yes I would rather give my data to the US I mean they already have it I guess it depends if you're from the US but China is going to use your information for different ways to actually use against us so if you're an American most people should probably want the United States to succeed and be the ones who have your data and as far as I know the US government doesn't have it anything you are creating you have the choice to actually turn off just a toggle switch and you don't need to have anything that you're using be seen inside of chat GPT unlike deep sleep where everything you use is considered direct property of CCP obviously probably that's what you have it on your computer a lot of people don't have that capacity and we want all the other benefits that come with the product not just the model itself. It's all about what we're going to do with these ai's make them more reasonable for the common man so to speak sorry for babbling I'm using voice to text so I'm not going off on a tangent just a bad habit LOL😂
@@michaelspoden1694 Don't get me wrong. I am not a China supporter. I was just saying that when we use online AI interfaces, our data is going away (even in the US, don't be foolish). And yes, I think we will have the power of using AI and build AI agents locally (it is already sufficiently powerful). Never the cognitive skills of human beings will be replaced (my thinking), we will just have to adapt and know how to use AI, totally different approach.
Thank you for your reply and thoughts.
I should have precise that I am in the middle (from France);-)
And I am confident that Chinese will set up more GPUs despite the embargo to be capable of dealing with the really fast increase of users.
Plus, using WebUI give us a way to use our models locally but, on our preferred web browser (Arc in my case).
@@michaelspoden1694 My point is different. Imagine if you are an ordinary American citizen, even if ccp got some of your data, they can't make use of it or at least they won't be able to knock on your door. But if CIA got it and you said something they don't like you will be in trouble just as those whistleblowers recently (for example Suchir Balaji from openAI)
R1 beast
We demand a recount - the tests are rigged :)
WAIT a sek?! Are you sure your "html snake game" is a challenging task? Because i just asked my local Qwen2.5-14B-Instruct-1M.Q4_K_M (a freakin 14! B model), and it gave me working code! Ok, not gonna lie, at first it had a little bug, when opening html, it said game over in a pop up window, but i saw the board and snake+food. So i just asked it to fix the bug (telling it exactly what i see). And it fixed it! 14B model that you easily run locally. It's not even a thinking version R1-Distill-Qwen-14B, you read it correctly.(actually thinking version is BAD imo)
Hindus are already worshipping R1 and I heard they've built a temple already 😂😂🤣🤣
You my friend deserve more likes
I don't think so. Source?
Remember when he did SEO?
R1 feels better overall for me
Thanks for watching
01:29-04:29
I tried o3 and it sucks
most of my job candidates fresh grads behave like chat-GPT. Fast answers and totally wrong. While people with just high school they think and if they don't know, they tell.
sounds really nice, just like I made deepseek uncens0red on my youtube channel
thanks
Thanks for watching
Please change the name to CloseAI
I testes my articles (written by me) and it showed 98% AI, so either the ai detectors are BS, or I'm a terminator to destroy humanity.
You asked the most basic questions possible for the most advanced models out right now 🤦🏽♂️
I create marketing content and honestly o3 mini high surpass R1 deepseek just need clear instruction and clear example I copy paste from R1, sad part is open ai will not make that "o3 mini high unlimited" if it wasn't to challenge deepseek R1, so open ai still untrustworthy. I do not mind to pay 20$ monthly to pay fast and great AI assistant, but openai is my very last option knowing they try stab us with 200$ PRO already.
😂 call the R3 as if they're going to skip the two like open AI did SMH
It's not a fair comparison, not even sure why you bothered to do a video like this.
Open AI has all the benefits of using the best chips and unrestricted by sanctions.
Deep seek is sanctioned and running on a budget. Despite that They even open sourced. Closed AI learned from them after they dropped the manual on how to be more efficient with ai algorithm on top of opening up the hood for all to see.
To top it off all they were running on limited chip hardware due to sanctions.
How about tulu?
You should test 01 pro
I've been using that model for a month. I don't see the difference between r1 and o1 pro
Deepseek hasn't been connected to the internet for 6 months. it's not even close to GPT imho
Glazer. Its close for the reason stated, plus if you need internet, use google or gpt4, not being close is incorrect
Thanks for watching
Something is off with the coding test
Deepseek is facing huge traffic . So it is unfair
IF NOT BECAUSE DEEPSEEK
OPEN AI WILL CHARGE US $2000 PER MONTH 😂😂😂
Denfinitely a R1 biased testing video. Try this not so simple math question: the sum of 4 real numbers is 9 and the sum of their respective squares is 21. What is the minimum sum of the 2 bigger numbers?
Also, using a machine detector to evaluate text is human like or not is ridiculous. On 1 side, it only says how much it’s gpt like, not human like. And it could be just designed to test chatgat and doesn’t know much about deepseek. On the other hand, if a machine can detect how much human like , it can already output human like texts
After 117 seconds of thinking, deepseek provided the answer 5. It is uncertain whether this is correct. In its reasoning, it employed the Cauchy-Schwarz inequality and the Lagrange multiplier method.
So ClosedAI is coping the Chinese, but it's still worse.
What about Qwen 2.5 max?
远远超过deepseek和openai的任何产品
中國不可以“抄襲“我們的東西？然後川普可公然説：格陵蘭，巴拿馬運河，墨西哥灣（美國全要）？？
美國道德人士的聲音呢？
US AI industry quiet nowadays. Once China starts to dominate in AI they have lots of China AI industry 10x more than American counterparts. If that's the case China already dominate. Deepseek then Qwen, and the other China AI follows. Wake up US. Where are you? You need cooperation with Allies encourage your allies to have strong AI development in order to win.
中国国内AI竞赛也很剧烈，下个月R3发布（没有R2）
Allies? Like France's Lucie?
@@cherrylee9591 Hahahaha..mmm
🤑🤑🤑
third
nice
The test was funny where you had to change the text to check or AI wrote - and doesn't it bother you that it is the system that determines ChatGPT? Even its name is like that) The question is that such services look for patterns in the answer - accordingly, GPT was the most popular and on its basis they almost trained models for detection. But DeepSeek has not been tested and has not been searched for yet) That's the whole difference
For the Newbie if you are actually trading in the crypto space and you don't have a sound mentor. Then you are certainly going to get liquidated in 90% of your trades. Yeah that's sad truth. I remember when i just got into crypto back in 2019 but later in 2020 i ended up selling it because i have lost alot trading all by myself without a guide. Got back into crypto early in 2024 with 2.3b"tc and now I'm up to a decent 18B'tc in just a short period of time. Thanks so much coach Evelyn Mya...
I’ve seen many positive comments about Evelyn Mya. Can someone guide me on how to reach her?
She mostly interacts on Telegrams, using the user-name,,
@Evelynmya
The beauty of Evelyn Mya’s approach is her dual focus: while she aggressively pursues profit opportunities, she's equally tenacious about shielding investors from potential pitfalls. It's a balance few can achieve.
Investing has proven to be an incredibly beneficial decision. My cryptocurrency profits continue to play a substantial role in growing my overall wealth, reducing my reliance on my salary.
First
nice
waiting deepseek new version 💀
3月份r3发布，官网确定的
i'm running r1 offline. it's so much better. i run it locally on my gpus.
GpuS??????? I don’t even have 1!
@Antileaves you can run the gguf models locally on your phone too. do you have a decent phone? min ram required is 8 gb
Chatgpt is the winner 👍
how? did you watch the video?
@eihero yes, let's wait for sometime. The accuracy will fine tune.
I use both deepseek and chatgpt.
But in some extent chatgpt is the most refined answers.
For building some gaming task cannot determine chatgpt strength.
@@eihero to be fair I tested the same prompts in o3-mini-high and it did well on the first attempt. Maybe he was just unlucky.
Disappointing message limitations though.
Obviously not, according to this video.
Deepseek sucks in coding even provide official docs