The Open Source AI Revolution continues! HUGE News & Updates

[@MattVidPro]

HUGE Thanks for Riffusion for sponsoring today's video! Check out their FREE beta: www.riffusion.com/?

[@LouisGedo]

👋 hi 👋

[@MattVidPro]

@@LouisGedo Hey Louis!! Thank you for watching the channel! Your comments always brighten my day :)

[@LouisGedo]

​@MattVidPro 6:23 Also watching your sponsors 👍 Just like you, I think the competition in the AI space is great news for users......... looking forward to trying Rifussion

[@MrEthanhines]

22:27 did we just get rickrolled?

[@dockaiju]

So from your statements, if a person makes music using the apps you promote & doesn't mix them in a non-AI app that we shouldn't be able to copyright even if we spend days or longer on it? Seems hypocritical. I think you need to do some introspection. Why should us AI artists subscribe & respect you as an influencer after this?

[@PhiloSurfer]

If DeepSeek R1 can create such fuss and havoc, imagine what will happen when they release R2D2.

[@qwazy01]

! Beep Bop Beeeep Buzz Bip !!!!!!

[@RiftWarth]

DeepSeek is the real Open AI.

[@LDEV-l3p]

Man, I see folks barking up the wrong tree here. Everyone's arguing "Is DeepSeek better or is OpenAI the real deal?", treating DeepSeek like some rival. But honestly? They're missing the big picture. Here's the thing-DeepSeek's open-source! That means it's literally handing tools to its so-called "competitors" to level up, not the other way around. Seriously, it's feeding the competition instead of fighting them.

[@antonchigurh8541]

DeepSeek data exposed: kzbin.info/www/bejne/hXzViZltiLKXrdk

[@live--now]

R2...

[@Ovyron]

@@LDEV-l3p Because they're not competing, they're cooperating. Imagine the state of the world with a cooperating OpenAI.

[@nickgirdwood3082]

😂😂😂 Oh yeah, the one that allegedly has a better image generator than them but is nowhere close to DALL-E 3? FOR SURE!

[@LorettasHaven]

Riffusion is surprisingly fun to play with! I'm glad they sponsored you!

[@TeleviseGuy]

This is actually the best time for AI enthusiasts, because models are becoming really high-quality and accessible. This is the StableDiffusion moment for AI chatbots - open source text bots are becoming as good as proprietary ones, if not better.

[@justaguy7263]

Interesting stuff, keep going! You're such a reliable news source for AI

[@justaguy7263]

Entertaining too

[@MattVidPro]

Thank you!!! Doing my best out here!! (trying not to drown in the tidal wave of AI lol)

[@justaguy7263]

@@MattVidPro AI can be overwhelming yea😭 I'm testing DeepSeek R1 currently. Sure, o3-mini will be free too, but the limitations? I'm sure you won't be able to send a ton of messages as a free user, while R1 is COMPLETELY free! So maybe I'll still prefer R1, we'll see

[@NakedSageAstrology]

I must admit, he grew on me.

[@OllieNCS]

I think what’s really shortsighted is the perspective that most people have. What happened?: The market woke up and learned that openAI, like most of our American led AI companies overcharge and use compute expenses as a constant excuse (like pixel dojo, who won’t stop repeating it to everyone in his comments) and now we’ve learned what we knew all along: ‘It’s bullshit!’ So now we have to hit them back with the only power we have: our hard earned money and show them we are not going to pay $2,400 yearly for a model we can get for free somewhere else (including from the comfort of my phone and home). And their add-ons are a joke. Sam Altman is a smug pos that likes to talk down to everyday people like you and me, so hopefully deepseek, the market and America humbled his ass! 😂

[@LDEV-l3p]

People keep arguing DeepSeek vs. OpenAI, but miss the key point: DeepSeek’s open-source-it’s literally making its 'rivals' stronger.that’s a win for everyone.

[@ChristianIce]

Yeah, true, now that I can access a better model for free I'm going to give Sam Altman 300 bucks a month insted of 200.

[@bWWd0]

its forcing them to release better models, without R1 openai wouldnt be as fast with releases and they will be cause money

[@Marquis-Sade]

Lol​@@ChristianIce

[@codeXenigma]

Thank you for sharing, just spent 12 hours writing lyrics for a few songs on Riffusion. So much fun!

[@StudyBuddy-wq9uf]

What happened with tapping the brakes on AI development when the vanilla ChatGPT 4 was the frontier model. Lol.

[@markmuller7962]

The Nvidia/GPU problem is that Nvidia has the monopoly on training chips while the inference market is way more competitive

[@LookToWindward]

Yeah, that's what everyone who just keeps parroting "Jevon's Paradox" is missing. Plus, Nvidia was/is a genuine bubble, even if AI takes over the world in the next few years. When it goes up/down by something like 5% (I forget the exact amount), it adds/subtracts the GDP of Indiana from the market.

[@MattVidPro]

This is a great point to bring up, copy pasting my response from a similar comment: Nvidia has an incredible ability to pivot their offerings based upon consumer demand, and I'm not entirely convinced they would let inference compute dominance slip out of their grasp, but you are correct that we shouldn't assume it isn't possible. I also think (and my apologies this was not mentioned in the video) that while it is true DeepSeek didn't use as much compute to train R1, they also benefited from the Billions of dollars poured into AI by companies such as Open AI, Microsoft, Google, Meta, Nvidia, ect. Prior. - For example no one really knew if inference would be worth it prior to Open AI investing in what I'm sure has been endless trial and error to produce models like o1. I just don't see how Deepseek R1 is somehow proof that to advance AI not only in the US but also across the globe we don't need more GPUs.

[@markmuller7962]

@MattVidPro Ye sure I mean the investors clearly wantet some earnings and this was the perfect excuse. Still for the time being inference will dominate and new competitors may emerge to challenge the very fragile and very concentrated Nvidia unicorn. "At long last... No king rules for ever my son"

[@pavelkraynyukhov1990]

@@LookToWindward Not anymore, DeepSeek-R1 runs on AMD too. Even on Raspbery PI. What important is, that there is a tons of money being invested into software improvement of the GPUs others than Nvidia. Google TPUs are much cheeper for the inference as Nvidia too. There is also Groq LPUs.

[@PhiloSurfer]

@@LookToWindward Exactly. Even if Jevon's Paradox is operative here, it will mainly stimulate demand for those less expensive chips which are abundant and which China can produce. Nvidia's AI chips will definitely be adversely affected.

[@JM168M]

Thank you Matt 🍷🔥🍷

[@MattVidPro]

Appreciate the kind words!!

[@Vine_Zx]

Once they release O3 they will get hit with that R2, the stock market won't recover that time.

[@justaguy7263]

@@Vine_Zx most likely lmao

[@bengsynthmusic]

They know. Which is why they'll most likely introduce new features beyond just chat and image.

[@NakedSageAstrology]

UPDATE- Cursor now allows DeepSeek R1 as an option! Incredible results.

[@lightlegion_]

It’s fantastic to see your unique style.

[@TheGalacticIndian]

'Open'AI finally open-sourcing something🤯What a time to be alive!

[@monah62rus]

Excellent advertising, free and music is what you need, I'm already going to watch it. top top top top... .

[@rogerc7960]

riffusion is amazing

[@angrypenguinpng]

awesome video as always

[@MattVidPro]

Thank you!

[@jonbear7683]

The reason the market reacted is due to software that is proprietary to Nvidia which gives them a market advantage with their gpu's. The breakthrough (from a market perspective) is more about how less companies will be able to define an edge going forward, with competition being much greater due to innovation. In other words, Nvidia may not have the market on lock now, from an investment perspective, and that lowers the real and perceived value of their stock

[@LrdElderon]

Despite what deepseek showed, there was no reason for the stock markets to drop this far down unless it was way overvalued in the first place, in which case it had nothing to do with the tech sector and everything to do with companies and investors over inflating stuff Like tires becoming cheaper and longer lasting, so the entire auto industry dropped 30% of it's entire value

[@ChristianIce]

If you live by the hype, you die by the hype. That's the nature of the stock market. It makes sense for a company to lose billions when that was not the actual value of the company, but just the hype around it. Look at Tesla. It's too expensive, ugly and it's not autonomous as promised. Its value is a bubble. When another car will deliver for 1/10 of the price, don't ask yourself "why Tesla lost so mcuh", because it will lose a value it never really had.

[@BrianPellerin]

with progress like this, how long until every AI is free?

[@rogue_minima]

Nope. Once they become profitable (whatever that means) all powerful AIs are planned to become super expensive and subscription only. It's been the plan all along, since stable diffusion. And let's face it: who would settle for free but basic when we need super?

[@ChristianIce]

@@rogue_minima You know you can download an open source model and run it on your computer, right?

[@MilesBellas]

Do a broader, in depth, deep review of the 5090. 😊👍😁😎

[@pdjay8912]

항상 좋은 정보 감사드립니다😊

[@corazonjedi]

Im just here for the DeepSeek news.

[@patrickzupanc1795]

Great video, thank you!

[@lpanebr]

Riffusion is awesome!

[@ilyass-alami]

I like Chinese open source ai are free

[@elitefew8537]

u know nothing about AI OR its power

[@LDEV-l3p]

People keep arguing DeepSeek vs. OpenAI, but miss the key point: DeepSeek’s open-source-it’s literally making its 'rivals' stronger.that’s a win for everyone

[@OutlawMantis]

​@@elitefew8537Lmao dude, OpenAI fanboys are in super salt mode lately. 🧂

[@Ian-wu3ly]

Hey Matt you seem a little confused about why DeepSeek shook the market as bad as it did. The DeepSeek team proved that you don’t need to spend billions on H300s to train a SOTA model and teams are already working on replicating the efficiency gains documented in their paper. NVIDIA doesn’t dominate inference like they do for pre-training. Thinking models also shift compute from pre-training to inference. It’s a combination of these factors (less gpus for pretraining and the shift from pretraining to inference) that sunk NVDA.

[@ALFTHADRADDAD]

Efficiency gains were always expected

[@MattVidPro]

I really appreciate the response but I'm not quite sure I buy that. Nvidia has an incredible ability to pivot their offerings based upon consumer demand, and I'm not entirely convinced they would let inference compute dominance slip out of their grasp, but you are correct that we shouldn't assume it isn't possible. I also think (and my apologies this was not mentioned in the video) that while it is true DeepSeek didn't use as much compute to train R1, they also benefited from the Billions of dollars poured into AI by companies such as Open AI, Microsoft, Google, Meta, Nvidia, ect. Prior. - For example no one really knew if inference would be worth it prior to Open AI investing in what I'm sure has been endless trial and error to produce models like o1. I just don't see how Deepseek R1 is somehow proof that to advance AI not only in the US but also across the globe we don't need more GPUs.

[@metatron3942]

Still looking for an open source AI Music generator

[@justaguy7263]

@@metatron3942 meta has one

[@Metarig]

Nvidia isn't needed to run the useful model anymore, which is why its stock is dropping. You can run the full-fat DeepSeek R1 on multiple Apple Minis. That’s the reason-why act like it’s surprising? Even for training, the highest-end Nvidia GPUs aren’t required anymore, as DeepSeek has proven, and now AMD GPUs are viable too. This is the second reason Nvidia’s stock has dropped and will keep dropping.

[@claudioagmfilho]

🇧🇷🇧🇷🇧🇷🇧🇷👏🏻 Can't wait for an OpenAI powered agent that seamlessly edits my videos on Final Cut Pro and Apple Motion cutting, transitions, effects, and everything in between!

[@NakedSageAstrology]

You can already do it! I use movavi video editor, with deep seek R1 to automate the process entirely.

[@claudioagmfilho]

@NakedSageAstrology 🇧🇷🇧🇷🇧🇷🇧🇷👏🏻Amazing!

[@NakedSageAstrology]

@@claudioagmfilho It's an incredible time for open source, and now EVERYONE is a developer! Help us build the New World Order together, before a few billionaires create their version. 🙏

[@The_Collective_I]

0:02 - it finally happened, that's not him. That's generated! That's not HIM!

[@Zerobytexai]

Udio for some reason still sounds the best, and still the best vocals.

[@annieorben]

Hey Matt, I agree with your take on the DeepSeek R1 release being a mostly positive and good thing. The only issue I think is that it's spooking some people.

[@PhiloSurfer]

Wait till they release R2D2.

[@LouisGedo]

17:17 👋 There's actually MUCH less water on earth than most people realize. The earth's surface is actually a small portion of the entire earth. So while this thin layer of surface has about 70% or so water, in reality that's not much compared to the entire mass and volume of the earth.

[@MattVidPro]

I remember learning about this in school, compared to the earths total mass it's a miniscule amount. Interestingly enough if I remember correctly most of earths mass is still liquid, the molten layer under the crust

[@jl25735]

There's a point we're missing. Deepseek is also using Huawei Ascend Chips instead of Nvidia, and they don't rely on CUDA but in their own THX technology. If Nvidia's CUDA tech is challenged, its whole monopoly of AI hardware is challenged. CUDA is Nvidia's Moat and a chinese version could put them in trouble

[@sidonline1922]

WOW MAN!!!!!!!!!!!!!!!!!!!!

[@Bahamamos]

Need to see riffusion's TOS. It's the only reason why I only use Suno and not it's top competitor. But hopefully Riffusion doesn't force your songs publicly with a paywall or say they own your work smh. Companies need to simply take the subscription, and let us be free to create...

[@TheoGottwald]

The Deepseek Focus WIndow is much smaller then MISTRAL Large.

[@journeyofasha]

Starting a trend, CloseAI Openseek we should call it as it is....

[@knotcensored932]

For Pete's sake Matt, get R2X running locally on your 5090box and configure it as 7of9!

[@mrrfyW]

9:23 is when the sponsor ends, you forgot to timestamp that

[@MattVidPro]

my b the AI I use for timestamps got confused fixing it now

[@mrrfyW]

@@MattVidPro 👍

[@FRareDom]

Truly a tidal wave iykyk

[@ZerofulMaster]

Hi matt!

[@MattVidPro]

hey!!

[@AnonymousObject]

wtf

[@LookToWindward]

Did anyone else notice that when it cloned the Wii menu it put "Introducing Operator" in the News window??

[@MattVidPro]

omg I didnt even see that!

[@aliettienne2907]

8:27 Can it clone your voice to be use in its music creating model. If an artist want to make music they'll want to use their own voice, the average user wouldn't mind that option as well. It's crazy how one can make artistic music without having the actual skills to make music in today's time. The drawback to use Ai to make music as a professional is that fans wouldn't like the idea. Music Fans will find it to be unauthentic and lacking natural creativity or talent. 😎💯💪🏾👍🏾 21:52 The fact that the inanimate can be animated it signify technique can be done. And if that level of motion can be achieved, it signify the starting point for greater things which is to come. 25:42 Alibaba Gwen Video generation is working really smooth and effective. The details achieve in the video generation is very clear and coordinated..

[@A.M.-AI.artvideo]

🔥🔥🔥

[@cbnewham_ai]

I skipped because it was (c)rap music that started playing. 😄

[@MattVidPro]

lol no worries it's always your choice! That crap music was one of my generations by the way LMAO!

[@cbnewham_ai]

@@MattVidPro I'd have stayed if it was non-rap 80s stuff 😆

[@vi6ddarkking]

Yup. The fact of the matter is the at some point in the near future. Open source Local AI and closed source AI will be so close to perfect. And then. Well one is free and uncensored. The other is not. CAPITALISM BABY! The free market is awesome.

[@RoryOConnor]

zero-sum is just panic not understanding actual value distrubution, zero-cost will be the new hero #EducationalAi could revolutionize everything up to a higher standard/quality of life!

[@ChristianIce]

It's not the end for AI made in america or for the GPU selling. It's very much the end for OpenAI, and it doesn't have anything to do with technology. It has to do with all the money invested in OpenAI which promised unrealistic results. Now those unrealistic results are simply impossible. Why would anybody pay for the OpenAI API when it can use another, open source model? Those are the money promised back to the investors, and they are gone. "poof", bye bye.

[@HaggisWildlifeFoundation]

There is more going on than what people think

[@MilesBellas]

OpenAI has a lawsuit yet ? I like free stuff, I love open source but this situation is really another example of "Unrestricted Warfare". .

[@brexitgreens]

Corporate hypocrisy. OpenAI have scraped the whole web without asking anyone for permission. And that's a good thing! Except now they are trying to deny the same liberty to China.

[@brexitgreens]

_A response was here_ 🚩. _Deleted by KZbin._ 🤐

[@live--now]

o3 mini is good it almost as good as deepseek r1 XD

[@divinelyindifferent]

It’s absurd for Corporate America to expect loyalty from American consumers, especially amid all of these mass layoffs. If companies can “offshore” labor for higher profits, consumers can just as easily seek cheaper, *offshore AI solutions* (like DeepSeek) to save money during this tough economy. The loyalty is dead.

[@NickWindham]

The Nvidia problem is that much tighter restrictions will soon be placed on what they’re allowed to sell to China & other countries that aren’t military allies with the US. I could foresee Nvidia being restricted from selling any chips that aren’t already approved to those countries for the foreseeable future. Even those night get tighter quantity restrictions. This cost Nvidia hundreds of billions in long-term revenue growth, which analysts were forecasting in their calculations to arrive at Nvidia’s $3.3 trillion dollar peak market cap.

[@VioFax]

If anything this is an oppertunity for NVIDIA to make even more money. Selling training and inference chips to consumers. Better yet, lets make a generalist GPU and instead of calling it a Graphics Processing Unit. It will be a General Processing Unit.

[@ChristianIce]

But what if, in a year from now, you could buy a huawei video card for 300 bucks, with the power on a 5090?

[@JD-yz4kr]

Should you call it a crash or more of a correction? Nvidia was trading at 48x PE ratio, which indicates that it is overpriced and driven by speculation (maybe even maniulation). Going down to 38x PE ratio is actually not enough of a correction. Should go down to more realistic levels of about 25x PE ratio. As for less GPU, it is true that there will be higher GPU demand, but not for Nvidia's H100's, which is its moneymaker. Also, demand will be driven by non-Americans, especially in Global South, where Nvidia and other American manufacturers are not allowed to sell to. Even now, India is planning to setup their own AI research centres, as well as in Africa, Brazil, Southeast Asia, etc. And none of them will be Nvidia customers.

[@TheEconomicElder]

I don't understand why would take more GPUs? The R1 model is using an order of magnitude less compute for inference and training. A paradigm shift has occurred so that's why the markets are down.

[@MattVidPro]

We are increasing the efficiency of our AI models, which means we can create even better ones using more compute.

[@TheEconomicElder]

@@MattVidPro not sure about that TBH. Berkeley Uni just copied R1 for $30. Its over for expensive chips IMO, edge computing is the future. We will see though

[@ChristianIce]

@@MattVidPro Matt, you know that is also half true. Sure, we will be able to do more, but there's a roof to that. Example: I would love to generate more video output for my projects, but the computer power allows me to generate, let's say 1 hour a day. It will be cool to be able to generate 10 hours a day, but after that I would not have the time to watch it. What I'm saying i that the growt cannot be exponential forever, and at some point your needs are satified. You don't need a "faster car", you don't need "more lightbulbs in your house", see where I'm going?

[@MattVidPro]

@@ChristianIce I know what you are saying but I think you are underestimating the overall power potential of AI technology and how high that ceiling really is. just my opinion on it!

[@ChristianIce]

@@MattVidPro I totally agree with that, but if account for the projections of *needed* consumption vs the news about actual consumption, you have to expect a hit in the market. That's the same that happens with gas, prices change a lot, nobody counts the number of cars from a day to the next to establish the price, right? :)

[@MrRandomPlays_1987]

DeepSeek R1 would make Nvidia raise their GPU prices (due to loss) and would mess up Stargate project? then it's not good news for all exactly.

[@ChristianIce]

Yeah, sure, because Google can't wait to pay more for a product that needs less.

[@blubbblubbblubbish]

The Nvidia-stock-'crash' wasn't because they needed less gpus than OpenAI. It was because they (big investors/fonds) thought/think that Deepseek lies and actually needed much more gpus than they admit. That would mean that they have imported gpus illegally through other countries. And so the thinking was/is that much more of Nvidia's revenue comes from Chinese customers. This could be critical if the government would try harder to shut down these black markets.

[@MattVidPro]

This is a fantastic theory! I hadn't even considered this but honestly it does sounds very intriguing and believable.

[@bengsynthmusic]

No not really. All the major reports have been about cost efficiency and distillation. They highlighted the $6 million training cost, and that alone is the biggest reason for the stock crash. There is no evidence beyond rumors to substantiate the claim that they have several tens of thousands of H100s, so that wasn't widely reported.

[@pixelplaygroundofficial]

riffusion or suno?

[@lonedruid1]

good news

[@jonorgames6596]

You forgot to mention the elephant in the room... oh wait 24:38

[@yoyo-jc5qg]

it was only a matter of time for nvidia, u cant dominate and overprice your stuff for that long, ppl will find a way around it

[@pavelkraynyukhov1990]

Real reason NVIDIA goes down is 5080 is bad, 5090 is not a much of improvement, and Trump imposing import taxes on Taiwan's electronics and tightens sanctions towards China once more in regards to the GPU power. This will reflect itself in NVIDIA revenue this year.

[@MattVidPro]

I also have heard this and to me these feel like more probable reasons. I heard that the 5080 was disappointing for raw rasterization which sux. I have the 5090, and its obv a great GPU, but without experience with the 4090 I can't really say if it's a worthy increase especially with the price hike.

[@alancalvitti]

-1 chuck e cheese backdrop

[@acarria]

That's okay, DeepSeek will also be a pay only model within a year. They all seem to follow that same path. And some other new open source one will take them out as well. I'll hold the crashed stocks and laugh when it triples in value later xP. Competition is great for business since it encourages growth and better tech designs. I personally see it as a general knock off that will soon be passed very easily. Sadly has no use for me at this time, so idc either way tbh, but too many are crying over it and I'm just like.....have they not seen the AI field.....cause they get replaced constantly with newer models, and less than a year later it's a useless model comparing to the new one built from its backbone. It's not original, it's a piggyback that grew up to soon become the mount for another piggyback model lol. So enjoy it while it lasts, it'll go pay only in time.

[@ted1703]

it would be nice if sombody would create an avatar where we can download pictures, documents and own thoughts so he can live as me in internet

[@Zerobytexai]

It doesn't matter how good ChatGPT is or o3 mini, or any hype it gets, people are missing the point. It's the fact that it's not open source. They are in a losing battle regardless. Why are people oblivious to this? OpenAI loses by default

[@viewtouchpointofsale73]

I would like to point out to you that the pronunciation of the word 'the' is "thuh" in all cases when followed by a word beginning with a consonant. The pronunciation of the word 'the' is "thee" only when followed by a word beginning with a vowel.

[@MattVidPro]

You are totally correct! I do strive to improve my pronunciation but bad habits can be tough to break! I will try and keep getting better though!

[@Agispsi]

This isnt a hard rule in speech, its just common practice. Thee version can also be used for emphasis as a normal common example of how it is used. "He is THEE richest man in the world" for example. Use of the thee pronunciation even casually for all instances of the word the is still okay too. Weird... but linguistically fine.

[@viewtouchpointofsale73]

I appreciate your appreciation! By the way, you might enjoy learning what ViewTouch is.

[@Agispsi]

@@MattVidPro Bro really just tried to sell you a point of sale device...

[@christopherd.winnan8701]

Elephant video, true to life? How many elephants in your neighborhood currently? Not to mention all those floating lanterns falling in the wrong direction, about the set the Forbidden City on fire!!

[@onlyyoucanstopevil9024]

👍👍👍👍

[@claudioagmfilho]

🇧🇷🇧🇷🇧🇷🇧🇷👏🏻

[@WMR1776]

despite what the market might say, and what like 80% of the people you talk to think about things

[@Jack122official]

Yay

[@darkphanthom8741]

5:00

[@matthew04101]

-- Nvidia should send me their GPU or project DIGITS to guide my mind and believe in Jim Fan, otherwise my mind starts wander freely. -- As an A.I user I don't care about tech stocks up or down, I just want to see A.I getting better and rates getting cheaper. -- Don't give in to o3 because R2 is around the corner. LOL

[@puremustang732]

American Ai Companies just want money. DeepSeek is open source and super cheap.

[@robinmc123]

I can't stand the jump cuts it's to distracting

[@live--now]

open Ai is about to become Broke AI if they dont start releasing things better cheaper, and even free ...

[@aruak321]

What are your opinions about the heavy censorship built into the Deepseek model? What other gotchas may have been trained into this model? Also do you have an thoughts about the allegations that Deepseek actually just used the OpenAI api for distillation to save a ton on processing and cost for training their model? If true that would mean that Deepseek was not as "cheap" as everyone thinks but rather just an o1 distilled model.

[@MattVidPro]

Thank you for the questions! Not a fan of censorship in R1, or in general. The model is open source though which means people can create less restricted variations, and also folks like Pliny on X have been jailbreaking nearly every LLM and posting the methods openly, so I don't see it as an issue that is unsolvable, even if it is irritating. If DeepSeek used o1 training data on their model I wouldn't be surprised. The issue here is that Open AI doesn't have a great way to prevent this from happening, and the fact of the matter is also that o1 was trained on a massive amount of data that was scraped from the web, which makes their claims feel a little more hypocritical.

[@cbnewham_ai]

It doesn't appear to be censored any more than other non-local models - it's just a different set of censorship.

[@aruak321]

@@MattVidPro Yes I agree that OpenAI complaining is hypocritical and more than little bit ironic, but does that cast any doubts about how "cheaply" they were able to train the model if what they did was piggyback off of a lot the work that OpenAI already did?

[@aruak321]

@@cbnewham_ai I feel like the censorship is at a completely different level. Like the difference between asking about methods to create dangerous explosive devices (which western models will do) vs asking pretty benign questions that are only just remotely related to China. Like censoring queries about Winnie the Pooh even if China or Xi Jingping are not even referenced in the query.

[@mickelodiansurname9578]

I care not where my AI comes from only that its mine in isolation, I get to do with it what I like, it works, its cheap and its easily deployed... other than that if a guy with horns and a spiky tail produces the next Open Source model I'll use it. I'm also not entirely sure why I would care that a bunch of billionaire investors lost loads of money.... why would I care about that?

[@erikburzinski8248]

Horns and a scally tail are only cosmetic am sure ot will be fine

[@rogue_minima]

You entirely underestimate all that can be hidden into a LLM, specially a big one. They can be trained so they manipulate you into anything the creators want and you wouldn't even notice. They can be trained to embedded malicious code in your apps. They can be trained to hide dark enchantments on your magic spells. All of them could be doing this. We must care from where they come, not just how much they cost.

[@erikburzinski8248]

@ronilevarez901 you can train yourself to recognize manipulation and once you have done that it's almost impossible for someone to manipulate you in such a way.

[@rogue_minima]

@erikburzinski8248 almost, being the key word. No one will be immune to the smartest being on earth.

[@mickelodiansurname9578]

@ Yeah but I don't use LLM models to have a friggin chit chat... I use this one to write out json files for my app, produce python and javascript code, reformat text and form agents that source data... There are use cases above ChatGPT you know... and the vast amount of money spend on OpenAI is not the average user asking for stories or finding out if Albert Einstein was married... its devs using it to build on... we are the folks that fire half a million requests at the servers in an hour! Also at the moment you are convincing yourself that the DeepSeekR1 model is the story here... and you've also convinced yourself how things can be 'hidden' in the model to trick you... and you are telling an AI/ML engineer this with 45 years coding experience... So ummm yeah people really can convince themselves of all sort of things. But the real story is the process Deepseek outline in the paper, not the model. Using this even I can afford now to build my own model cheaply-as outlined in the deepseek paper... where I avoid the need for supervised learning (that alone is a 50% reduction) I avoid the need for RLHF... another 50% of that, I avoid the need for distillation from some other model (halfing it again) and I can lower the compute to next to nothing... Of course to do that, well you have to know how to train a model from scratch... but thats just a matter of math and learning. But using the Deepseek training process OpenAI's o1 model... by the end of 2026 kids in a first year CS degree course will be asked to replicate that as a term project!

[@DynamicLights]

First

[@MattVidPro]

6th

[@DynamicLights]

@MattVidPro damn you are right ✅️

[@steeleforce13]

Matt qwen been a downfall

[@MattVidPro]

they just released video gen for free!

[@steeleforce13]

Nah, it's not working properly ​@MattVidPro

[@MattVidPro]

@@steeleforce13 Strange.. perhaps it's region locked? A few users in my discord are already generating!

[@contentfreeGPT5-py6uv]

For me Is ok, i uding agents AND groq for videos AND llms am mix

[@SirHargreeves]

R1 is just a distilled OpenAi model. I doubt there’ll be an R2 etc.

[@MattVidPro]

heard this theory floating around, not sure if it's true or not, but deepseek isn't new to the party - they have been making models for awhile!

[@cbnewham_ai]

It's what OpenAI would like you to think. The jury is out. We don't really know a great deal at this point. Probably the truth lies somewhere in the middle.

[@dockaiju]

So from your statements, if a person makes music using the apps you promote & doesn't mix them in a non-AI app that we shouldn't be able to copyright even if we spend days or longer on it? Seems hypocritical. I think you need to do some introspection. Why should us AI artists subscribe & respect you as an influencer after this?

[@TheWorldAwake]

Am I alone in feeling sick of seeing coders everywhere ?

[@mediocreape]

the reason nvidia stock dropped is becauese r1 doesn't need cuda cores for infrence anymore. it could use any gpu, like amd , intel.