@@KyleKabasares_PhDnot moving... Actually Logan, of Google AI studio is trying to troll OpenAI before releasing Gemini 2.... These are not official model but probably some previous trained checkpoints of Gemini 2
3:40 Multiply by sqrt(2) up and down, to eliminate the sqrt(2) from the denominator - you gain a sqrt(2) in the numerator, and the denominator becomes just 4.
Ran it on deep seek and what it did was really clever and human like. It split the integral into separate parts that it knew from memory and solved it.
@ One of the most impressive experiences I had with it is when I asked it some common sense physics problems about accelerating a spaceship via a pushing laser. I gave it the constraint of limiting itself to a 1 terawatt laser and maximizing the acceleration of a 500 ton payload and it figured out not only to increase the number of reflections between the source and the payload (making an analogy to optical resonator cavities) but also to split the payload into multiple pieces, accelerate them separately and then re assemble them on route. Another question I asked both it and gpt was to derive coloumbs law in a hypothetical situation with three different types of charge. Both came up with the solution of representing the charges as vectors in an abstract charge space and using the dot product to calculate force between them. A great thing to do with them is to give them open ended problems where you can see how they come up with solutions and whether or not they understand the implications of these solutions.
MathCad 4.0 from 1993 could do the integral right away, in less than a second. I bought this wonderful program for Windows 3.11, and it still runs on Windows 10 and is my favorite math program.
@@KyleKabasares_PhD MathCad is a relatively expensive, proprietary math software. Today's version costs $780 per year. I bought it on special offer for $150 back then and still use it to this day. MathCad integrated the then leading MAPLE (computer algebra system) software.
Put the ultimate pressure on DeepSeek until you see its limits. This is the lite version as well, Oh yes, i forgot, it is a preview of the lite version, and it will be open sourced as well.
I think we need more challenge .. something that the models cant find in their data ... Like some calculation that humans right now are working on ... And then when we confirm the right answer ... We can check if the AI got it right
You wouldn't believe me if I write here that even ChatGPT-3.5 could initially solve such integrals provided you used CoT (but since the nerf, CoT doesn't work with ChatGPT-3.5 anyway.). To make Dan do it, you would need to decompose the question into stages and feed him in TeX, just like you do here, the equations, explain the variables, give example of a solution of a stage and ask him to solve in the same way. Sure, once they started resetting every prompt this stopped working, but it did work. And I have screenshots of Dan actually solving such problems and even more difficult ones.
So, solving those integrals by human was not pratically useful in past, now some AI would be doing these job. If that's a useless job we want to have AI doing, I'm all in.
I broke deepseeks chain of thought thing and it just started infinitly spitting out the same Chinese character. I had to start a new chat because it just wouldn't stop. Maybe it was because I was asking it for uncensored content, which it will make if it's certain that you're not asking for something unethical. It seems to work a bit differently to 01 preview. I know 01 preview hides the full chain of thought, but still it seems like deepseep is using a different technique. I forget the difference.
Gemini experimental 1114 is already outdated... they just released experimental 1121. No joke 😂
Woooow 🤠
Already?!!! Things are moving way too fast
@@KyleKabasares_PhDnot moving... Actually Logan, of Google AI studio is trying to troll OpenAI before releasing Gemini 2.... These are not official model but probably some previous trained checkpoints of Gemini 2
Damnnnn Google came back swinging! Super impressed
you are back just in time, the future seems bright with that kind of progress speed
3:40 Multiply by sqrt(2) up and down, to eliminate the sqrt(2) from the denominator - you gain a sqrt(2) in the numerator, and the denominator becomes just 4.
Ran it on deep seek and what it did was really clever and human like. It split the integral into separate parts that it knew from memory and solved it.
@@MT-xu7dh Wow that’s amazing
@ One of the most impressive experiences I had with it is when I asked it some common sense physics problems about accelerating a spaceship via a pushing laser. I gave it the constraint of limiting itself to a 1 terawatt laser and maximizing the acceleration of a 500 ton payload and it figured out not only to increase the number of reflections between the source and the payload (making an analogy to optical resonator cavities) but also to split the payload into multiple pieces, accelerate them separately and then re assemble them on route.
Another question I asked both it and gpt was to derive coloumbs law in a hypothetical situation with three different types of charge. Both came up with the solution of representing the charges as vectors in an abstract charge space and using the dot product to calculate force between them.
A great thing to do with them is to give them open ended problems where you can see how they come up with solutions and whether or not they understand the implications of these solutions.
MathCad 4.0 from 1993 could do the integral right away, in less than a second.
I bought this wonderful program for Windows 3.11, and it still runs on Windows 10 and is my favorite math program.
@@OrbitTheSun where was this when I was in college lol well we had WolframAlpha by then
@@KyleKabasares_PhD MathCad is a relatively expensive, proprietary math software. Today's version costs $780 per year. I bought it on special offer for $150 back then and still use it to this day. MathCad integrated the then leading MAPLE (computer algebra system) software.
glad you're back!
Thank you! Glad to be back as well :)
Bro Try 1121 Gemini newly released
My o1-pre got it correctly twice, what is this?
bro, you can make into an official llm math torture channel!
Yup. Talking with DeepSeek with DeepThink ON is weird. It's like talking with someone and see his thoughts.
Waiting for more.
Really liked this testing. Do more questions like this.
Will do :)
Can you make nanay solve the integral?
lol i don't think she would like that
Put the ultimate pressure on DeepSeek until you see its limits.
This is the lite version as well,
Oh yes, i forgot, it is a preview of the lite version,
and it will be open sourced as well.
Still waiting for full o1 been in preview for too long now seriously
I think we need more challenge .. something that the models cant find in their data ... Like some calculation that humans right now are working on ... And then when we confirm the right answer ... We can check if the AI got it right
You wouldn't believe me if I write here that even ChatGPT-3.5 could initially solve such integrals provided you used CoT (but since the nerf, CoT doesn't work with ChatGPT-3.5 anyway.). To make Dan do it, you would need to decompose the question into stages and feed him in TeX, just like you do here, the equations, explain the variables, give example of a solution of a stage and ask him to solve in the same way. Sure, once they started resetting every prompt this stopped working, but it did work. And I have screenshots of Dan actually solving such problems and even more difficult ones.
DeepSeek and Gemini 1114 are very intelligent and i hope you will ask these new AI's more math questions!
Thank you sir. Always a pleasure
@@ziadnahdi4343 thanks for watching!
Finally you're back, was wondering where's the new vids
DeepSeek is amazing, also as you said, a bit unsettling
Gemini Experimental 1121 is out!
DeepSeek is pretty good at coding as well: writes more efficient code compared to o1.
So how did the AI handle the infinity?
Or you could do it by contour integration.
other than math, i dont find deepseek that impressive
waiting for more upcoming livestreams'
Brilliant
❤
It writes without proper grammar because it's Chinese company that made this.
Oh got it, thanks for the clarification! Wow their model is really good and kind of came out of nowhere!
3:22 lol phd bro ?
lol yes even Physics PhDs can struggle with basic math
So, solving those integrals by human was not pratically useful in past, now some AI would be doing these job. If that's a useless job we want to have AI doing, I'm all in.
Can it do my laundry and my dishes and take out my trash yet
@@KyleKabasares_PhD man, it's already automated. Dishwashers, laundry machines ...
@ but i still have to put the dishes in the dishwasher and collect my laundry and fold it!!
Meanwhile his response :
There are 2 r's in 'strawberry'
Whatever a chinese sees today, you'll get a copy tomorrow
shit talk all you want, but the chinese have consistently lead the open source efforts so far, in LLMs and Video Gen at least
That's why they are so good
I broke deepseeks chain of thought thing and it just started infinitly spitting out the same Chinese character. I had to start a new chat because it just wouldn't stop. Maybe it was because I was asking it for uncensored content, which it will make if it's certain that you're not asking for something unethical. It seems to work a bit differently to 01 preview. I know 01 preview hides the full chain of thought, but still it seems like deepseep is using a different technique. I forget the difference.