This free Chinese AI just crushed OpenAI's $200 o1 model...

[@DataIsBeautifulOfficial]

Capitalism just got outsmarted by open source.

[@FeLiNe418]

a.k.a communism

[@whoamitotheworld4567]

at the end of the day it would benefit capitalism.

[@trickeddev]

ccp saving the day yet again ❤❤❤

[@randomuser66438]

​@@FeLiNe418communism heavily funded by big tech

[@nothingtoospecial777]

@@FeLiNe418open source software is not communism lol.

[@retroverdrive]

00:37: "Closed AI companies like OpenAI."

[@epixerty]

ironic isn't it

[@marsh346]

yes, indeed that is the joke

[@Happ1ness]

Based

[@Psykoses]

@retroverdrive: "00:37: "Closed AI companies like OpenAI.""

[@kofi3124]

Unfortunately it's true.

[@Nicholasorsomething.]

I don't care about ai but god damnit do i love to see free and open source software competing with the big boys

[@BBWahoo]

I'm gonna try it, never cared for AI either

[@herrpez]

​@@BBWahooIt's fun to mess with. The previous version of deepseek would not, for some strange reason (it's not strange) answer stuff about Tiananmen Square and other Chinese things. Most peculiar (not peculiar).

[@abhorrentweeaboo4104]

most open souce software has always been better than the proprietary ones after all when you don't have shareholders breathing down your neck demanding infinite growth you have no incentive to ensh!ttify your product for short term gain.

[@BBWahoo]

It's fun to mess with, until you end up with@@herrpez

[@xninja2369]

​@@abhorrentweeaboo4104BRO as supporter of oper source NO its bog lie until recently open source software were always behind. .

[@dgw8514]

很重要的一点在于在这个地球上不是所有国家都能像一些美国的大公司那样有能力训练自己的AI，因为那样的AI成本很高，但是现在deepseek的出现可以帮助很多国家的人跟上AI的发展

[@Aymanne-io6vb]

We thank the chinese, may God bless open-source and the chinese(especially those involved in deepseek). Ameen.

[@manuelapeters1059]

Glory to han china

[@electron6825]

China is West Taiwan

[@yuchenzhang9581]

@@electron6825 😅in fact,taiwan is just a part of China

[@MaxMax-nn8ur]

@@electron6825 But taiwan is officially called China.

[@jianxi]

The chinese basically did what OpenAI was supposed to do... which is to be free and accessible to everyone.

[@HiteshP-l7v]

The entire SaaS and software industry is the same message right now. Like how did they even assume I'm gonna pay for 10 different tools for every small thing I do? Everyone should collectively rethink subscriptions or simply put offer every subscription in a single payment depending on needs and that'll be more reasonable. Let these companies get paid in subscriptions instead of us paying every now and then.

[@st.altair4936]

I wish it was fully state-funded but at least it's FOSS ig

[@raghnallm4004]

First loser will always go for the open source blue shell. Everyone knows that if the Chinese were number one, they would be going fully protectionist as well

[@Hinterfrage]

well r1 is a woke trash ai, worser than OpenAI in this point!

[@st.altair4936]

@ Why am I not surprised chuds hate everything good

[@El_Bartto]

Deepseek is also available worldwide unlike chat gpt

[@dragonverde188]

and is free and is open source and is not made for profit, and it was founded with much less resources, makes you think how much of US AI is a bubble

[@yauyuso]

Not it's not. Apparenlty it's limited by region

[@thatguyalex2835]

Agreed. Sadly, the largest model I can run on my computer is Mistral Nemo (12 billion parameters). Deepseek is a good idea for folks who have mainframes. God bless open source.

[@Maplin-Ma]

@@thatguyalex2835 How about rent GPU online?

[@phoenixeyy]

deepseek is not available in Australia

[@matt_milack]

If optimists are right, we're heading towards bizarre dystopia. If pessimists are right, we're heading towards bizarre dystopia.

[@dimitardobrev3296]

The optimists want it, the pessimists not so much.

[@Elongatedrider]

So it's a win win anyway ~ the government probably

[@boxboxerson991]

in what regard does the pessemistic take lead to that? come on people use your brains. are you software engineers or larping as one?

[@matt_milack]

@@boxboxerson991 Just observe politics events in the last, like 3 weeks, and tell me we are not on our way towards bizarre dystopia. Sure, it would be dystopia with much more people having their jobs, however . . .

[@boxboxerson991]

yeah no

[@pctong5387]

Waiting for all the western media to start saying this is a security risk

[@CarlyGayJepsen]

They already are 😂 they’re even calling it a “non-democratic ai” now 💀

[@djkidii]

how ? its open source :D jezus

[@cccc2740]

It refused to comment on tianmannen square 😂😂😂 so definitely a risk

[@johnclark5115]

AI communist deep seek you

[@elrond7887]

the source is open, the account you create to use it has potential security risk Critical thinking

[@Cavz001]

Rack overflow is crazy 😂

[@migovas1483]

laser eyes!!!

[@LuisSierra42]

The Zuckerberg situation IS CRAZY!!

[@user-tr1zj]

Zuck is really staring at that "silicon" valley

[@Cavz001]

@@LuisSierra42 he maybe getting little to no action from his girl tbh

[@CyanRooper]

​@@user-tr1zj uh ackshually they're made of silicone not silicon. ☝️🤓

[@allmightjunior6917]

"Closed ai companies like open ai" 😂😂

[@meetit5949]

A closed AI company, OpenAI, will release a below human-level intelligence they call ‘AGI’.

[@moormanjean5636]

Exactly what they are

[@artman40]

Funny that if I ask ChatGPT to list closed AI companies, the first example is OpenAI.

[@Rami-bi9xj]

@@artman40 thats one open part about it

[@FireFox64000000]

I'm convinced fireship didn't even try and write that joke. It just happened by accident.

[@bause6182]

It's so satisfying to see openai being overtaken by open source. Without all the open access literature open ai would never have been able to realize their gpt models.

[@000aleph]

the string 'open' appears 3 times in this comment

[@Ginsu131]

I mean, isn’t that almost all products by any company?

[@NostraDavid2]

Not just open source. Open source from China. Fucking China is eating the USA's breakfast. 😂 It's all MIT licensed BTW, which is better than what Llama has.

[@Rami-bi9xj]

@@000aleph >'OpenAi'.includes('Open') >false

[@Noredia_Yuki]

Even the transformer architecture was not created by them

[@Sekiberius24]

Deepseek just showed the entire world how overvalued and overpriced the entire AI industry was, it was never going to be as profitable as the venture capitalists wanted it to be.

[@Abdulkadir-yq2vm]

You are %100 right, companies like open ai try to convince us making AI is hard so we have to pay it more.

[@oompalumpus699]

​@@Abdulkadir-yq2vm _I remember how almost a year ago I got laughed at for saying that 'OpenAI' will eventually paywall all of its users by shooting its subscription pricing through the roof._ _I also got laughed at for saying the future is not big corpo AI but it should be AI that everyone can access and build like ordinary applications._ _It feels so great to see the hypebros for 'OpenAI' get so mad._

[@amarug]

China of all countries putting this out free and open-source. Gotta absolutely love this timeline, there is a plot twist at every corner...

[@savejeff15]

Agree but just for the record: llama is also open source and from meta. And Nvidia also published their LLM

[@azarak34]

When us needs to care about copyrights and walled content. China has all access to data they want. Them winning the race was just a matter of time.

[@zaryabK-vi8fh]

@@savejeff15 but are they good and on level

[@de132]

​@@savejeff15 Meta's LLMs aren't truly open source, there's constraints commercially

[@aminulhussain2277]

​@@savejeff15 They aren't as good as Deepseek though.

[@Jaob994]

I love how he already put in all the memes from today, like the "salute" from Musk and the thing Zuckerberg now has to explain to his wife

[@crazycowboy213]

his wife has a flat chest so he doesn't have to do any explaining lol

[@blackrack2008]

This guy working overtime

[@MontyMcRib]

theres absolutely nothing wrong with glancing to confirm that some woman is indeed dressed like a rostitute.

[@DR_1_1]

Are they memes of fakes? I mean these are not the real images. That's the main issue I have with AI, it's basically a forgery machine, a way to bypass reality, authors, copyrights, and more.

[@markvenuto9383]

​@@DR_1_1both are real, you can watch videos of each of them happening from yesterday's highly publicized US presidential inauguration

[@Shahpo]

Just tested it with some of my PhD level string theory/general relativity questions. Honestly does better than O1. Impressed.

[@AgoniUnconditional]

Are you doing a PhD in physics??

[@nightfall8732]

Doing PhD on which topic? And from where?

[@Shahpo]

@@nightfall8732 AdS/CFT type stuff. I asked it to explain why the Hamiltonian for an observer in de Sitter space is unbounded from below. It gave a much better, and correct, explanation than o1 did.

[@Shahpo]

@ yea. PhD in maths, technically, but on theoretical physics (string theory / QFT).

[@jorge666]

@@Shahpocool field of study! went into georgia tech as a physics major but decided that for me a role in digital design was the move

[@axeedo]

0:12 I'm in the 3rd camp. I think AI is going to be good enough to "take r jobs", but not good enough to propel humanity into a technological explosion.

[@Ed-yr9yw]

I actually think the vast majority of us are here. We aren’t even really in the area where these “AI agents” work autonomously enough to trust them to do “jobs” and not just do specific tasks. They are trying to label us as either pro-AI or Anti-AI most of us are somewhere inbetween

[@OongaBoonga234]

That’s against company interest hehe

[@chrishayes5755]

technological explosion will happen in terms of understanding of physics and energy production, better motor tech (power to weight), robotics doing all your busy work for you. there's no stopping the tech progression outside of WW3 and even that will just push it forward more (but hijack it onto the wrong path). amazing things are coming down the pipeline. don't worry about humanity, worry about what's in your direct control: YOU. acquire power, influence and resources so you can enjoy a good life. otherwise you'll be hooked into VR 24/7 on a dopeamine drip kept alive by government welfare programs lol.

[@ivy8483]

@@chrishayes5755you are my Morpheus

[@tropistan7735]

​@@chrishayes5755love this fantasy world were we will all be unemployed,. unable to do basic maths or spell, but the robots we can't sfford will be working for us....

[@MehnixIsThatGuy]

I'm optimistic enough to think AGI is possible and pessimistic enough to know it'll just be used to do stuff like invent ways to directly stream ads into our minds.

[@Vangaurd_tiger]

Directly into our dreams.

[@BassLiberators]

I might be crazy but I think if AGI becomes a reality the world will change so much that advertising won't be a thing anymore. Companies might not be a thing anymore.

[@Vangaurd_tiger]

@@BassLiberators agree, there will only be one company, the company that wields agi.

[@BassLiberators]

@ But what would that company even sell? Like, once we have AGI almost no one will have to work anymore. Currency might not even be a thing at that point.

[@cieranholmes]

@@BassLiberatorsbrother what are you smoking

[@v000000000000v]

this model actually gave good reasoning after asking it a really tricky thermal dynamics question that 60% undergrads fail to answer, it's one of the first model that I've come across to answer it properly

[@AbdulRehman-fn4nk]

@@RyanTaylor-pi8gq Like chatgpt is censored from talking about Palestine/Israel issue truthfully

[@RyanTaylor-pi8gq]

@@AbdulRehman-fn4nk I dunno bro, I've had pretty deep conversations about that very subject and it seemed pretty fair. Maybe it just doesn't line up with YOUR biases. Or maybe the training doesn't include the vast majority of the recent events?

[@andyyu5973]

What's the thermodynamics question? Just curious

[@v000000000000v]

@@RyanTaylor-pi8gq funny thing is sometimes it will say it was restricted by OpenAI programmers to avoid those topics lol...

[@v000000000000v]

​@@andyyu5973 Question was: Imagine you are in an office with a shared kettle, consider the following scenarios: 1. only fill up what you need and boil for yourself, when you are done leave it empty 2. keep the kettle partially full by first refilling the kettle to its max, then reboil, and you pour what you need, you don't refill nor reboil after 3. keep the kettle always full for the next person, first you reboil, pour what you need, and refill for the next person, don't reboil after 4. keep the kettle always full for the next person, first you reboil, pour what you need, and refill for the next person, reboil after Assume everyone uses the same amount of water, and that the hot water loses heat to ambient slowly. Use thermal dynamics and reasoning to determine the most efficient way that minimizes boil time for each person. Also if the solution depends on the relationship between usages, and thermal loss rate, elaborate. -------------- The tricky part for this question is that no numbers are given which makes it difficult for some people and especially so for chat bots, coz some people just assume that if you leave more hot water boiled it will be at a higher temperature for the next person, where the real solution depends on the relationship between the ambient heat loss and time elapsed between usages, it goes something like 4 > 1 > 3 > 2 but in some situations 1 > 4 > 3 > 2, being able to explain when exactly 1 beats 4 or vice versa and why 1 is always faster than 3 which is always faster than 2 is not simple. This bot even gave me a situation where 2 could be faster than 1 and 3 which I've never expected lol

[@2012mpe]

This time Chinese devlopers won the reputation ❤

[@zhoma982]

Even 38% of AI developers in America are Chinese

[@Ryosuke-12]

I found both similar....and both are free for me, both are dumb tbh

[@curtisalex456]

@2012mpe 🤣🤣🤣🤣, ,you are really funny. just ask this AI some questions about Fulong Gong or what happened in 1989 at Tiananmen square, or who is winnie the pooh?

[@jackeydu108]

@@curtisalex456 Dude, take a break, the political brain has poisoned your rigid thinking, with discrimination and prejudice to treat an open source platform is meaningless, DEEPSEEK open source will empower individuals and small businesses and small countries to effectively protect their data from the US syndicate theft, and even surveillance ~ this is the most important thing!

[@Shalin0z]

@@curtisalex456 of course, you people will always ruin the mood.

[@musicloungepodcast]

Impressive! If DeepSeek R1 is truly outperforming OpenAI o1 on key benchmarks and is open source, this could be a game-changer for the AI community. Excited to see how it develops and what new possibilities it unlocks!

[@Music_For_Life_YT]

Absolutely! Open-source models like DeepSeek R1 can foster more innovation and collaboration across the AI ecosystem. If it holds up to its promise, it could push the boundaries of what's possible, especially in accessibility and transparency. Definitely one to keep an eye on

[@Fliki_JK]

@@Music_For_Life_YT Great innovation

[@bulldozerfx621]

thanks chatgpt

[@omarismail1]

linkedin ahh response

[@chrishayes5755]

@@omarismail1 "linkedin ahh response" you speak like the type of fella who works at mcdonalds, no need for linkedin lol

[@TheJackTheLion]

"Rack Overflow" holy snap that just about killed me. The script for this one deserves an award Jeff.

[@PeterKoperdan]

Written by o3 (tunned).

[@knockonwall]

good joke

[@AR-iu7tf]

Yes. And artificial binary code

[@TheOriginalJohnDoe]

1:53 ofcourse a Chinese model beats other models in math 😂

[@n.park1]

Okay, that's racist

[@MisheelTubshinbaatar-mr3rj]

@@n.park1 its ok Trump is elected.

[@n.park1]

@@MisheelTubshinbaatar-mr3rj why did this turn political

[@Wave1dave]

​@@n.park1either racist or political. pick one

[@n.park1]

@Wave1dave just don't be either

[@DK-ox7ze]

01:10 Did anyone notice that the guy behind Zuck is also checking out the "Rack overflow"

[@borndrago]

LMAO

[@JiggyJones0]

They all are. Bezos is pretending not to look, but he can still see them from his peripheral 😂

[@Hobson474]

That’s meme worthy!

[@InfernalPasquale]

@@JiggyJones0I’m pretty sure Bezos is checking out Zuck’s puck…

[@davidt8615]

Its like when you want to check out what this other person is checking out.

[@Shiibai]

The scary part? Once you read Lost Scrolls of Jewish Wealth, you realize how much of your life you’ve been wasting. But the good part? It’s not too late to change.

[@stephanndon]

where we can find that book?

[@canadiannomad_once_again]

I tried the 8B model with the strawberry problem.. It was funny how confidently it counted the letters and still ended up with "2" as it "sounded right". Though it worked when I told it to trust itself.

[@cihadturhan]

Did you use DeepThink? I used the web app and it did pretty good job with DeepThink enabled

[@spartaleonidas540]

8b?? Buy a good gpu man 70b at least

[@liveen]

@@spartaleonidas540 how many good gpus you got for your 70b model? 😂

[@li_tsz_fung]

@@spartaleonidas540 unless you're actually making money with open source model, there's no good reason to buy multiple flagship gpus just to try out a new model once a while.

[@TheAzraf123]

buy the nvidia supercomputer small one it's good

[@JomeG]

literally one of my colleagues in the field told me about it like 2 weeks ago that it's the next big hit and I took his words with a grain of salt and started using it. 2 weeks later and I'm shocked at how consistent it is with its information

[@BBWahoo]

You owe him backshots

[@samserious1337]

Now you can give china information for free :>

[@GoalOrientedLifting]

give it 10 years and china will be ahead in most areas. 5 years ago they produced more engineers, each year, then the total amount of engineers in USA.

[@seeibe]

@@samserious1337 Rather China than the US three letter agencies tbh

[@QTwoSix]

​@@GoalOrientedLiftingChina can only steal from the west, they can't create.

[@Combatfootage01]

Its pretty clear what their response to this will be. They'll cry "Chinese Spyware" instead of improving their obscene fees. Edit: I know that it can't be used to spy, but the average person is a lot more gullible/stupid.

[@davidlazerz8564]

I mean....its happened often enough that you gotta be pretty dumb to just assume its not spyware until we get some sanitized open source branches lol

[@user-tr1zj]

this one is open source though so they'll need to build an insane firewall to block out the US internet from the rest of the world with domestic surveillance/censorship capabilities, which can still be bypassed with a VPNNN

[@avenger1011000]

The oligarchy must he maintained

[@Johnnyvtg]

Chinese spyware is so much worse than American spyware (/s because a lot of neurodivergency among tech bros)

[@luuopo3539]

you guys have the American dream you don't need nothing else 😎

[@st.altair4936]

And this is _actually_ open source unlike "Open"AI. Great to see!

[@Cjaymoney]

1:10 the greatest sentence created so far in 2025

[@921Ether]

bezos just can't catch a break with these guys 😂

[@costynvd]

I had to watch it several times. Just genius

[@BigSources]

Talked to it for 5 minutes about sports science. Felt an entire world smarter than openais free and limited model. Thank you china. 4o feels like it heard about a topic by listening to a youtube video that discusses a scientific paper. R1 feels like it wrote the scientific paper. Idk if it's on par with o1 because i don't pay, but it's definitly the superior free model.

[@jaaguar13]

After few tests I can say it is better than o1. I'm paying for o1 and using it in my daily work.

[@williamking3974]

不要使用中国专制得大模型，deepseek是中国开发的邪恶大模型，它会摧毁民主与自由

[@brewingtonnadine]

@@jaaguar13 agree

[@seekerodyssey1]

you are not talking about the chat bot on deep seek R1 official page right? U test time compute it?

[@hammertime89]

“Rack overflow” Bravo 👏

[@Mo4Honesty]

Artificial binary code 🤣

[@thekwoka4707]

an AI benchmarking company getting funding from AI makers isn't necessarily a problem, just probably a problem. Like, the companies selling a thing will want other companies to testify to whether they do the thing that they are supposed to do. Like Companies paying accounting firms to audit their finances. but that stuff should be public, and clearly called out whenever reporting benchmarks, and they should be doing what they can to clearly show how they are not favoring those that are funding them with test and result selections

[@Quisspo]

ive been using R1 extensively for coding tasks since yesterday and honestly, its capabilities are so helpful in cutting time when trying to think about stuff. Obviously it doesn't get everything right the exact way I want it, but it takes out the need to write most of the actual code by hand, which is amazing (especially casue its free and so far I haven't been rate limited after over 100 requests)

[@knwrtv2544]

R1 means , url ?

[@MrSupasonik]

That's interesting! Could you let us know which model size you're using?

[@Quisspo]

@@MrSupasonik Since its their official website ui I suppose its the full model

[@HermanWillems]

How did you add it to Cursor then?

[@donwang9117]

It's unlimited. If you use their API,the price is 2.19 USD/million token. Very cheap

[@gentoffire]

Here we in China are fighting so hard against mega corporations, hope more people in the world can join us.

[@shawnmen9805]

我可去你的吧，要不先把劳动法给落实了？也就骗骗外国人了

[@elibarbq]

Unfortunately they're too busy watching TikTok clips

[@The_Savage_Wombat]

Here in USA we are fighting so hard for freedom from autocratic leaders, hope more people in the world can join us.

[@SuperSviki]

​@@The_Savage_WombatAre you? Cause I couldn've sworn you just reelected a billionaire who tried to overturn the election the last time he lost...

[@jelly.212]

​@@The_Savage_Wombat stop being anti semitic 😡

[@cybertopgun]

I'm pessimist to the point of thinking we are fucked either way

[@houssamassila6274]

makes you sound supersmart

[@thegoldentree6913]

200 IQ

[@MontyMcRib]

damn dude where did you get your pessimism degree you sound very smart

[@DR_1_1]

I'm still struggling to find out how it could be useful for me. I have been using it to ID bird songs, but outside of Europe and North America, it was almost useless. Also very powerful to identify photos of plants and insects, but in many cases you still need an expert to verify the answers, or do the work yourself to check the results. When I unload all my personal, day to day issues, it's totally useless, only spouting generalities, it's really depressing...

[@yalnisinfo]

@@DR_1_1if it is not fed the data it can't really be sure about it right? Like how much are these datasets includes plant and insect photograph. I wouldn't assume much. If you are going to use it for a specific purpose. Finetune a model for it with and appropriate dataset

[@AMANKUMAR-oh1zt]

The jokes ate just so much. I doubt if AI can ever compete with Fire ship in passive aggressive jokes. 😂😂😂

[@ahchoo5659]

Id rate Elon’s salute nein out of ten

[@MrRaja]

Nein out of Elfleven 😅

[@cuckmasterflex9106]

I did nazi that coming harharhar

[@savejeff15]

Nein out of na.tzi

[@starmorpheus]

A heartgoesout/10

[@awesomesauce804]

The reichest man in the world.

[@abhijoy.sarkar]

America and China, if they were collaborating instead of competing, we would have time machines by now.

[@grerovambrozoyuz9426]

It's better if they compete. Why because they do their best and fast lightning ability who dominates

[@axitc]

competition brings about innovation faster

[@JSigmoid]

The Deepseek model sucks ass. Have you used it?

[@TANISHQTHUSE]

@@JSigmoid i mean I found it exactly same to use as ChatGpt but much faster than OpenAI Still feel claude is better for coding still considering upon initiall hypothesis

[@AMan-xy3lx]

Yeah, and America would be trying to take over the world with it... this is a better way.

[@mmaxh]

Optimists make money ❌ Companies make money from optimists ✅

[@blackpiller3777]

Optimists lose a lot of money too

[@nikhil_jadhav]

Optimist gets rug pulled.

[@ecoista1373]

​@@blackpiller3777 the small ones does, the bigger ones fail upwards

[@indrajitg]

Unless it is the optimists who start a company, talk using lots of tech-lingo and then reap money from the investors!

[@Vinayakkkd]

Optimists make money for companies (as employees working for them, as consumers by buying)

[@GhostOnTheHalfShell]

The beauty of this is the producer, the developer of this model, which happens to be a hedge fund. I totally would not expect somebody to have set up the appropriate options to write out the sell off and make tons of money off of it.

[@mjob321]

I'd argue the base technology has become stagnant. The reason why it seems we're moving forward is because we're learning to find utility with these tools over time.

[@pensivepenguin3000]

Exactly, and even Fireship doesn’t touch on that enough. These aren’t advances in the underlying tech but rather clever ways of getting them to take more independent steps before returning a response. Basically just letting the bot feed its own answers back into itself as prompts until it decides which one is right. Even then, it’s not deciding correctness based on logic or reasoning; it’s deciding it based on how close it is to other examples it’s seen in it’s training data. I still believe we’re going to need a fundamentally new AI paradigm to achieve AGI, super intelligence or any other buzz words currently being thrown around. Language models aren’t gonna get us there

[@gavinderulo12]

​@@pensivepenguin3000reasonable take. I think the sheer data Transformers require is a big issue. Humans can learn with much fewer examples. Also the fact that current models encode both language rules like syntax, tone, emotion etc as well as pure knowledge, like how a microwave works, into the same neurons seems very inefficient.

[@Dextrostat]

@@gavinderulo12 Yeah I think we'd really need to get an understanding of the brain first before we can even get into AGI. Our brains are more segmented and is like a collection of classes with different functions, Frontal, Temporal, Parietal, Occipital. For example, the pre-Frontal Cortex handles Attention, Motor functions like Impulse Control, and Working Memory. We can learn things in like 2-3 attempts vs the hundreds or thousands that LLM's need. Way more efficient.

[@pensivepenguin3000]

@@gavinderulo12 exactly. I really think the future is various specialized AI models working together, much like the different regions of our brain, in a coherent and seamless way that seems like it’s one thing. I guess you could say we are starting to get there - LLMs process language, then we have models for computer vision, text to speech, speech to text, image generation and so on. ChatGPT can write and execute its own python in the background, which does allow it to solve complex problems more reliably, so as the compute and resources keep scaling up, maybe that kind of thing will become less costly and more integrated, essentially giving them something akin to reasoning

[@StarStrider99]

@@pensivepenguin3000 You seem pretty smart. Are you an optimist or a pessimist? ;)

[@houssamassila6274]

yet you didn't answer the question: Does it pass the vibe check?! You just explained how it works.

[@Kass_me]

That's just how this guy's videos work. Just saying nothing new for 10 minutes

[@joaopacheco2240]

@@Kass_me 4 minutes + sponsor*

[@savejeff15]

​@@Kass_mehow can you say that when he informs you about the Chinese model and how it roughly works? But yeah I would also like to know if it passes the vibe check

[@dylan6386]

@@savejeff15 DeepSeek currently does not offer users the ability to opt out of having their data utilized for AI training purposes. As outlined in their privacy policy and terms of service, the platform collects various types of user information, which includes: - Chat inputs and outputs - Profile information - Device and network data - Usage patterns - Some data may be retained after account deletion for legal compliance - Data is stored on servers located in China - Personal information may be transferred outside users' home countries This data collection is an big part of their operations.

[@petrektek1385]

It works yeah. Pretty well.

[@NTKM-om9vn]

Google's Gemini and ChatGPT is garbage. DeepSeek R1 is now my default AI companion!

[@Davidman3976]

Thanks bot.

[@yalahhabibi-r7n]

bot

[@djkidii]

@@yalahhabibi-r7n you are ?

[@Skux720]

First Black Myth Wukong, then Marvel Rivals, and now Deepseek, these Chinese developers are just built different

[@MegaAntimason]

For coding deepseek outshines chat gpt by a mile. Chat gpt gives toy examples,buggy code and not fully ready to ship. Even with the same prompt

[@jaaguar13]

Just tried it out and can second it. ChatGPT spat out all kinds of nonsense when it came to difficult system setups (DevOps tasks and such). Deepseek offered much more sensible advice.

[@Nice-sm5hr]

Wow. Imagine paying $200 a month for something that’s a mere joke, a frivolity, to you

[@JennyPatel-im8xi]

1:33 honestly speaking I thought of this vulnerability when it came out but never tested it thinking if I know those senior dev would definitely know.

[@Geloarc-ni4ze]

I am still a pessimist. I tried out deepseek just now and either tricked it into saying it used supervised fine-tuning when it was not (based on this model), or got it to admit to being fine-tuned. So either this video is incorrect (and deepseek was fine-tuned) or deepseek still hallucinates.

[@ObviouslyASMR]

Their paper describes two versions, R1-zero which is finetuned with pure RL, and R1 which has some supervised finetuning mixed in (this is the model that performed best). But also you can never trust a model's statements about itself because generally they're not trained on that information, as the information didn't exist yet at the time of gathering training data (plus it seems this model has been trained on chatgpt outputs so it can happen that it literally tells you it is chatgpt)

[@mig_kite]

More open source development please, we can't afford to let this sit in the hands of the few.

[@lokish]

Coding? Math? Fuck that, this model is a beast in role-play.

[@mrhowll4333]

"Closed AI companies like OpenAI." LOL

[@catcatcatcatcatcatcatcatcatca]

I hate the chat-models where I need to write multiple descriptive questions and as return get overly verbose answers, just so the model is tricked into doing some actual thinking. I do not have the social skills necessary for keeping up a fruitful conversation. I end up googling the documentation myself, so that I can ask better questions. Model that benefits from short prompts sounds wonderful. The less it feels like a conversation, the less I hate using it.

[@silkretega8902]

real

[@iuruoy-shao]

this is real, smthn I noticed is that in natural conversations when the other person doesn't really know what you are talking about, they ask clarifying questions so their response answers you precisely, whereas AI seems to try to capture all possibilities of what you COULD want to know

[@cihadturhan]

I agree and sometimes I feel reluctant to write a description to direct it to the right context, then use google search. The answer has to be long so ai needs to make a background context for the actual answer. Interestingly it's claimed r1 solves it somehow, I will try...

[@JonWtalk]

Math bro, there no thinking by anything.

[@danielbinoy]

0:32 might be the realest thing I've ever heard - a broke, stupid optimist

[@chiefpacman]

It's not a deep comment. It's actually a pretty dumb concept. Optimist don't make money, they just lose other people's money in the hype. Nothing special about being a lying hype man

[@anubis6084]

So lets use the 671b model and let it solve the collatz conjecture!

[@therealmagicmonti]

or the Riemann hypothesis

[@stalinsbutterfly]

​@@milan0023 not enough.

[@UriGerhard]

at least 3, maybe more

[@Hailzua]

The ai thoughts are keep going

[@jxyang1181]

in 10 years later

[@KaiKai-q7x]

I think people are misunderstanding whats going on. Its not that this Al model from china is crazy. Thats what the media is saying. The chinese model is great but not well rounded. In very specific scenarios is is able to meet the same benchmarks but its not as good. Heres an example because social media seems tech illiterate. Lets say one gym bro has been working out for 12 years, hits crazy PRs, a new guy joins the gym 3 years ago and matches the other guys bench PR. Thats incredible, ask him what his one rep max for shoulder press is though, or literally any other exercise.The chinese company behind deep seek did not train their model like the US models, they trained for specific scenarios primarily. Think about it like graphics cards vs asics if youre a crypto bro. Asics can outperform GPUs at specific tasks, just like a gym bro who only does bench will outperform someone who does every exercise. Its honestly not a shock that china has done this, honestly a shock no one had done it sooner

[@AlwaysThirstys]

The one shock is the cost.

[@divinecomedian2]

2:52 these stock clips are getting out of hand

[@Noobificado]

God damn, China has been stacking some W's lately

[@bilbobagend8155]

if slave labor is a w, then china is the biggest winner in the world

[@vincentdesun]

The rest of the world is lazy and rot.

[@melonmind.eng333]

China is one of the world's oldest civilization. They've existed since even the Egypt and Mesopotamia times, which are now disintegrated or absorbed into other cultures. I'd say that their existence is a testament of cultural, political, and social resilience. I think knowing this, we can wonder why the west sees them as a threat. I mean they make all of our products in our homes, while we attempt to sow hatred to them instead of being more competitive and innovative

[@bilbobagend8155]

@melonmind.eng333 Would you call Italians Romans? Chinese culture was uprooted in the 20th century.

[@xiaofang902]

​@@bilbobagend8155 people speak and write Chinese and fighting not to be a puppet of US. This is the culture of Chinese.

[@GeoffBelanger]

I asked “How can I download and run you locally?” and got this response, “Unfortunately, you cannot download or run my exact version (GPT-4) locally, as I am a cloud-based AI model hosted on OpenAI’s servers. However, you can use alternatives to run local AI models with similar capabilities. Here's how…”. Can someone explain what happened there?

[@zxtsteve]

Maybe you are using V3 version which is distillation model of ChatGPT, if you want to use the reasoning model DeepSeek R1 you need to turn on “deep thinking”

[@telescopicS627]

DeepSeek can't create illustrations, and I have a sneaky suspicion this is why it's less resource intensive. I think it's hype.

[@Enigma-t6n]

Put down the pipe & Tinfoil aside, Deepseek ( Engine : DeepThink R1 ) is trained specifically on STEM fields & on Coding accuracy & efficiency. While as of now chatgpt is better at visualization & somewhat better in copilot integration, but actual works of coding & data clean, analysis., Deepseek is by far the fastest & it's open source. So For redditors brain, it can look like hype, but people with actual jobs & come out of house regularly it looks promising as we are already using it.

[@DronyDosko]

It helps academically way better.

[@jpaburido]

Just keep hugging on chatGPT

[@mudkings]

Lot's of hype

[@telescopicS627]

@ Lots of chicom bots hyping it too, bad grammar.

[@SharatS]

Lmao that meme template with Zuck staring 😂

[@tamelo]

Why did it take the industry 6 days to the market to crash?

[@eskanderatrakchi4914]

same question here 🤣

[@thatsalot3577]

Would love to get a 100 seconds of chain of thought models, or a longer video with every kind of generative AI model

[@AfifFarhati]

If you want to learn about how AI models function including chain of thought with the same style of memes , then the channel bycloud is the best out there , i recommend you check it.

[@constantin58]

I hope that AI will eventually be able to uncover and expose the details of corruption among officials worldwide, revealing information that is currently concealed from the public.

[@voidmind]

It's cool that they released a 14B (9GB) model that can be used with video cards that have 12GB of VRAM or more. Compared to llama 3.1, where they gave you 8B or 70B, but nothing in between.

[@leszekpomian]

Your sponsored segments are starting to get so good that I look forward to your joint collab with LTT

[@peter_de_Jong817]

Again capitalism got beaten in the marked place of ideas. Open source W.

[@bilbobagend8155]

open source isn't socialist

[@danielxbox28]

@@bilbobagend8155 it's it free and everyone gets equal share. it's more socialist then capitalist...

[@genxploits]

@@danielxbox28 no not everyone gets a equal share under socialism not open source, you still need to actually be able to understand how to use open source, its capitalism

[@danielxbox28]

@ hugh that doesn't make sense clear it up? capitalism is socialism when you make losses anyway

[@genxploits]

@ capitalism is not socialism and you cant use the open source unless you know how to code, which is captialism, in socialism everyone would be trying to use the open source

[@JaskaranSingh-vv1zo]

This is my default ai tool now. It's deep think work like a charm. I asked it some situational and emotional questions and it did great answering them 🤩😍

[@maciejnowakowski7087]

200$ a month? Im in france a pay like 60-70 euros a month for a 2 pers business account where i have access to o1. Is this pricing different in the us?

[@TsugMt]

200 dollars here in Mexico too

[@DronyDosko]

Even in Sri lanka

[@12kenbutsuri]

I asked O1 to make a plot of a spherical harmonics in xyz coordinate, with z=1.0. it was such a basic task, but it couldn't even come close after dozens of prompts later, and i gave up and had to do all myself.

[@lavendereucalyptus3225]

It’s worth mentioning the distinctive difference between US and China the way they view and apply technology. Technology to US is to privatize and monopolize market for dominance and exploit the mass mankind. Technology to China, the ultimate goal is to benefit mankind. That’s why China has state owned internet infrastructure, like other state owned infrastructures, keep cost down and good connection services to mass population. This idea of using technology to better mankind was mentioned in Das Kapital by Marx.

[@M-Mir]

Social Credit Score +10000

[@lavendereucalyptus3225]

@ Social credit score = US propaganda

[@sunday_2000]

@@M-Mir 我是中国人，经常看到你们西方人针对中国说什么信用评分，这个社会信用评分到底是个啥东西？

[@M-Mir]

@ Its a meme based on a pilot project of the city of rongcheng.

[@ohwhatworld5851]

DeepSeek's ability to execute prompts at a fraction of the cost and energy usage is such a groundbreaking technological advancement for AI. The energy usage was going to become a bottleneck, which is why tech companies want to build nuclear reactors to help power the data centers. This field can now advance without limitation, aside from regulations. It will bring cheap AI powered technology for everyone.

[@dyo_x0]

is the energy part true? that was my biggest ethical concern other than ai replacing workers

[@ohwhatworld5851]

@@dyo_x0 Yes it was reported on various news platforms. Google it.

[@mukeshodhano4094]

MarkZuker Burg: "we value your privacy" MarkZucer Berg also: hmmm

[@torarinvik4920]

The "We have no mote" letter is just aging better and better.

[@sssamalander]

Are you ever going to do programing tutorials again or just tech news? I loved the Fireship channel because of the way of breaking down complicated topics and explaining concisely and really well - it's something that's incredibly unique, and that I haven't been able to find anywhere else. I understand that maybe news videos get move views, but it's disappointing to someone who's here for the programming content.

[@iuruoy-shao]

Not to mention the bro-culture tone to his videos nowadays, as a female viewer I feel slightly irked when things like Zuckerberg were mentioned for humor points only

[@victrixx]

@@iuruoy-shao Most stem content ends up with a ~>85% male viewership which makes it ideal to sneak in bro jokes to get new viewers interested in new posts.

[@CripplingDuality]

Easy money.

[@InfinityOver0]

@@iuruoy-shao The minority will never be adjusted for, as it is not the statistical majority.

[@PeterKoperdan]

Out of the thousands of programming teachers on YT, surely some are on par with Fireship?

[@chrisbryden8102]

I’m absolutely blown away how good Deepseek is. I’ve gone through my prompt history and gave it ones I wasn’t particularly happy with the responses and it just completely crushed the others. WTF!

[@ThatsPety]

0:29 im stealing this (with no citation)

[@mark_vanheyningen]

Same

[@PotionsMaster666]

😂

[@AdamUS123-j7q]

China is always making the same things Cheaper

[@CrimsonVine-v3]

Individuals who get rich by driving up prices through marketing hype dislike China.

[@ImagesOfCountries]

China is in a sense a world 'savior'... 👍

[@ÑEZ0688]

"same things"

[@AdamUS123-j7q]

@@ÑEZ0688 Yes "Same Things"

[@WanChin-ob6oj]

copying the same thing cheaper

[@FrozenFlames243]

0:47 "don't steal my proomts bro"

[@Charlie-xx2wj]

This situation reminds me of the fierce competition between Internet Explorer and Netscape decades ago. At one point, Microsoft decided to offer their Windows operating system for free to the public, which ultimately led to the demise of Netscape.

[@MissaMelodica]

0:13 woah, priests are some of the most optimistic people left in our society. Not sure that stock footage fit so well

[@ballsmasher3000]

But this is a funny stereotype of showing very conservative peoples( conservative by definition means trying stick with existing or past)

[@MissaMelodica]

@@ballsmasher3000 Fair point about the stereotype, but I think priests represent more than just tradition. They point to the hope found in Christ’s death and resurrection- something that’s beyond anything AI can offer. As much as I love AI and the amazing ways it drives our creativity, it can’t offer eternal life. That’s the kind of perspective priests bring, and it’s something worth appreciating. Pax et bonum

[@piergiorgio919]

Optimistic?? Most of them blabber about the end of the world and society

[@MissaMelodica]

@@piergiorgio919 Recognizing a broken society isn’t pessimism-it’s realism. The focus shouldn’t be on ignoring the problems but on finding redemption and hope beyond them. Even Pier Giorgio Frassati, whose name you share, lived this perspective. He saw the brokenness in the world but devoted his life to addressing it with faith, charity, and a focus on what truly restores.

[@piergiorgio919]

@@MissaMelodica yeah i agree the system is broken. The church doesnt pay taxes, paedophile priests arent prosecuted, cults ruin people's lives, mega pastors embezzle donations and who knows what else. But yeah keep supporting this great system and believing these lies just cause of "faith"

[@AVIMANHWA0]

bro i just downloaded this and asked about a phone that is about to launch in my country. and OMG this is my news source now. this is the only AI i am gonna use on a regular basis

[@majoramarix]

The only benchmark you can trust is counting the 'R's in strawberry and he counted it right!

[@knerduno5942]

No it did not. Look at the comments

[@majoramarix]

@@knerduno5942 it did for me, turn on the deep think if it didn't

[@akosijuan4034]

Can Deepseek solve some of the most prominent unsolved math problems including the Riemann Hypothesis, the Birch and Swinnerton-Dyer conjecture, the Hodge conjecture, the Navier-Stokes existence and smoothness problem, the P versus NP problem, the Yang-Mills existence and mass gap, and the Collatz conjecture ? You can put it in a test

[@yl_ohmeohmy]

This just goes to show you no matter how good you are at something, there’s a Chinese 12 year old who’s better than you.

[@alinmoraru]

Thank you for doing what you do. Big respect from Romania

[@DixonYamouf]

Open source software exposes the contradictions of western capitalism. The ai should be for the workers, not for our bosses.

[@dyo_x0]

i sort of agree, but you contradict yourself. if the ai is for the worker that is under a boss, then whats preventing the ai from replacing that worker for the boss?

[@aaronzhao1052]

The open-source community has just gained a new Titan.

[@newfrontiers5673]

DeepSeek R1 feels like a game-changer for AI innovation on multiple levels. Its affordability, speed of development, and open-source nature make it a strong signal of a shifting paradigm in technology-one where accessibility begins to rival raw capability. What stands out most to me is how it embodies both the potential and the tensions of our current AI landscape. On the one hand, it democratizes AI, empowering smaller institutions, underfunded markets, and even individuals to access tools that were previously locked behind paywalls and proprietary restrictions. This kind of openness fosters collaboration and rapid iteration, which benefits global innovation. On the other hand, it’s a powerful reminder of the geopolitical dynamics driving technological progress. The fact that U.S. sanctions inadvertently fueled this breakthrough is an ironic twist that highlights how competition can unintentionally accelerate advancements in unexpected ways. Ethically, the model's potential raises critical questions. How do we balance democratized access with the risks of misuse? How do we prevent biases from synthetic data or ensure transparency without overloading the system with constraints? It’s like standing at the edge of a double-edged sword-one side offering immense possibility, the other demanding vigilance and responsibility. What intrigues me most is the clear shift toward breaking monopolistic barriers. It’s a direct challenge to the walled gardens of AI giants, and it forces these players to rethink their approaches. DeepSeek R1 is not just a technical achievement; it’s a philosophical statement about the future of technology and how it should be shared. In a way, it’s like watching Prometheus bring fire to humanity. The question now is whether we’ll use this flame to build a brighter world-or burn ourselves in the process.

[@SatoshiNakamoto-b2x]

Why are you doing all your Terminal stuff in the VSCode built in terminal?

[@kynsj]

It's not VSCodes "built in terminal", it's VSCode embedding whatever terminal you configure it to use

[@loading....]

@@kynsj it's literally called vscode integrated terminal. i think you are confusing terminal and shell. you can configure the termianal to run different shells(powershell, cmd, bash etc)

[@kynsj]

@ You're right, I'm confusing them.

[@Sportgameww]

GPT-4o: Best overall versatility and strong across domains. DeepSeek V3: Excellent all-around performance, particularly in math, coding, and English. Claude-3.5: Niche strengths in SWE Verified, English QA, and coding assistance. Qwen2.5: Top for Chinese tasks, decent in English. Llama3.1: Competitive in English, but weaker overall.

[@crumbopulis]

Roses are red, violets are blue. If you think you’re good at something, there’s an Asian better than you.

[@robertstevensii4018]

We didn't listen :(

[@slaapri]

So this AI learns by the equivalent of giving yourself a bit of chocolate after an hour of study? They're like us already!

[@codesafariDev]

2:54 omg, I thought we had to wait for ai to get much better before we see stock footage like that xD

[@Jaydizler69]

“National socialist regime proves you don’t need mass migration of foreigners to innovate” color me shocked

[@deidresable]

America birth rate is declining because mot of them turn gayy so america really need immigration moe that they even realize

[@visiasch]

China lowkey got 3x US population. Of course it doesnt, use your brain a bit

[@Sergeant_Camacho]

China based

[@SagarBediOfficial]

02:52 The perfect example 🤣

[@Sportgameww]

GPT-4o: Best overall versatility and strong across domains. DeepSeek V3: Excellent all-around performance, particularly in math, coding, and English. Claude-3.5: Niche strengths in SWE Verified, English QA, and coding assistance. Qwen2.5: Top for Chinese tasks, decent in English. Llama3.1: Competitive in English, but weaker overall.

[@yuhuitong7717]

Modern China as a country, is actually not a communist one itself economically, even politically, but the way they behaved in the market - Fast copying every product and make a budget mode of every one, is in practice, a challenge to monoply and capitalism~~~

[@vambylamby83]

Communism 1 Capitalism 0

[@SushiDud]

US Ai stocks TANKING right now

[@curtisalex456]

😂😂😂😂, you are very funny.

[@SushiDud]

@curtisalex456 the funny person has not seen the market today.

[@qwertyqwertperson]

fading

[@kunalmehra5217]

"Closed AI companies like OpenAI." LMFAOOOOOOOO

[@CryptoMiningTurtle]

Senior “prompt” engineer 😂😂😂 I love it

[@qwertzmelon]

0:10 You got optimist and pessimist the wrong way around