Deepseek R1 - Quick Testing!

[@1littlecoder]

Local Installation of Distilled Deepseek R1 tutorial - kzbin.info/www/bejne/nqPYeGCAobGYh8k

[@miken3d]

another awesome video, thanks!

[@1littlecoder]

Glad you enjoyed it!

[@henkhbit5748]

Really impressive, will test it also. Thanks for the update👍

[@1littlecoder]

@@henkhbit5748 thank you

[@jamespat7975]

@@1littlecoder Question : 8 students(namely 1 to 8 ) are arrange to sit at the seats in two row of 5(meaning there are always two empty seats). Find possible numbers of arrangements in which students 1,2,3 sit next to each other, 7 and 8 not sit to each other and that two empty seats must be next to each other. My mannual hand calculations answer = 15,264, anyone can use ChatGPT o1 Pro to calculate it ? Can deepseek get the same/correct answer ? Any can try to calculate with deepseek ?

[@설리-o2w]

Absolutely stunned-this truly feels like we're inching closer to AGI, especially for open source! This should prompt OpenAI to focus on innovating rather than resting on their laurels. And those criticizing it just because it's a Chinese model are showing their ignorance and lack of open-mindedness.

[@Atheist-Libertarian]

Chip export restrictions on China 🇨🇳 must be lifted. Imagine what they would create if there were no Chip constrain 🤯 And they will eventually Open Source it, So it's good for whole world. 🌎

[@1littlecoder]

@@Atheist-Libertarian what if they don't open source in that alternate universe?

[@abdelouahabtoutouh9304]

The constraints are what sparked their creativity! They had to accomplish all this with minimal resources. Ironically, if they'd had access to abundant GPUs, they might not have achieved such success!

[@KevinKreger]

The paper is fascinating.

[@1littlecoder]

Thanks for sharing, I need to spend some more time on it!

[@amoledzeppelin]

Please also test the distilled 1.5B and 7B models. I was impressed by the 7B one.

[@emport2359]

Great Testing man, much better than other youtubers, you actually think about it's thoughts and outputs

[@itlearner1175]

Why do india has so many IT engineers but not building any model like deepseek?

[@NadeemAhmed-nv2br]

Money, very poor country and very high taxes. Importing one GPU, the taxes triples the price in the india. We're a country in which ninety percent of the population has a standard of living lower than sub saharan africa so most people cant afford to pull something like this off, china's standard of living is atleast 5 to 10x higher which makes alot of things possible

[@KevinKreger]

@@itlearner1175 all the top talent was drained to US?

[@52joelsharonr60]

Fund, India has the capability to do it, but not necessary fund for the R&D.

[@ShivamPradhan-c1x]

india has so many IT engineer but not so many IT product company service based company work for other so they are not interested in building funds for GPU only for researching , that we don't do here

[@Trials_By_Errors]

​@@NadeemAhmed-nv2brYes, taxes on GPU's must be lowered. And the living standards of 90% of Indians isn't below Sub-Saharan Africa.

[@alexsov]

i have done a test for o1 - large specification for complex function implementation (with acceptance tests it must conform). and o1 implemented fully function code in ONE prompt! and all tests was ok! 4o - implemented but not all tests passed. same with deepseek - 1/3 of tests is errored...

[@1littlecoder]

that's interesting, which language is it?

[@alexsov]

​@@1littlecoderIn TypeScript

[@jamespat7975]

Question : 8 students(namely 1 to 8 ) are arrange to sit at the seats in two row of 5(meaning there are always two empty seats). Find possible numbers of arrangements in which students 1,2,3 sit next to each other, 7 and 8 not sit to each other and that two empty seats must be next to each other. My mannual hand calculations answer = 15,264, anyone can use ChatGPT o1 Pro to calculate it ? Can deepseek get the same/correct answer ? Any can try to calculate with deepseek ?

[@mrrubel8841]

Very nice , informative content. I am your regular viewer.

[@1littlecoder]

@@mrrubel8841 thank you sir!

[@jezf4240]

Excellent review brother!! I'm heavily impressed with DsR1. Squeaky bum time for VC's I think...🤭

[@1littlecoder]

thankk you! i just built a stupid local agent but that's a great start to do with these models locally!

[@TheRealHassan789]

amazing vid master-ji :)

[@1littlecoder]

Thank you ji!

[@SteveGamesOnline]

i would consider this model as superior to o1, not just because of the quality of the responses, but also because of the efficiency this model can operate.

[@VaneNickOke]

I love your scientific approach. Well done.

[@1littlecoder]

thank you, any feedback to improve further?

[@nickernara]

oh boyy, this one is amazing. i shd start running it locally and try it out.

[@1littlecoder]

@@nickernara in case if you need assistance kzbin.info/www/bejne/nqPYeGCAobGYh8ksi=BrCfigb2p8xQ1dd5

[@thoughtfulcomet]

whatever paid version of LLMs US companies launch chinese reverse engineer it & open source that

[@1littlecoder]

at least they're open sourcing, which is a good thing!

[@thoughtfulcomet]

​​@@1littlecoder i believe in robinhood

[@ZuckLogic]

If it werent for china's cloning , cheap mass production , majority of world would not have afforded anything due to US , Europe's Monopolies

[@minimal2224]

I love that they work so damn fast too lol

[@alan83251]

I’d rather that than the US companies being the only entities who have the capabilities.

[@GNARGNARHEAD]

great set of tests 🤯that's impressive

[@swamchem]

Great demo with impressive test set

[@aculz]

whatever LLMs US companies launch , then chinese will reverse engineer it & open source that. and dont forget, they lowering down the price to be extremely cheap and beat their value! which is AWESOME you know what is more shocking ? their even made this super huge good model free if you using it on their sites. and no need to pay $200 to use it on their own platform like o1 i love this company !! glad to know this also thanks for your amazing videos 🤩🤩🤩

[@1littlecoder]

thankk you! rightly said, Innovation at the core is expensive - definitely respect for openai!

[@Saurav-xx]

R1 model really impressed

[@CarlosValero]

Amazing!

[@ElevateMotivatee]

Ok another AGI model lol, ok here is a question for you. How does this model adjust to dynamic changes, adapt to changes like our brain, what's its reasoning process for changes, how does it adapt? Our brain uses a process called Neurol Plasticity to correctly adjust our neurons and synaptic connection to changes and adapt to changes. Can this model dynamically adjust its weights and bias, adjust its connections based on different situations/patterns. The test time and inferent scaling, is basically a fancy abbreviation for Reinforcement Learning, where the model is rewarded and penalized for its answers.

[@jamespat7975]

Question : 8 students(namely 1 to 8 ) are arrange to sit at the seats in two row of 5(meaning there are always two empty seats). Find possible numbers of arrangements in which students 1,2,3 sit next to each other, 7 and 8 not sit to each other and that two empty seats must be next to each other. My mannual hand calculations answer = 15,264, anyone can use ChatGPT o1 Pro to calculate it ? Any can try to calculate it with deepseek ?

[@RideauVert-k7v]

Dammmmm 🎉🎉🎉🎉 thanks for sharing !!!

[@1littlecoder]

🤗

[@07Mihai07]

Guys, to try out LLMs is nice to have a list of questions most of the get wrong. I want us to make a list here so I will start below.

[@07Mihai07]

The United States's succession of presidents in famous for switching from a democrat to a republican then back to democrat and so on. There are very few cases in which a president (either republican or democrat) that enjoyed 2 consecutive mandates is followed by a president from the same party. Can you make a list of such situations and the presidents names?

[@07Mihai07]

Anna has 3 brothers. Each brother has 2 sisters. How many sisters does Anna have?

[@Aiworld2025]

Inside out is a great way to describe how this happening, I mean without all the emotions 😅

[@1littlecoder]

@@Aiworld2025 thank you. I'm glad someone noticed it!

[@yajush-t4w]

13:12 wait! who is 'we'? Who are they??

[@john_blues]

Hi. Just wanted to share this chat I had with R1. It was on Deepseek's website. I was asking how to enhance a Google Sheet. It got totally confused. Figured I'd share this real world test with the community: (Excerpt) They need to type "🔍 Search term..." directly into cell B1. Then, create a conditional formatting rule that changes the text color to gray and italicizes it when the cell is empty. Wait, actually, if the cell is empty, the text wouldn't be there. Oh, right, the idea is that when the user clicks on B1 and starts typing, the placeholder disappears. But Sheets doesn't support dynamic placeholders like that. So maybe the correct approach is to have the placeholder text appear when B1 is blank. But how? Ah, perhaps I made a mistake earlier. The user needs to have the placeholder text in B1, but when they start typing, it replaces the placeholder. But since Sheets doesn't have built-in placeholders, the alternative is to use a script or a formula. But the user might not want to use scripts. So the best way is to manually instruct the user to type the placeholder into B1 and use conditional formatting to make it gray and italic. Then, when the user types something, the formatting changes. Wait, no. Conditional formatting can change the style based on a formula. So if the cell is blank, apply the gray italic style. But if the user wants the placeholder to show when the cell is empty, that's not possible without a script. Wait, maybe the user can type the placeholder text in B1, and then use conditional formatting to hide it when the cell is not empty. For example, set the text color to white (or the cell's background color) when there's input, making the placeholder disappear. But that's a hack... (Excerpt)

[@Sri_Harsha_Electronics_Guthik]

Absolutely love it

[@wardakhan4911]

Can you show some use cases of how to build ai agents to solve a problem using this model.

[@karanon6367]

This is a wow moment similar to when I used the GPT for the first time

[@rohulamin496]

With sincere appreciation from 🥰 Pakistan 🥰 for dedicating your valuable time to share enriching content.

[@1littlecoder]

Thank you!

[@rohulamin496]

@@1littlecoder Thank you for your response! May Allah bless you with happiness in your life.

[@sohamnimbalkar66]

Wow it is amazing I mean it is Opensource my god although it is chinese but has to be appreciated 🎉🎉

[@jsbgmc6613]

Thinking process is more interesting than the answers 😊. I wonder if this is a learning opportunity ... We/kids? learning from AI how to think.

[@awaqken]

Build a SaaS, move to production, and maintain it using only Deepseek R1.

[@AlmightyGod.321]

Is it able to solve coding problems?

[@ysy69]

Impressive!

[@RideauVert-k7v]

Can eait to get it woth the nvidia hardware comming in may

[@1littlecoder]

You mean wait ?

[@RideauVert-k7v]

@@1littlecoder yes the Project Digits

[@ysy69]

Mind blowing

[@indrajitnaskar6851]

Is there any limit in deepseek api key?

[@1littlecoder]

I think it's paid

[@kirantej6690]

no it has no limits for real

[@indrajitnaskar6851]

@@kirantej6690 What do you meant by real

[@alan83251]

I’d rather some of the most capable models be open source and the CCP also has them, than they all be closed-source and solely in the hands of US mega corporations.

[@xolo2617]

after trying 2 or 3 times it actually solved recent leetcode weekly contest hardest problem , super amazing , really excited for the future of AI.. I am testing it and will test in live contest how it will perform.

[@leeme179]

You should do a video about "Suchir Balaji" an OpenAI whistleblower that died

[@mdmishfaqahmed2138]

And also has the hardware infrastructure for it. 🤷

[@1littlecoder]

after restrictions on GPU purchase!

[@swamchem]

Try checking with different languages

[@256chiru]

Please make video for kimi k 1.5 similarly scoring and just now released benchmarks please check x

[@xXWillyxWonkaXx]

They released another model other than their v3 wtf 😅😂 it’s impossible to catch up with all this

[@Aditya-rs5dj]

TRY THIS : Check the model's reply when you ask questions of Geopolitical importance framed against china..

[@Tanvir1337]

Open Source FTW

[@Saurav-xx]

btw i'm IIT aspirant

[@captainoddessy]

Please try improving your video quality using gemini live. Show it your video, and ask for suggestion to improve. I think using you pixel phone ( i saw it on one of your video) you could vastly upgrade the visual quality. I really wanna see your channel grow.

[@1littlecoder]

Thanks for the suggestions. I recorded this with Mac camera. I used Pixel but somehow it was a mess to setup in the right way. Perhaps first step is to buy a light

[@1littlecoder]

Btw is the lighting bad in this? I used a slightly different setup

[@captainoddessy]

@@1littlecoder it is. Your face is over saturated. And some part is over exposed. Buy a tripod. Lock focus and exposure and transfer it to your Mac. And do the editing

[@jackbauer322]

I really don't like the way your going through your video names ... you were more down to earth at your beginnings

[@dadadies]

At least it's not made by the US or Israel.

[@sohamnimbalkar66]

As it is chinese I don't know why but I think data used for training might not be completely legal 😅😅 because it is China at the end Although One of the Best model out there 🎉🎉

[@1littlecoder]

haha

[@TaHa-nf5vc]

Similarly, is OpenAI really transparent about their data ? factually we don't know for sure, we can speculate yes, and it'll be stupid to believe they're all trained on 100% legal data, for any of them.

[@mysteriouswonders528]

Who makes it legal😂😂😂

[@sohamnimbalkar66]

@@TaHa-nf5vc agreed

[@sohamnimbalkar66]

@@mysteriouswonders528 😅

[@Useryskk9079]

0:28 spyware

[@1littlecoder]

🤣🤣🤣🤣

[@thakkalikuttu8896]

😂😂😂

[@leisureclub_]

You should ask Questions about Palestine from OpenAI models (basically American models) and let me know what response you get. Same goes for Chinese models... so no surprise..

[@1littlecoder]

Absolutely true and it is of my opinion as well that people hardly criticize us models

[@sagarranjan3385]

please call it ASI

[@wilfredomartel7781]

Pardon my ignorance, but why ASI?

[@1littlecoder]

same question - why?

[@sagarranjan3385]

@@wilfredomartel7781 Going back to the basics of building LLMs (Large Language Models): traditionally, these models are trained on human data to copy our behavior and outperform us in specific tasks. This is what I would call an Artificial General Intelligence (AGI). However, they are now leveraging Reinforcement Learning (RL), similar to what was used in AlphaGo. AlphaGo was able to defeat the best human players because it wasn’t solely trained on human data. This enabled it to make the famous Move 37, a move that no human had ever considered before. Similarly, in this context, they are creating synthetic data through RL. They understand that relying solely on human data won’t lead to Artificial Superintelligence (ASI). To achieve ASI, they are generating this data through deep, strategic thought processes. ASI will require data that goes beyond human limitations,so basically we internally asi achived by deep seek and oepn ai ,we have build machine god

[@jimigoodmojo]

Artificial Super Intelligence?

[@sagarranjan3385]

@ Going back to the basics of building LLMs (Large Language Models): traditionally, these models are trained on human data to copy our behavior and outperform us in specific tasks. This is what I would call an Artificial General Intelligence (AGI). However, they are now leveraging Reinforcement Learning (RL), similar to what was used in AlphaGo. AlphaGo was able to defeat the best human players because it wasn’t solely trained on human data. This enabled it to make the famous Move 37, a move that no human had ever considered before. Similarly, in this context, they are creating synthetic data through RL. They understand that relying solely on human data won’t lead to Artificial Superintelligence (ASI). To achieve ASI, they are generating this data through deep, strategic thought processes. ASI will require data that goes beyond human limitations.

[@alpha_centauri_]

Ask about Tiananmen Square

[@joneslee9084]

sb

[@jsbgmc6613]

Answers without thinking... Its a no brainer question 😂