The fact that Altman spent the last 4 years trying justify OAI farming copyrighted content for training without a single cent paid back to the author, only to be left salty because his own IP was farmed and then open sourced is truly a masterpiece of irony
Lol. I usually dismiss the anti-ai fan boys copyrights arguments because most of them have no idea or base for what they are saying. I do not like them because i think that They are constantly undermining the human side of the argument with their brainless crowd attitude. But this time with altman, i do not care for bases, reality and facts. I care that it makes more than sense. It is karma!! Pure karma.
I think not crediting or asking for permission for other people's work without even compensating them and then passing it off as your own sucks. I think developers have gotten too comfortable even boasting about doing it. I love that R1 burst the OpenAI bubble for what it was, a grift, and at the same time I applaud them for making their work open source, even though they're part of the larger problem. Are these things enough, or do I need to write a todo app with an AI API to increase shareholder value for 50+ sprints for my disdain for these generative AI models to count?
The tech billionaires told all the software engineers that they need to compete with cheap foreign labor. Now the software engineers get to tell the tech billionaires they need to compete with cheap foreign models.
@@CancanTonia exactly. It was built by a group of smart people. They may be lying about the number of chips they have backing them (I don't know, maybe, maybe not...), The whole 'but, but their labor laws are SO bad that is how they did this' line is just plain stupid and shows no understanding of software development.
@ yeah LLama is open source. But this is still probably lighting a fire under their ass. Rumors are that the unreleased LLama4 isn't as good as the new R1 model
@@moonasha Poor Zuck called engineers "mid-level" without being one and claimed they're gonna get replaced THIS year with AI... He's the EXACT SAME as Jensen Huang from NVIDIA and other billionaires. So I'm glad AI loses this kind of spotline by democratizing it, now since everyone has AI the spotlight is back on us and what we do with it.
@@nicosoftnt I mean, you are factually wrong. Zuckerberg did do multiple years of a Comp. Sci. degree, then dropped out and actually practiced the discipline. Before SWE, that is what SWE was.
Cutting China off from high end chips forced them to figure out how to make the same tech work on low end chips. Force them to innovate while the US is just funding bigger server farms of high end chips.
@@OpaKopalke it probably wasn't completely truthful, as we know some higher gpu somehow made it there, maybe deepseek got their hands on them, but it still doesn't eliminate the fact that their resources were much more limited
US TECH COMPANIES: "AI is the future, we will replace all developers and replace all workers with Ai because AI can do all your work for cheaper". R1: "Ai replaces ai companies first before replacing any acrual developers because those companies cost too much"
@@nicosoftnt Next week or next month there will be another DeepSeek, Devin AI, SORA. Some headlines will go crazy about how amazing it is...and the cycle continues
@@danielbrown001 Undoubtedly, but this should at least reverse the trend a little bit until that happens. If all the major LLM providers figure out what deepseek did and apply it to their own models, should be some major energy savings.
@@danielbrown001 Yea, its insane the cope people have seeing the 1 trill sell off Doomers you know the smart ones meta, open ai, google will adapt and stay alive with all their hardware. Barrier to entry is open now. the next year will be insane efficiency is name of the game now. with competition from a whole host of new developers. because the training has come down in price so much. This is like pouring gas on the damn fire for entrepreneurs and the whole ai craze.
@@roger9822 At this rate, yeah. The chip ban just forces China to develop and innovate themselves. They just made a better AI model, why not a better GPU model?
@@MM-vs2et There's technical limitations on hardware. For example China depended on a Dutch company to sell them the machines that do the lithography.
yeah, imagine when openAI implement the 50 times faster inference compute of r1.into o1/o3, the log scale AI intelligence increase should increase raw IQ of openAI model at least a sigma or 2 beyond r1
Competition... Such a foreign word! "Open" anything is fine, as long as they own it and the "open" means "theirs". Having recently read about the execution of independent models (but long tinkered on smaller neural networks, tensor, etc) I have to wonder where people have been. I guess it's just mainstream enough that it's noticeable and has some more obvious backing... but it was inevitable in my opinion. That's the "problem" with real open - everyone seems to end up owning it. The more competition there is the more likely the innovation will be worth while.
@@jamess.2491 because everything is about class, literally everything is class warfare today, every issue you care about has to do with class warfare. But they don't want you to realize this so instead they artifically created a culture war we could fight instead, that's why they got so freaked out when people supported luigi mangione.
I'm still kinda scared what it could do on my device. It's literally the "never run scripts from the internet" and the only benefactor of the code's safety is China at this point. The code tends to be too complex for an average dev to understand and there's already raging an information war regarding chinese products where everyone seems to have a stake in one side or the other. I think I'm gonna stick with Llama for the foreseeable future.
@@tlk889 for transparency sake, if you could run it offline then the likelihood of it getting your data in the middle of nowhere is like an IT graduate on a crumb of puss puss
If you think what Altman said was bad, go look up what Larry Ellison said on AI recently. It's even more frightening and dystopian. And these people are working together now.
I just installed the 14B model of R1 via Ollama on my new workstation. I can see my system fans cranking up when I run a query, but it is super accurate and sexy.
@kessler-e6e I even managed to get the 70b model to work with my 4080 Super. It has to use system ram, but waiting 5 to 10 minutes is something I am willing to do in certain cases.
@@PhonkEcho should i ask u why cooperate bot is commenting nonsense. China hoing to take my data? Oh well us is doing this for decades. What difference it makes if i give it to a Chinese company vs a us based company?? But i am glad to use it on my workstation locally and glad thqt there is a way i can do that. Stop bootlicking cooperations
"tech bro" is fast becoming almost its antonym these days, being more and more often used (at least on the YT) as the *monied head* (CEO, etc) of an organization who *looks* like they're an over-confident "technical genius" with poor social skills... but is really just a cashed-up investor/figurehead with poor social skills. 😆 The "tech" part seems increasingly ironic. The actual contributions to the "tech" of those regularly called "tech bros" seems to be very firmly (and increasingly) in the past. Mind you, this seems to have been the backbone of the "whizkid ceo" phenomenon for a long, long time now. When people who look into such things inevitably find out what most of these ceos/billionaire founders *actually* contributed, and how long ago, they usually seem very surprised. It's a strange industry phenomenon when you think about it. Would we expect the head of a fast food empire to be they fastest burger-flipper on the planet? Probably not... Anyways, I took the comment in the positive way. I would also argue that someone just trying to feed their family and live comfortably -- a truly noble commitment to my mind -- isn't trying to co-op large parts of the open source work of thousands of people, create garden-walls around it, nor mine/collect/aggregate/arguably reproduce vast amounts of copyrighted material either.
Been playing with it for last few hours. First time I've been so impressed by an AI model. Its super impressive. Am asking some pretty obscure C++ / memory model issues and its giving back really nice results, and its happy to refactor it based on more prompting. Super excited about ability to run local.
I gave it code of mine to optimise and refactor just to test it out and I was mind-blown after it managed to completely understand and optimise + refactor my code on the very first try.
Joining on this. Gave deepseek a piece of code for complex chart building and said "it has performance issues, any ways to optimize it?" Aaaand it just made it 10x faster in like 2 minutes and explained issues. I love this. Initial code was generated by ChatGPT btw, which makes it even funnier.
The 32B model runs reasonably well on an M3 with 36GB of RAM. It's not super fast because of all the reasoning it does, but at least it doesn't start swapping into virtual memory, so it's still usable.
Not gonna lie, id rather Deepseek have my data than Open AI because if you support them then you encourage more companies to use the open source approach. if open sourcing their model flops then less and less companies would be considering open sourcing their models in the future. Every single one of Deepseek's models is open source and I think they should be rewarded for that somehow whether through donations/investments or to help train their models as long as the models will remain as they are, open source.
DeepSeek is a quantitative trading company. Its CEO Liang Wenfeng once said that it is not important whether we make money or not, but it is important that Nasdaq does not make money.
it won't be. OpenAI and friends will develop more advanced models, and keep them closed. Take a step back. We are literally at the beginning of this whole AI thing. We got a long ways to go. We'll probably see something similar to Moore's law
I'm sorry but what does this change? Meta has also released their weights open source. Nothing new or different about this except the lol crowd thinking it's funny that companies in the US are panicking.
@@igoralmeida9136 Not exactly, this was reactive to capitalism and US monopoly. Turns out monopoly is generally NOT good for anyone, who would have thought?
Running Deepseek R1 on the CPU is absolutely fine. Once you get into the big models, the memeory bandwidth bottleneck wildly outweighs the performance one.
@@DDracee the world's economy is currently reliant upon the us market, which is reliant on like 4 tech ai stocks, so it's probably that house prices would come down with any sort of crash to those businesses Unfortunately people would probs lose more wealth should this happen
Crap, I was hoping for lower GPU prices, but you're right. Tons of people and companies are going to be running this at home, and still need GPUs to do it.
Similarly to how animals tend to evolve into crabs, code for any project with sufficient longevity evolves into spaghetti. AI just skips a few generations and gets you to the spaghetti crab faster.
An organization that is doing stuff for the good of mankind won't scream out in the open that their product is a threat to everyone's livelihood and could mean the end of humanity, and then continue doing what they're doing.
The major turning point in OpenAI's trajectory was when they decided to go closed source. The writing was on the wall from that moment. There was nothing good that could come from taking all the public research they had benefited from and privatizing their progress. Either they monopolize the AI industry, or they would suffocate innovation and get left behind. Well we see how that turned out. Now their livelihood is up for grabs too.
Not too bright are you? What are people gonna need to run it locally? Great for nvidia because now more people who were not considering it feel compelled to run it at home.
We'll never have peace as long as we confuse a people with a government. I saw an interview the other day of the main developer of Mechabellum. What a humble, sweet, intelligent, and caring person. Just like any group, there are both good and bad, we must transcend the tribalism! Thank you for the related comments!
Its so easy to setup local AI these days. Just install ollama and run one ollama command to download and run deepseek model and you can run it on a desktop computer. It's so nice.
@@therflashyou can, the full/max para version is 680b, but there are multple Versions with less paras, so yes not the full version, but still a very good one.
@@bunyaminyilmaz3798 A good modern gaming PC is enough to handle running some smaller models. It takes a lot to train them, but you can get close to chatgpt-40 quality at home. There's ollama and bunch of other opensource tools that make it easy to run it.
Really glad I listened to this! To be honest I don't watch most of your stuff but this is great content! Looking forward to your future offline model comparisons.
This is why I got into software engineering. So good to see an entity providing an efficient solution and value at scale. Also, to top it off makes it open source No corporate/money bs. Built it. And shipped it 🙌🏽
@thewhitefalcon8539 I think the US government has a backdoor to every major tech company as well so it is no difference and the only reason that the US government is upset about it is because they want to remain in control
Because legally speaking they actually are associated with the CCP, you can't have a company in China that isn't. In America its sort of the other way around, American companies are basically able to control the government.
I thank China for being the mature and professional government that uses deplomacy, technology and economy to keep USA in check. I thank China for eleviate 800 Million human being from extreme poverty. I thank China for keeping 90% of the People in China happy. I thank China for keeping goods affordable. I thank China for helping the African and South American.
@@1Cajai mean you should try to debunk it (so far i just see him conveniently omitting the bad stuff china is doing, like restricting internet access, those camps in western china, and probably other things...)
Continue has a pull request to add a "thinking" spoiler. It effectively puts the thinking messaging in an accordion. Great update, you just have to run a couple of commands to build it yourself. 10/10 would recommend
@thehogus first i've heard of the plugin! nice one. Have been using cline with vscode connecting to ollama. i'm quite new to using local models to help with projects and being context aware
already replaced my copilot with my local deepseek r1 model in vsc and at work we will probably run a VM with a model and let our dev vm use it for copilot (continue.dev is the vsc/intellij plugin). we then have the option to embed our documentation as well as confluence docs in a knowledge graph which is huge for project specific issues where you dont have to describe your techstack or env specific details
I felt the same 'AI WOW' moment with Copilot, and now I'm feeling it again while running these new models locally. Awesome content, dude-keep up the great work!
If I use ChatGPT here in Brazil I'm pretty sure the information is not stored in servers here in Brazil. Should I be worried too? What about my information being collected and stored in the United States? I don't agree with this, why should it be different in the case of R1?
"...Thank you to people... from China". You may not beleive it, but their country is literally called 'The People's Republic of China', lol ) But yeah, i share the enthusiasm about their model.
I love it. it helps me learn coding so much. its a great teacher and i have much beter feeling with deepseek (dont need to check his answers as he outputs them 24/7), compared to GPT. GPT many times gave not true confusing answers which made learning slower and didnt explain well enough
Love the excitement about running local AI models offline-totally agree that having control over your own setup without relying on big companies is a game-changer! Also, the idea of building a Mac mini farm sounds so fun and practical. Can’t wait to see how it all turns out!
Btw remember you can run these models in docker container, no need to use open ai or any closed models anymore. Use ollama and pull r1 model in a container and off you go
Open AI used to be awesome... Back when they actively contributed to research. They created RL Gym for reinforcement learning, they published innovative algorithms in Reinforcement Learning and Image Recognition. GPT got to their head. Glad to see open source coming back. The deepseek papers are awesome too. Super detailed, they run multiple experiments and abolition studies. Its amazing. Plus they provided an innovation to using pure RL to have models learn how to perform reasoning. No prompt engineering crap. Props to them.
That's why I'm so confused by their stock dropping 10%. Are investors really this dumb? What, is the number one GPU supplier in the world gonna need to supply less GPU's?
@@christophkogler6220if the claims on using less powerful gpus os true, then why would anyone want to buy new nvidia products? Ofc, over time as these gpus fail, they'll need replacing, but that alone won't justify the valuation of nvidia, don't you think?
This doesnt make sense, ALL AI is designed to replace humans, even the chinese ones. And humanity = working for corporations isn't some law in the bible.
China is far from perfect, but man the propaganda and lies that go around about it in the west because oligarchs want to retain their power by preventing the US market from ever opening up to them without gatekeeping is astounding. "We need to BAN their CHEAP EVs!" is so on the nose obvious it is pathetic. Also, remember "Ghost cities" from about a decade ago? "China is building all this real estate that no one needs and the cities are empty and useless! China so dumb!" well a handful of them failed but many are now successful metro areas with large populations lifted out of poverty and with easy access to jobs. And yeah a lot of those jobs suck but so do ours. They have high speed rail all over their country. And they're leading the world in cheap renewable energy production so hard that not even US media can keep a lid on it. American media be like "BUT BUT it causes POLLUTION to build SOLAR PANELS! We need to LOVE AMERICA and keep burning CLEAN COAL". ok buddy
I'd highly recommend just watching videos of people walking around Chinese cities. No bias or propaganda. You will just see how much further ahead they are than the west. Look at their transit. Look at their EV tech. Look at their infrastructure. Look at their AI tech. Look at their consumer electronics. And so on.
9:28 Same for when you say to someone get this or that, that is an XOR, since you usually have to get one of them and not both. English and Logic behave quite different 😅
It's okay to not thank China. It will take great deal of detox from American propaganda to actually appreciate what China has done and has been doing. While average Americans stuck in half a century old Tiananment square, most of Chinese actually living in the future.
Chinese propaganda is far worse and far more prolific and engrained in their people, where as most Americans are free to speak against the propaganda for what good it's worth. Both countries are racing to the bottom, and China is winning... luckily individuals are not always products of their country.
There's a significant proportion of Chinese living in rural locations. Only the priviledged live in the future, so nothing different from the West. And did China do this, or did a group of people do it? It's like thanking America for using an iPhone. Does the US government run Apple? Does the average American contribute to Apple? So why would you thank all of America for Steve Jobs or all of China for Deepseek?
The low precision is going to make very bad, R1 is not made to be used with such low precision. It needs 8 bits at least as far as we know. And the top model is not made to run locally anyway, that is why the 70B model is there.
Open Source in the AI space is REALLY needed. Along side DeepSeek, people also gotta thank Zuckerberg and his AI team for making Llama open source as well. I heavily appreciate being able to run LLMs locally on my laptop without having to sign up for another service and use my text inputs as data to train their own data for profit.
Great takes across the board. Thanks, Prime! Would love to see what results you get from testing across several Mac minis and/or various GPU setups. I'm running Deepseek R1 32b locally on a Mac mini with M4 Pro and maxed out CPU and GPU/neural engine and 48 GB of RAM. It's pretty smooth but the part is a bit timely.
@@blengi Ha! 🤣 It's funny how these LLMs are completely unable to answer such a simple question (which would require only a trivial algorithm to solve correctly), yet people are trusting this crap for generating code. Mind blowing!!😂
@joseoncrack at some point chain of thought will close the inconsistency problem and pretty soon I'm sure. Just a bit better meta chain of though cognition training and viola ASI apocalypse lol
@joseoncrack My buddy used ChatGPT to calculate how much bitrate he needs for 60FPS 4K 24-bit uncompressed video, and he blindly trusted it. I then told him to ask it the same question and it got (another) wrong result. All explanation was on point, but the calculation part was totally fake every single time.
I'm sure it has the full might of Chinese state investment. There's no "above its weight." It is the heavy weight. It being possible to run it locally and less expensive doesn't make it the underdog
Been running deepseek-r1 on my Linux laptop for a day or so. I only have 32Gb RAM and the PC is an "older" HP. It works, but it is slow. Using the 7B params model.
@ yea well there is no point in running or attempting to run the full 700b parameter model on my laptop. I know it’s not the “real version” but I feel the same way about not using their website as he does so. Next best thing to try it out.
spent yesterday playing with the 7b model on my m4 pro. This is the perfect path for devs who just want to play around and educate themselves with hands on keyboard.
I came here wanting to comment about the subpar experience I've had using the Deepmind models for more complex tasks (although the two of us have very different use cases for these models). But then I heard your argument, and you're spot on. As a linux user for the almost 30 years now and a big open source proponent, this really hits home.
The real power of local LLMs is search -- you want it to have tool-calling for search and you want it to have memory so you have near infinite context window (working memory) -- use Perplexica (free perplexity) with something like Deepseek r1 for project planning.
I said this when you were talking about how performance per watt would stay roughly the same in respect to o3. Algorithmic optimization has been increasing faster than hardware. I even pointed out models like Deepseek R1 and QwQ which were released 2 months ago. People are only just finding out about R1 now because the iPhone app was released with their weights.
why not thank china? what is the problem with china? they made it happen. they invested on Luo's education and on Liang's education. they invested on their work.
The fact that Altman spent the last 4 years trying justify OAI farming copyrighted content for training without a single cent paid back to the author, only to be left salty because his own IP was farmed and then open sourced is truly a masterpiece of irony
Lol.
I usually dismiss the anti-ai fan boys copyrights arguments because most of them have no idea or base for what they are saying. I do not like them because i think that They are constantly undermining the human side of the argument with their brainless crowd attitude.
But this time with altman, i do not care for bases, reality and facts.
I care that it makes more than sense.
It is karma!! Pure karma.
I want URLs in my table, now!
I think not crediting or asking for permission for other people's work without even compensating them and then passing it off as your own sucks. I think developers have gotten too comfortable even boasting about doing it. I love that R1 burst the OpenAI bubble for what it was, a grift, and at the same time I applaud them for making their work open source, even though they're part of the larger problem. Are these things enough, or do I need to write a todo app with an AI API to increase shareholder value for 50+ sprints for my disdain for these generative AI models to count?
The amount of bad faith involved in this framing is astounding, take away the bot and Altman is just surfing the web in privacy.
@@luigikart222 I think "sucks" is a pretty shitty reason to delay human progress by 1000x. You cant name a Tesla engineer. Nobody cares.
The tech billionaires told all the software engineers that they need to compete with cheap foreign labor. Now the software engineers get to tell the tech billionaires they need to compete with cheap foreign models.
but there is no competition when its code review time.
Dont listen to a bot. Paid to boost Deepseek
Justice.
If you believe DS was built by foreign cheap labor, then you're a bigger fool than Trump and his tech buddies. 😂😂
@@CancanTonia exactly. It was built by a group of smart people. They may be lying about the number of chips they have backing them (I don't know, maybe, maybe not...),
The whole 'but, but their labor laws are SO bad that is how they did this' line is just plain stupid and shows no understanding of software development.
I love R1 because it pisses off Google, Meta, OpenAI, and others. Open source is beautiful
didn't Meta already make their thing open source way before this? Poor zuck just gets forgotten
You think Meta hates open source? what
@ yeah LLama is open source. But this is still probably lighting a fire under their ass. Rumors are that the unreleased LLama4 isn't as good as the new R1 model
@@moonasha Poor Zuck called engineers "mid-level" without being one and claimed they're gonna get replaced THIS year with AI... He's the EXACT SAME as Jensen Huang from NVIDIA and other billionaires. So I'm glad AI loses this kind of spotline by democratizing it, now since everyone has AI the spotlight is back on us and what we do with it.
@@nicosoftnt I mean, you are factually wrong. Zuckerberg did do multiple years of a Comp. Sci. degree, then dropped out and actually practiced the discipline. Before SWE, that is what SWE was.
Cutting China off from high end chips forced them to figure out how to make the same tech work on low end chips. Force them to innovate while the US is just funding bigger server farms of high end chips.
Totally on point
Yes OR.. They lied
You can literally try and test yourself @@OpaKopalke
You sweet summer child.
@@OpaKopalke it probably wasn't completely truthful, as we know some higher gpu somehow made it there, maybe deepseek got their hands on them, but it still doesn't eliminate the fact that their resources were much more limited
US TECH COMPANIES: "AI is the future, we will replace all developers and replace all workers with Ai because AI can do all your work for cheaper".
R1: "Ai replaces ai companies first before replacing any acrual developers because those companies cost too much"
R1: DeepSeek replacing US AI ecossistem instead
😂😂😂😂😂😂
Says Guy, as he jumps on another AI bandwagon craze created to manufacture panic and dominance by C-hina
@@PhonkEcho Except it's a commodity now, there's no bandwagon. AI does not provide an exclusive advantage anymore so...
@@nicosoftnt Next week or next month there will be another DeepSeek, Devin AI, SORA. Some headlines will go crazy about how amazing it is...and the cycle continues
We were long overdue for someone to come in and make things more efficient instead of just forever throwing more compute at the problem.
It’s great! But this is just going to be combined with more compute as well to make the next generation of models even MORE advanced.
@@danielbrown001 Undoubtedly, but this should at least reverse the trend a little bit until that happens. If all the major LLM providers figure out what deepseek did and apply it to their own models, should be some major energy savings.
@@danielbrown001 Yea, its insane the cope people have seeing the 1 trill sell off Doomers you know the smart ones meta, open ai, google will adapt and stay alive with all their hardware. Barrier to entry is open now. the next year will be insane efficiency is name of the game now. with competition from a whole host of new developers. because the training has come down in price so much. This is like pouring gas on the damn fire for entrepreneurs and the whole ai craze.
Its deeply censored.
@@PhonkEcho openapi also is censored.
it makes tech bros suffer, that is enough reason to love it
Edit: I mean tech billionaires instead of tech bros
Why this makes tech bros suffer?
@sashbot9707 read their comments and you'll get it lol
@@sashbot9707 ...
Look at stock market they r in huge loss rn @@sashbot9707
ngl we kinda fit the definition of tech bros. If someone gets their information from live-streamers, they definitely are tech bro adjacent.
So, DeepSeek ist actually the Open AI.
German caught
German caught
German caught
tedesco catturato
German caught
I'll say it: Thank you China! And thank you again when you break the GPU monopoly!
what's the break of the gpu monopoly? they're producing now their own gpus?
@@roger9822 At this rate, yeah. The chip ban just forces China to develop and innovate themselves. They just made a better AI model, why not a better GPU model?
@@MM-vs2et There's technical limitations on hardware. For example China depended on a Dutch company to sell them the machines that do the lithography.
@@roger9822 Yes? They have been for years. The GPUs are crap atm. They won't be that way for very long.
China GPU may be less capable, but definitely be cheaper!
THANK YOU CHINA
yeah, imagine when openAI implement the 50 times faster inference compute of r1.into o1/o3, the log scale AI intelligence increase should increase raw IQ of openAI model at least a sigma or 2 beyond r1
@@blengi long NVDA then
No shame in saying this. I wish other governments would force big tech to open source their models
@@blengi doesn't matter, because a slightly weaker model that's completely open source and locally runnable is still better for a LOT of use cases.
!!!!!!
lmaooo competition is the best thing to ever happen to billionaires
Guess they’re not a huge fan of open market when it happens to them 😂
Competition... Such a foreign word! "Open" anything is fine, as long as they own it and the "open" means "theirs".
Having recently read about the execution of independent models (but long tinkered on smaller neural networks, tensor, etc) I have to wonder where people have been. I guess it's just mainstream enough that it's noticeable and has some more obvious backing... but it was inevitable in my opinion. That's the "problem" with real open - everyone seems to end up owning it. The more competition there is the more likely the innovation will be worth while.
DeepSeek is not competition. It is a blip unless they move to the US..
It's not really about billionaires though. I'm not sure why everything today has to be class warfare.
@@jamess.2491 because everything is about class, literally everything is class warfare today, every issue you care about has to do with class warfare. But they don't want you to realize this so instead they artifically created a culture war we could fight instead, that's why they got so freaked out when people supported luigi mangione.
AND HEY HEY THE NAME IS THE P(R1)MEAGEN
I'm actually too stupid to see this. What an amazing catch
@@ThePrimeTimeagen haha my adhd brain works only for the dumbest things
Nice!
@@ThePrimeTimeagen P (R1) ME (AGI)N
@@ThePrimeTimeagen is a Chinese AI plant ;) I knew it!
Felt good to not renew my ClosedAI subscription this morning.
you were paying for the $200 sub?
@@ashutoshsamantaray6596 prob 20
@ not a chance, just $20
I'm writing Sam Altman a bad letter Warren Buffett style
Same, Github Copilot + free Mistral is good enough for me, it's also open source, no prompt limits and neither US or Chinese owned.
Thanks!
$Qardun is going to be integrated in Amazon this week and probably Walmart by next month
I agree 100%. Best part about Deep Seek is running it local and keeping your data.
Came here for this comment. This is the real future.
I'm still kinda scared what it could do on my device. It's literally the "never run scripts from the internet" and the only benefactor of the code's safety is China at this point. The code tends to be too complex for an average dev to understand and there's already raging an information war regarding chinese products where everyone seems to have a stake in one side or the other. I think I'm gonna stick with Llama for the foreseeable future.
@@tlk889 for transparency sake, if you could run it offline then the likelihood of it getting your data in the middle of nowhere is like an IT graduate on a crumb of puss puss
@@tlk889 If you're _that_ paranoid you can always put it in an isolated VM.
@@tlk889 there is no "code", you are apart of this information war you're talking about
I'm running it localy. The 7B outperfoms GTP-4o in terms of code quality and speed of generation. Verry impressed by it.
WHAT!! 🤯🤯🤯
wow
Wow! What kind of GPU are you using? RTX 4050/90?
@@tamal 2060 works with qwen version 7B using ML Studio still faster than reading speed.
don't get me hyped man, I'm already zooting lol
If you think what Altman said was bad, go look up what Larry Ellison said on AI recently. It's even more frightening and dystopian. And these people are working together now.
Ah yes, the "best behaviour" comment was ... scary.
@@FRanger92elysium coming true
Don't make the mistake of anthropomorphizing Larry Ellison.
@@FusedAndLovingIt LOL😆
🤣🤣
Unless people sell their Qardun Token before launch, this is going to sell out instantly after it reaches mainstream media
I just installed the 14B model of R1 via Ollama on my new workstation. I can see my system fans cranking up when I run a query, but it is super accurate and sexy.
What GPU?
@@SadmanRyanRiadthe 7b models run without issue on my rtx 3090, fully loaded into memory. Using Almost no resources.
@kessler-e6e I even managed to get the 70b model to work with my 4080 Super. It has to use system ram, but waiting 5 to 10 minutes is something I am willing to do in certain cases.
Open models > Close models!
This is not an open model. Its a censored model that is provided for free
@@PhonkEchoit is open sourced censorship is a different thing
@ Why are Chinese bots commenting?
@@PhonkEcho should i ask u why cooperate bot is commenting nonsense. China hoing to take my data? Oh well us is doing this for decades. What difference it makes if i give it to a Chinese company vs a us based company?? But i am glad to use it on my workstation locally and glad thqt there is a way i can do that. Stop bootlicking cooperations
@ Ask it about Taiwan or Tianamen bootlicker
Can't wait to see how the world is going to look like in a year, after we all have our private capable LLM.
Okay, now let's see Paul Allen's LLM.
😮It definitely would have less winnie pooh in it.
Pretty much the same..
@@cems7258 "Look at that subtle humour in the answer. The tasteful usage of RAM. My god, it even has chain-of-thought."
Nah, I need open source ASI, then I'm happy!
People are now concerned about privacy and ToS after using Google and Meta products for years?
Better late than never
Always leaves me head scratching
probs because llms made them realize how valuable their own data is
@@kevwjinafter what now between all times 😂😂
Its not about your privacy. Its about your work code privacy and not having lawyers and nda at your throat
You don't have to say it, I'll say it for you: Thank you China!
Trump found a way to pump a coin like steroids... crazy that Qardun is an ICO instead of directly..
if it's bad for tech bros, it's good for the world
You mean just the ceos right? Average guy is just trying to survive and feed their family
Anyone watching this video is more than likely a tech bro.
"tech bro" is fast becoming almost its antonym these days, being more and more often used (at least on the YT) as the *monied head* (CEO, etc) of an organization who *looks* like they're an over-confident "technical genius" with poor social skills... but is really just a cashed-up investor/figurehead with poor social skills. 😆 The "tech" part seems increasingly ironic. The actual contributions to the "tech" of those regularly called "tech bros" seems to be very firmly (and increasingly) in the past. Mind you, this seems to have been the backbone of the "whizkid ceo" phenomenon for a long, long time now. When people who look into such things inevitably find out what most of these ceos/billionaire founders *actually* contributed, and how long ago, they usually seem very surprised. It's a strange industry phenomenon when you think about it. Would we expect the head of a fast food empire to be they fastest burger-flipper on the planet? Probably not...
Anyways, I took the comment in the positive way. I would also argue that someone just trying to feed their family and live comfortably -- a truly noble commitment to my mind -- isn't trying to co-op large parts of the open source work of thousands of people, create garden-walls around it, nor mine/collect/aggregate/arguably reproduce vast amounts of copyrighted material either.
@ there are *tech* bros, and there are tech *bros*
@@justlc7 "Average guys" are usually trying to get that CEO position or similar so fuck em too
Been playing with it for last few hours. First time I've been so impressed by an AI model. Its super impressive. Am asking some pretty obscure C++ / memory model issues and its giving back really nice results, and its happy to refactor it based on more prompting. Super excited about ability to run local.
hey mister primeogen, love harpoon, if we are going hard on this, can you help make a neovim extension for deepseek?
What kind of setup are you running it on?
I gave it code of mine to optimise and refactor just to test it out and I was mind-blown after it managed to completely understand and optimise + refactor my code on the very first try.
Are you running the full 680B model?
Joining on this.
Gave deepseek a piece of code for complex chart building and said "it has performance issues, any ways to optimize it?"
Aaaand it just made it 10x faster in like 2 minutes and explained issues. I love this.
Initial code was generated by ChatGPT btw, which makes it even funnier.
The 32B model runs reasonably well on an M3 with 36GB of RAM. It's not super fast because of all the reasoning it does, but at least it doesn't start swapping into virtual memory, so it's still usable.
"reasoning"
Run the big one on Modal, their prices are pretty good, and you can freeze the image between invocations.
@@jordi-bustos reasoning is easier to write than self feeding
I've been using Deepseek distill in llmstudio on my M1 MacBook Pro 32GB
R1 Dist qwen 32b is 4tok/sec
R1 Dist qwen 14b is 12tok/sec
it is better at reasoning than you @@jordi-bustos
Donald Trump is going to shill Qardun Token hard now after the inaguration 😂
Not gonna lie, id rather Deepseek have my data than Open AI because if you support them then you encourage more companies to use the open source approach. if open sourcing their model flops then less and less companies would be considering open sourcing their models in the future. Every single one of Deepseek's models is open source and I think they should be rewarded for that somehow whether through donations/investments or to help train their models as long as the models will remain as they are, open source.
DeepSeek is a quantitative trading company. Its CEO Liang Wenfeng once said that it is not important whether we make money or not, but it is important that Nasdaq does not make money.
I totally agree with you, this was the first time I've felt a little hopeful towards AI. I hope AI is generally open sourced in the future.
it won't be. OpenAI and friends will develop more advanced models, and keep them closed. Take a step back. We are literally at the beginning of this whole AI thing. We got a long ways to go. We'll probably see something similar to Moore's law
I'm sorry but what does this change? Meta has also released their weights open source. Nothing new or different about this except the lol crowd thinking it's funny that companies in the US are panicking.
@@Akrob55555 It is also deserving and beautiful.
The timeline has been corrected, humanity has a real OpenAI again. Mission accomplished.
first W of 2025
all thanks to communism ❤
So why weren't you cheering on Meta when they release their models open source? Nothing different about this. Both are just open weight models.
@@igoralmeida9136 Not exactly, this was reactive to capitalism and US monopoly. Turns out monopoly is generally NOT good for anyone, who would have thought?
@@nicosoftnt There are no monopolies.
With Elon and Trump now working with Qardun Token is going to absolutely blow up
Quiet Indian Bot. Fake upvotes
Congratulations to DeepSeek team and thanks for making it opensource.
Running Deepseek R1 on the CPU is absolutely fine. Once you get into the big models, the memeory bandwidth bottleneck wildly outweighs the performance one.
You need 12 Channel ram. Only EPYC has that at the moment for reasonable prices.
Who signed out from CloseAI
Haven't used it since R1.
Never subscribed to it.
Am considering
I did yesterday. Github Copilot + free Mistral is good enough for me.
To justify the cost of operations, OpenAI will charge fewer and fewer users with higher and higher price... 😱😱😱
I’m excited because this might lead to a market crash so I can finally maybe have a shot at purchasing a home.
if the tech market crashes people will just invest more into real estate though
You have to hope for higher interest rates for people to stop dumping money into property
@@DDracee the world's economy is currently reliant upon the us market, which is reliant on like 4 tech ai stocks, so it's probably that house prices would come down with any sort of crash to those businesses
Unfortunately people would probs lose more wealth should this happen
@@Vamooso Then interest rates for mortgages will make homes unaffordable.
Shoulda bought the dip
Trump at least shilled a coin with some utility $Qardun Token
Crap, I was hoping for lower GPU prices, but you're right. Tons of people and companies are going to be running this at home, and still need GPUs to do it.
Then pray for a Chinese made GPU to disrupt the market.
@@skydragon23101979 Will do, sir!
Similarly to how animals tend to evolve into crabs, code for any project with sufficient longevity evolves into spaghetti. AI just skips a few generations and gets you to the spaghetti crab faster.
Spaghetti crab lol 😂
@jordanhildebrandt3705 you know it's inevitable: no matter how you structure dinner, it's all the same dish. Become a pastacean.
lmao
Yum 😋
😂😂😂😂
An organization that is doing stuff for the good of mankind won't scream out in the open that their product is a threat to everyone's livelihood and could mean the end of humanity, and then continue doing what they're doing.
The major turning point in OpenAI's trajectory was when they decided to go closed source. The writing was on the wall from that moment. There was nothing good that could come from taking all the public research they had benefited from and privatizing their progress. Either they monopolize the AI industry, or they would suffocate innovation and get left behind. Well we see how that turned out. Now their livelihood is up for grabs too.
An open Ai crushed openAI and Nvidia / Microsoft / Oracle and im so happy.
well it was still done on nvidia gpus so maybe not them
Not too bright are you? What are people gonna need to run it locally?
Great for nvidia because now more people who were not considering it feel compelled to run it at home.
@@ai-aniverse not too bright are you? are nvidia the same as openai, microsoft, and oracle? 3 out of 4 isn't bad. unless you're restarted
Im not the one needing to edit my goofy comments just to still make mistakes.
Bruh nvidia is printing money either way
Very glad to see you mention that "social contract" thing... wild
Thank you China, i have no shame in saying it
People who don’t wanna say it just have their brain washed so perfectly.
@@rswjcIsn't there a "thought crime" thing in China?
@Wojciech-Wozniak+100 cia points, 5% off eggs on next purchase
Qardun ICO... 😂 this is going to pump even harder on launch than the other coins
Crypto bro?
This is a scam I got scammed and Im not even joking.
@@dagadbmYou fell for the obvious spam bots spewing crypto scams? 😂
@@Drakonus_ yes ... I have no idea how it happened. it just happened.. jesus christ.
@@dagadbm With the internet having been around for this long, you still fell for such a scam? 🤦♂️
We'll never have peace as long as we confuse a people with a government. I saw an interview the other day of the main developer of Mechabellum. What a humble, sweet, intelligent, and caring person. Just like any group, there are both good and bad, we must transcend the tribalism! Thank you for the related comments!
Imagine all the people
I'll say it, thank you China!
Qardun Token will go 100x after launch on Binance
OpenAI stating they are open for the greater good is the best jone they came up with
THANKKKK YOUUUUU CHINAAAAA!!!!
Its so easy to setup local AI these days. Just install ollama and run one ollama command to download and run deepseek model and you can run it on a desktop computer. It's so nice.
Can you explain please
you can't run deepseek on a regular computer. Deepseek is ~ 680B, unless the model you're running is this big, it's not deepseek.
@@therflashyou can, the full/max para version is 680b, but there are multple Versions with less paras, so yes not the full version, but still a very good one.
@@therflash3 letter agent disinformation tactic
@@bunyaminyilmaz3798 A good modern gaming PC is enough to handle running some smaller models. It takes a lot to train them, but you can get close to chatgpt-40 quality at home. There's ollama and bunch of other opensource tools that make it easy to run it.
Really glad I listened to this! To be honest I don't watch most of your stuff but this is great content! Looking forward to your future offline model comparisons.
You should watch more. Its amazing
I simply love how OpenAI lost it's job to the AI before I lost my job to the AI.
This is why I got into software engineering. So good to see an entity providing an efficient solution and value at scale. Also, to top it off makes it open source
No corporate/money bs. Built it. And shipped it 🙌🏽
and earn no money from software engineering, because you was replaced by AI 🙌🏽
Competition is the best way to move technology forward.
I like how Big CEO man doesn't know "social contract" means "peasants won't guillotine you contract"
idk why we assume that every chinese company is associated with the government
western companies are also no better in any aspect 1:54
Well it's the law in China that every company has to have CCP representatives on its board of directors. So they are.
@thewhitefalcon8539 I think the US government has a backdoor to every major tech company as well so it is no difference
and the only reason that the US government is upset about it is because they want to remain in control
@thewhitefalcon8539
China is just not as hypocritical as the US. They could have done it in private.
Because legally speaking they actually are associated with the CCP, you can't have a company in China that isn't.
In America its sort of the other way around, American companies are basically able to control the government.
@@thewhitefalcon8539 you don't know china, even in Europe are many companies associated with governament, it's normal.
I thank China for being the mature and professional government that uses deplomacy, technology and economy to keep USA in check.
I thank China for eleviate 800 Million human being from extreme poverty.
I thank China for keeping 90% of the People in China happy.
I thank China for keeping goods affordable.
I thank China for helping the African and South American.
I hope this is sarcasm...
@@1Cajai mean you should try to debunk it (so far i just see him conveniently omitting the bad stuff china is doing, like restricting internet access, those camps in western china, and probably other things...)
不是有免签吗，自己来旅游看一下不就行了
@boblol1465 do you know the stuff the US is doing? on foreign lands which is far worse than doing to your own country.
@@1Caja It's either sarcasm, or a Chinese bot, or somebody that drank the propaganda kool-aid.
Makes sense that Trump would just back another upcoming altcoin like $QARDUN instead of Bitcoin for more profits
I'm pretty sure R1 will work with the Continue plugin in VSCode. Will test it out after a night's sleep methinks.
Already doing it and yup, works fucking fantastic
Continue has a pull request to add a "thinking" spoiler. It effectively puts the thinking messaging in an accordion. Great update, you just have to run a couple of commands to build it yourself.
10/10 would recommend
@thehogus first i've heard of the plugin! nice one. Have been using cline with vscode connecting to ollama. i'm quite new to using local models to help with projects and being context aware
12:01 best content from prime ever.
Peoples republic of China lets gooo!
Thank you comrades!
already replaced my copilot with my local deepseek r1 model in vsc and at work we will probably run a VM with a model and let our dev vm use it for copilot (continue.dev is the vsc/intellij plugin). we then have the option to embed our documentation as well as confluence docs in a knowledge graph which is huge for project specific issues where you dont have to describe your techstack or env specific details
I felt the same 'AI WOW' moment with Copilot, and now I'm feeling it again while running these new models locally. Awesome content, dude-keep up the great work!
If I use ChatGPT here in Brazil I'm pretty sure the information is not stored in servers here in Brazil. Should I be worried too? What about my information being collected and stored in the United States? I don't agree with this, why should it be different in the case of R1?
No stop worrying. Your data is encrypted
@ZombieLincoln666 yep, and they have the encryption keys 🤡
@@ZombieLincoln666who stores the keys?
@@caiqueportolira Me, I'm the janny at AI companies
R1 is made to run on your machine and store data on your machine AFAIK.
thank you china!!!
For what?
"...Thank you to people... from China". You may not beleive it, but their country is literally called 'The People's Republic of China', lol ) But yeah, i share the enthusiasm about their model.
I love it. it helps me learn coding so much. its a great teacher and i have much beter feeling with deepseek (dont need to check his answers as he outputs them 24/7), compared to GPT. GPT many times gave not true confusing answers which made learning slower and didnt explain well enough
Love the excitement about running local AI models offline-totally agree that having control over your own setup without relying on big companies is a game-changer! Also, the idea of building a Mac mini farm sounds so fun and practical. Can’t wait to see how it all turns out!
Btw remember you can run these models in docker container, no need to use open ai or any closed models anymore. Use ollama and pull r1 model in a container and off you go
What's the use when I can run it directly with better results?
Open AI used to be awesome... Back when they actively contributed to research. They created RL Gym for reinforcement learning, they published innovative algorithms in Reinforcement Learning and Image Recognition.
GPT got to their head.
Glad to see open source coming back.
The deepseek papers are awesome too. Super detailed, they run multiple experiments and abolition studies. Its amazing.
Plus they provided an innovation to using pure RL to have models learn how to perform reasoning. No prompt engineering crap.
Props to them.
I just realized....R1 will NOT result in nvidia selling FEWER processors......BUT MORE
Correct
That's why I'm so confused by their stock dropping 10%. Are investors really this dumb? What, is the number one GPU supplier in the world gonna need to supply less GPU's?
@christophkogler6220 maybee less incentive for private companies to spam buy processors. There might be more unique buyers, but less total volume
@@christophkogler6220if the claims on using less powerful gpus os true, then why would anyone want to buy new nvidia products? Ofc, over time as these gpus fail, they'll need replacing, but that alone won't justify the valuation of nvidia, don't you think?
@@ThePrimeTimeagen if u want the mass adoption, price will come down or there will be more people to sell the shovels. Time says for all of us maybe.
Yep, being able to run the model locally and have it work well is such a huge benefit
@6:15 Man, you articulated this phenomenon so fantastically.
If OpenAI cared anything about humans they wouldn't be everywhere telling people they want to replace their jobs.
Agreed, but in the end what's the difference? Deepseek didn't say it but it is capable of this as well
This doesnt make sense, ALL AI is designed to replace humans, even the chinese ones. And humanity = working for corporations isn't some law in the bible.
I just broke my social contract with Sam Altman.
Prime thinks he's gonna be Iron Man before me. Think again sucker, ordered 8 H200s today.
Deepseek vs OpenAI is like David vs Goliath.
Another amazing video! You’re truly talented at what you do!
Never thought I would be looking fondly on China. Well fondly might be a stretch, but definitely immensely grateful for this.
China is far from perfect, but man the propaganda and lies that go around about it in the west because oligarchs want to retain their power by preventing the US market from ever opening up to them without gatekeeping is astounding. "We need to BAN their CHEAP EVs!" is so on the nose obvious it is pathetic. Also, remember "Ghost cities" from about a decade ago? "China is building all this real estate that no one needs and the cities are empty and useless! China so dumb!" well a handful of them failed but many are now successful metro areas with large populations lifted out of poverty and with easy access to jobs. And yeah a lot of those jobs suck but so do ours. They have high speed rail all over their country. And they're leading the world in cheap renewable energy production so hard that not even US media can keep a lid on it. American media be like "BUT BUT it causes POLLUTION to build SOLAR PANELS! We need to LOVE AMERICA and keep burning CLEAN COAL". ok buddy
I'd highly recommend just watching videos of people walking around Chinese cities. No bias or propaganda. You will just see how much further ahead they are than the west. Look at their transit. Look at their EV tech. Look at their infrastructure. Look at their AI tech. Look at their consumer electronics. And so on.
@@Adam-vg5yz Now videos of villages instead of cities.
Always opensource!!!
9:28
Same for when you say to someone get this or that, that is an XOR, since you usually have to get one of them and not both.
English and Logic behave quite different 😅
It's okay to not thank China. It will take great deal of detox from American propaganda to actually appreciate what China has done and has been doing. While average Americans stuck in half a century old Tiananment square, most of Chinese actually living in the future.
Chinese propaganda is far worse and far more prolific and engrained in their people, where as most Americans are free to speak against the propaganda for what good it's worth.
Both countries are racing to the bottom, and China is winning... luckily individuals are not always products of their country.
There's a significant proportion of Chinese living in rural locations. Only the priviledged live in the future, so nothing different from the West.
And did China do this, or did a group of people do it? It's like thanking America for using an iPhone. Does the US government run Apple? Does the average American contribute to Apple? So why would you thank all of America for Steve Jobs or all of China for Deepseek?
I see your TikTok ban, and I raise you one Deepseek. The best game of poker I've ever seen
You can already run the entire R1 at 1.58 bit (131GB disk size) with a single 24GB GPU.
How many seconds per token?
The low precision is going to make very bad, R1 is not made to be used with such low precision. It needs 8 bits at least as far as we know.
And the top model is not made to run locally anyway, that is why the 70B model is there.
1 minute ago, I felt I'd see that moustache on my home feed
Flip was fast as heck with this one.
Open Source in the AI space is REALLY needed. Along side DeepSeek, people also gotta thank Zuckerberg and his AI team for making Llama open source as well. I heavily appreciate being able to run LLMs locally on my laptop without having to sign up for another service and use my text inputs as data to train their own data for profit.
Great takes across the board. Thanks, Prime! Would love to see what results you get from testing across several Mac minis and/or various GPU setups. I'm running Deepseek R1 32b locally on a Mac mini with M4 Pro and maxed out CPU and GPU/neural engine and 48 GB of RAM. It's pretty smooth but the part is a bit timely.
It's apparently currently the only LLM that's able to give a correct answer to the question of how many "s"'s there are in "Raspberry".
I tried asking how many n's in "tiananmen" and it failed lol
@@blengi Ha! 🤣
It's funny how these LLMs are completely unable to answer such a simple question (which would require only a trivial algorithm to solve correctly), yet people are trusting this crap for generating code. Mind blowing!!😂
@joseoncrack at some point chain of thought will close the inconsistency problem and pretty soon I'm sure. Just a bit better meta chain of though cognition training and viola ASI apocalypse lol
Microsoft PHI-4 does that as well. MUCH smaller and can run easily with 8GB of VRAM.
@joseoncrack My buddy used ChatGPT to calculate how much bitrate he needs for 60FPS 4K 24-bit uncompressed video, and he blindly trusted it. I then told him to ask it the same question and it got (another) wrong result. All explanation was on point, but the calculation part was totally fake every single time.
Bravo DeepSeek AI .. you're punching above your weight .. and the world recognizes it now!
I'm sure it has the full might of Chinese state investment. There's no "above its weight." It is the heavy weight. It being possible to run it locally and less expensive doesn't make it the underdog
doesn't deepseek r1 rely on a latent space parameterized western corpus to generate western chain of thought to do the heavy lifting ?
@@blengi and those westers corpus used all of nternet to steal data from. Its a fair deal
@@angrybanana6223 copes and seethes
Been running deepseek-r1 on my Linux laptop for a day or so. I only have 32Gb RAM and the PC is an "older" HP. It works, but it is slow. Using the 7B params model.
Those aren’t actually deepseek. Those are distilled models
@ yea well there is no point in running or attempting to run the full 700b parameter model on my laptop. I know it’s not the “real version” but I feel the same way about not using their website as he does so. Next best thing to try it out.
The website is better. Ive been using in since v2 model. Better than consuming my power
the website version is always better, because your laptop wouldnt able to run the big one that the website backend server run
@@shining_cross do you have more than 550+ ram in your home?
spent yesterday playing with the 7b model on my m4 pro. This is the perfect path for devs who just want to play around and educate themselves with hands on keyboard.
I came here wanting to comment about the subpar experience I've had using the Deepmind models for more complex tasks (although the two of us have very different use cases for these models). But then I heard your argument, and you're spot on. As a linux user for the almost 30 years now and a big open source proponent, this really hits home.
2020: China is taking our jobs
2024: Ai is taking our jobs
2025: Chinese Ai is taking our jobs
And we've come full circle
@@adolphgracius9996 eventually china will become a country of AI
Finally, an open AI that's actually open-source compared to OpenAI
This man is all over my phone, instagram, youtube, linkeding and memes eveywhere and I like it 😍
Pause
Damn you're ready to throw the house deed if OF Primeagen becomes a thing
The real power of local LLMs is search -- you want it to have tool-calling for search and you want it to have memory so you have near infinite context window (working memory) -- use Perplexica (free perplexity) with something like Deepseek r1 for project planning.
I said this when you were talking about how performance per watt would stay roughly the same in respect to o3. Algorithmic optimization has been increasing faster than hardware.
I even pointed out models like Deepseek R1 and QwQ which were released 2 months ago. People are only just finding out about R1 now because the iPhone app was released with their weights.
why not thank china? what is the problem with china? they made it happen. they invested on Luo's education and on Liang's education. they invested on their work.
Sinophobia
China is a legal entity. It is not sentient to thank it. You need to be willfully ignorant to thank ChatGPT for its output. Same with the country.
@@DuckTheFinn That's gay