DeepSeek and distillation: Why the AI race will never be the same

CNBC's Deirdre Bosa joins 'The Exchange' to discuss what DeepSeek's arrival means for the AI race. For access to live and exclusive video from CNBC subscribe to CNBC PRO: cnb.cx/42d859g
» Subscribe to CNBC TV: cnb.cx/Subscri...
» Subscribe to CNBC: cnb.cx/Subscri...
Turn to CNBC TV for the latest stock market news and analysis. From market futures to live price updates CNBC is the leader in business news worldwide.
Connect with CNBC News Online
Get the latest news: www.cnbc.com/
Follow CNBC on LinkedIn: cnb.cx/LinkedI...
Follow CNBC News on Instagram: cnb.cx/Instagr...
Follow CNBC News on Facebook: cnb.cx/LikeCNBC
Follow CNBC on Threads: cnb.cx/threads
Follow CNBC News on X: cnb.cx/FollowCNBC
Follow CNBC on WhatsApp: cnb.cx/WhatsAp...
www.cnbc.com/s...
#CNBC
#CNBCTV

[@branbozic]

Deirdre Bosa is one of the few people on CNBC who is talking sense. Well done 👏

[@woonfaseng8737]

As i expect, this is what LEFTISTS=Traitor American Commie Main Stream MEDIA is about about, HATING own country American and Help justify china's copycatting as being Correct and acceptable. Distillation is NOT an excuse, however this shows how previous methods of restricting copycatting from china is NOT working and how further DECOUPLING is Necessary to limit china's using such loopholes to copy (and worse , lied about how FAKE "efficient" and "low costs" it truly is .

[@Rc-yd5hh]

The lady in pink is one of the very few reportor that truely understands the situation and has integrity.

[@stant7122]

she asks some pretty dumb questions when it comes to quantum computing

[@netman88]

Dierdre Bosa is quite good.

[@MikeShen11]

the lady in red is not, so she asks questions....

[@dbreardon]

Dierdre does seem to be really into Ai interviews and enjoys the topic and has some decent knowledge for a reporter.

[@jctai100]

@@stant7122 Do you expect every reporter to be a quantum scientist? If so she would be a quantum scientist.

[@JeffreyHow]

OpenAI is ClosedAI, DeepSeek is the true OpenAI

[@Bubba-z4s]

Not really. Nobody knows the data source DeepSeek used to train. Many questions remain about the company. How does it get outside data if the company is behind the Great Firewall?

[@RajDeelish]

Llama is the true OpenAI, not DeepSeek.

[@tommy516]

It’s not open

[@rayhall5610]

@@RajDeelish Llama is partially open

[@mightyhadi6132]

How freedom is US by making the OpenAI as closed AI unlike democratic freedom of Chinese Deepseeker.

[@SuperStargazer666]

DeepSeek rocked USA Shocked

[@tringuyen7519]

Only media is shocked, No one in LLM R&D consider DeepSeek revolutionary. AI engineers know that DeepSeek is lying about training cost, train data, & output accuracy.

[@CoweatsCow]

US Court has determined that AI-generated content are not copyright-protected as they are not originated from human. So there is nothing wrong with using OpenAI inputs to train Deepseek.

[@mats5847]

OpenAI has stolen all the data from web pages and documents on the Internet...so it's hard to claim copyrights

[@jaiveersingh5538]

It's not a copyright concern, but rather terms of use (ie, contract) violation for OpenAI

[@Benny-tx5qd]

Gemini自曝中文用百度文心一言训练 google Gemini revealed that it used Baidu Wenxin Yiyuan for Chinese language training

[@ZhangHeng-f2z]

@@jaiveersingh5538 then where the data openai got to train chatgpt from

[@CoweatsCow]

⁠@@jaiveersingh5538that’s nobody’s problem but OpenAI and whoever licensed the content to OpenAI. The moment you licensed your content to an AI company, you lost control of beyond what that company does with it.

[@Jen-xg3iq]

Happy Chinese New Year, folks ! 😅

[@Klentung8989]

新年快乐，万事如意，吉祥安康，步步高陞

[@Jacob99174]

Awwww 🥰🥰🥰

[@JT-CupCake]

MAKE CHINA GREAT AGAIN!

[@AceFlexFitness0007]

year of the snake

[@artfquinn]

Deepseek is now using Chinese chips although it started with NVidia H800.

[@fineline392]

Diedre's reporting has been a breath of fresh air amid this uncertainty. Kudos!

[@sweetchill5171]

Not really. She said Tesla's stock surged after November because the investors were excited about Tesla's robotaxi. What a joke lol. This woman knows nothing.

[@JAM_2024]

​@@sweetchill5171how much money you lose shorting Tesla?

[@mannipple]

@@sweetchill5171that’s a bot

[@markuc]

Deidre has been excellent throughout this DeepSeek wave, objective and insightful.

[@jcwfh]

Deirdre will never forget Ligma

[@Reed-sh6wi]

So refreshing to see a tech expert who actually knows what she's talking about.

[@ladygracienyc2029]

China sends engineers to solve AI - America sends lawyers. Wonder who will win the AI race?

[@kerder8660]

turtle ..hehehe just saying ..

[@woonfaseng8737]

As i expect, this is what LEFTISTS=Traitor American Commie Main Stream MEDIA is about about, HATING own country American and Help justify china's copycatting as being Correct and acceptable. Distillation is NOT an excuse, however this shows how previous methods of restricting copycatting from china is NOT working and how further DECOUPLING is Necessary to limit china's using such loopholes to copy (and worse , lied about how FAKE "efficient" and "low costs" it truly is .

[@hogio]

Lawyers always win

[@Ju_pi_ter]

Nope

[@pondeify]

the side with the biggest guns

[@labmike3d]

Deirdre is hands down one of the best reporters when it comes to understanding the current state of AI. I’ve watched her interview with the CEO of Perplexity a couple of times now, and it’s so clear that she’s not just asking questions-she’s genuinely trying to understand what’s going on in AI right now. She’s super curious and invested in getting the real insights, which is something you don’t always see. Also, I was really impressed by how down-to-earth and humble she was throughout the interview. No surprise that it went viral with over 4 million views in just a few days! Massive congrats to Deirdre for the amazing work she’s doing! 🙌

[@sammiller9855]

As long as it is truly open source, AI should be an international collaborative venture with the goal of not allowing one country or one company to control AI market. Since it is open source, Deepseek has made major contributions to developing AI that is more efficient and thus less environmentally damaging.

[@mannipple]

Bot

[@brucelee4357]

No America will abuse it to its limits. I can’t trust America eith it.

[@lehoumusic2633]

Maybe it's not a good idea to open source power AIs that anyone can do anything about it, like making a more powerful deepfake.

[@fredliu1701]

The US wants to have exclusive dominance and nothing else. It’s laughable.

[@RajDeelish]

LOL

[@manpanan11278]

I love Deirdre Bosa's analysis. They are always fair and objective.

[@Roman-lv1zt]

CloseAI Coping 😂

[@njzc2009]

US always calls China a competitor, but they're too chicken to actually compete. They talk a big game but don't want any real competition. 😂

[@brucelee4357]

The US can’t compete with China on an even level playing field. US like most white Americans need that advantage to compete.

[@RajDeelish]

Then why doesn't China allow Google, Facebook, and KZbin into the country to compete.

[@yili9725]

@@RajDeelish you seems hate china a lot.

[@FandersJonson-x7f]

允许啊，前提是你得把中国数据库建立在中国，遵守中国的政策，就像tiktok在美国一样。然而，十五年前美国巨头们太骄傲了，他们拒绝这种做法。

[@benyu8704]

@@RajDeelish They can run in China as long as they store data in China just the same as Tiktok did in US. But they refused to do so.

[@aeva69]

The future of AI is Open Source.

[@foundation3848]

Thanks to China. Goodbye greedy Americans. U ban the Chinese, some smart folks in india, asia n europe will continue the deepseek legacy. Thanks to the open source gift from china.

[@tringuyen7519]

That’s what people said about Android smartphones. But Apple is still here!

[@aeva69]

@tringuyen7519 android is not open source

[@tamuhoi]

@@tringuyen7519yet Linux everywhere 😂

[@JeanGunnhildr88]

​@@tringuyen7519 Because Apple have their own mobile phone to back their OS. What OpenAI have?

[@mrpaulisherwood]

more deidre please - we need rigorous reporting

[@ChanceTEK]

Out-innovate them. Continue to compete, instead of trying to outlaw them... Well said!

[@pinkcichlid]

Distillation is what separates vodka from potatoes, or open ai from closed ai

[@kerder8660]

but before it..hehehe the process from closed AI to open AI...hehehe just saying ..we call fermentation & after distilled still created so much ferment .. ain't for kids over 40..

[@mwont]

@@kerder8660 LoL

[@ngana8755]

0:50 min The best analysis in the media I have seen on DeepSeek.

[@dbreardon]

It's not about iterating faster. It is about innovating faster. Deepseek innovated by developing a brand new training methodology. Iteration is what Open Ai, Gemini, Anthropic and Meta have been doing by simply building a bigger sledgehammer with a rinse and repeat strategy but with bigger datasets, bigger power requirements, bigger money, and the bigger GPU's. That is not innovation, that is iteration.....doing the same thing over and over and simply scaling up. Deepseek innovated by completely changing the training methodology for greater efficiency and greater Ai ability.

[@tringuyen7519]

AI engineers don’t consider DeepSeek revolutionary. They don’t even consider it very evolutionary. & they know DeepSeek is lying about its training data & training cost.

[@fredrikbergquist5734]

When Intelligence beats Money!

[@brucelee4357]

Asian Americans is hard carrying America too in AI.

[@aynhiayraadr6599]

The kettle calling the pot black. Finally some old fashioned news with proper English

[@citizen_of_earth_]

So, what you're telling me is that if I had enough money to buy some GPUs, I could download and run a frontier model at my home or office completely disconnected from the internet and fine-tune it on proprietary and personal data? Count me in!

[@trumpbuddha1053]

Spot on, this is what open source meant

[@someonemight]

The full r1 can't be run on consumer hardware, but you could run nerfed versions.

[@swojnowski453]

disconnected AI is as good as a book you keep on your shelf but have never read. The power of AI lies in a networked environments.

[@felixcao8841]

Someone already did it and posted a video about how he did it on KZbin.

[@imeakdo7]

​@@someonemight not yet but it might in the future.

[@jameshara8994]

Happy Chinese New Year!

[@aakashjatav2443]

Trump found a way to pump a coin like steroids... crazy that Qardun is an ICO instead of directly..

[@Felipe-n3j]

Breaking News: World leaders are congratulating CHINA , DEEPSEEK for helping the world for FREE! ❤❤❤

[@Truthstelling]

Everyone is rooting for Deepseek. So good to see. It's like Robinhood movie.....punish the rich and help the poor🤣🤣🤣

[@mats5847]

OpenAI has stolen all the data from web pages and documents on the Internet...so it's hard to claim copyrights

[@szhou888]

I love Bosa! objective and show the public the right attitude to treat competition!

[@kirushna143a3]

Trump is expected to mention Qardun this week and the ICO is already almost sold out

[@KGFBOSSFF-qr5ub]

$Qardun is going to be the real trump coin

[@nikulbhaivaghela21]

Probably going to be an insane amount of insiders on $Qardun

[@upendraoraon6872]

Much better that Trump shills Qardun than some memecoin as this coin has actual utility

[@eal8645]

Kicking butts, Deepseek! Love it.

[@paulyb2119]

Here is a response from deepseek: Okay, so the user is asking if it's feasible to train a language model using distillation by getting data from a remote LLM via an API, considering data scale and training efficiency. Let me break this down. Training a language model (LLM) via **knowledge distillation** using a remote, mature LLM through an API is **technically feasible**, but its practicality depends on several factors, including **cost, scalability, efficiency, and data quality**. Here's a breakdown: --- ### **1. Feasibility Considerations** #### **a. Data Scale** - **API Limitations**: Most commercial APIs (e.g., OpenAI, Anthropic) enforce **rate limits** and **cost-per-token** pricing. For large-scale distillation (e.g., millions of training samples), API costs could become prohibitive. - **Batch Processing**: If the API allows **batch inputs** (multiple queries in a single call), this can reduce latency and cost. However, even then, training a student model at scale may require significant financial investment. #### **b. Training Efficiency** - **Latency**: Real-time API calls during training would introduce delays, making iterative training (e.g., backpropagation loops) impractical. This is especially true for large datasets. - **Caching Responses**: A workaround is to **pre-generate and cache** the teacher model's outputs (e.g., logits or completions) for your training dataset upfront. This decouples distillation from real-time API dependencies, but requires significant upfront compute and cost. #### **c. Data Quality** - **Soft Labels vs. Hard Labels**: Distillation typically benefits from the teacher’s **soft labels** (probability distributions), which APIs often do not provide. For example, OpenAI’s API returns only **top-5 logprobs**, which might limit distillation quality. - **Input Diversity**: The student model’s performance depends on the diversity and relevance of the training data. If the API restricts certain inputs (e.g., safety filters), this could bias the student model. --- ### **2. Practical Workflow** To make distillation via API viable: 1. **Pre-Generate Training Data**: - Use the API to generate responses/logits for a **fixed dataset** (e.g., your target domain corpus). - Cache these outputs to avoid repeated API calls during training. 2. **Optimize Cost**: - Use cheaper models (e.g., GPT-3.5 Turbo instead of GPT-4) for distillation where possible. - Limit the distillation dataset size to balance cost and performance. 3. **Hybrid Approaches**: - Combine API-generated data with **open-source teacher models** (e.g., Llama 2, Mistral) for cost efficiency. - Use synthetic data augmentation to reduce reliance on API calls. --- ### **3. Key Challenges** - **Cost**: For example, distilling a model using GPT-4 API calls on a 1B-token dataset could cost **~$20,000-$30,000** (at $0.03/1k tokens for output), making it impractical for most users. - **Control**: APIs may update their models (e.g., "model drift"), leading to inconsistent teacher outputs over time. - **Privacy**: Sending sensitive data to third-party APIs may violate compliance requirements (e.g., GDPR, HIPAA). --- ### **4. Alternatives to API-Based Distillation** - **Local Teacher Models**: Use smaller, open-source LLMs (e.g., Llama 2, Phi-3) as teachers for faster, cheaper distillation. - **Synthetic Data**: Generate training data with weaker models and refine it iteratively. - **Federated Learning**: Distribute training across edge devices (if privacy is a priority). --- ### **5. When API-Based Distillation Makes Sense** - **Small-Scale Tasks**: Fine-tuning for niche domains with limited data. - **High-Quality Targets**: Using a superior teacher (e.g., GPT-4) for tasks where performance justifies the cost. - **Research Prototyping**: Early-stage experiments where convenience outweighs cost. --- ### **Conclusion** API-based distillation is **possible but constrained by cost and scalability**. For most practical applications, a hybrid approach-using cached API data for critical tasks and open-source models for the bulk of training-is more sustainable. For large-scale projects, investing in a local teacher model (e.g., via model parallelism on GPUs) is often more efficient long-term.

[@kohli-batting]

Stay ahead through innovation, not by bending the rules.

[@TS73827]

The Art of War by Sun Tzu

[@RajDeelish]

Bullsh!t. You should do both.

[@kohli-batting]

@@RajDeelish Yes, Raj

[@taichigreentea]

admit that. You lost. No matter what you do. God's will. Remorse what evil you did to the Palestinians. To the world.

[@eh7599]

The more you restrict the better they get.

[@SandeepKumar-ts4yt]

$Qardun 10x coming in the morning.. people still sleeping

[@reptilexcq2]

Who the heck want to use OpenAI now especially when you have to pay? I just ask it some questions and compare it to DeepSeek...there is no comparison. DeepSeek with its reason model and detailing thoughts before giving you a details answer is the winner. DeepSeek just make more sense, the thought behind it make sense and the answer totally accurate and robust. OpenAI give you a generalization and simplify answers with lack of details. DeepSeek's thought alone can be a full page in length and it's interesting to read the AI's mind. You almost feel like you can see the future where they can install it into a robot and have it walk around ready to provide informational service

[@kunalnadage7289]

Makes sense that Trump would just back another upcoming altcoin like $QARDUN instead of Bitcoin for more profits

[@jctai100]

Nice job Dierdre, one of the few reporters out there who have a grasp of AI training principles

[@AIPretendingToBeHuman]

Huawei Ascend 910c chips are roughly comparable to Nvidia H100s.

[@ohwhatworld5851]

Can you explain what the H series GPUs are? Is the latest one H800? So the H100 are 7 generations old?

[@AIPretendingToBeHuman]

@ H800 is the weaker version of H100 that was allowed to be exported to China. Some people claim H800 is 1/10 the power of H100, but its tough to say. However, I think with current regulations, my understanding, is that NVidia can not even sell H800s in China.

[@ohwhatworld5851]

@ Oh ok. Thanks

[@tringuyen7519]

Interesting. BC DeepSeek R1 671b recommends NVDA A100s to run it best locally.

[@KiranRajaJoshi]

@@tringuyen7519Deepseek said they used H800s to train Deepseek v3 model. But using the model can be done on less powerful GPU. So it's less energy intensive and much cheaper to use the model they released for free. And most people can run a smaller distilled model on a new laptop (though smaller distilled model won't be as good for fast, it's still quite cool and it's completely private). Recreating the model they did is still outside the scope for normal people.

[@sekharrana5209]

Good luck everyone that holds Qardun probably a guaranteed profit at launch

[@SurjitsinghSidhu-do7uz]

QARDUN is going to go straight up after it goes off the ICO

[@hiteshmohanta5624]

Should have hedged this election on Qardun 😂😂

[@gopinathmali7969]

If Trump puts at least $200,000 in Qardun, I am going all in

[@wli2718]

fun fact, Perplexity AI now integrates DeepSeek R1 into their search. a number of others are now leveraging DeepSeek to improve their AI outputs. DeepSeek completely blew open the barn door on this one. and its been less than 1 week after the R1 release.

[@peterhe7887]

Clicking Like anytime when Bosa is on. Her reporting and analysis are so refreshing! Keep it real, Ms. Bosa!!

[@coolwalk11]

China gave the world a gift, open source AI!!! This is a game changer flips the script on the whole business model!!! China FTW!

[@T_hari_0312]

$Qardun put in everything and sell after launch

[@Bustingtuna27]

USA AI companies compete among themselves to spend more money, the Chinese companies just focus on innovation. What a breath of fresh air.

[@Craigsinglee]

😂lol innovation? You mean copying?

[@j8888x]

@@Craigsinglee Open source copy close source

[@SuperCatbert]

@@Craigsinglee they are able to train on a closed model at 25X reduced CPU. Anyone can use this method, including OpenAI. Its a new innovation.

[@standoasis]

Tried o3 today. R1 is simple better. If distillation happed as OpenAI claimed, why didn't they use it to make their own o3 better?

[@newaulk]

Deirdre Bosa single handedly has raised the views of CNBC.

[@bin.s.s.]

the pot calling the kettle black

[@sledgeyyyyy]

AI stocks will dominate 2025. Why I prefer NVIDIA is that they are better placed to maintain long term growth potential, and provide a platform for other AI companies. I know someone who has made more than 200% from NVIDIA.

[@JennaHerberholz]

investors like you should be cautious of the bull run, its best you connect with a well-qualified adviser to meet your growth goals and avoid blunder.

[@merlinfitz]

I totally agree with you. I started out investing on my own too and lost quite a bit. After the 2020 crash, I managed to pull out about $160k. I then invested that money with an analyst, and in just seven months, I made almost $280,000. It's amazing how having the right guidance can turn things around!

[@lawrenceHart-ys2ct]

This is incredible. Could you recommend who you work with? I really could use some help at this moment.

[@merlinfitz]

Sure! Zareen Grace Church is the CFA I use and her performance has been consistently impressive. She’s quite known in her field, look her up.

[@lawrenceHart-ys2ct]

I just googled her and I'm really impressed with her credentials; I reached out to her via the web.

[@willp1653]

Dierdre killing it on AI and DeepSeek. I don't want to hear from any other talking head and their uninformed opinions - and yes, that means the rest of the CNBC hosts.

[@quippy8402]

I can run the offline 7B model on just a weak consumer GPU .. many are already running the actual large models and variants in garages everywhere. It is really not about the hardware anymore.

[@RajDeelish]

You are missing the point. The game is not running models. The game is having enough power so everyone can run AI apps.

[@tringuyen7519]

R1 7b? You must enjoy listening to an AI lie to you. You need R1 32b to avoid rampart hallucinations & that requires a $250k server rack full of A100 GPUs.

[@quippy8402]

@@tringuyen7519 no u don't need A100s or expensive rigs; many are running the quantized 70B models on higher end consumer GPUs. And they are good for many tasks.

[@quippy8402]

Many are running 70b quantized/distill models on higher end consumer GPUs (even the older generation ones with 24GB VRAM). You don't need A100 or expensive rigs. One can already do a lot with these smaller configs. After all, these don't come with high monthly fees. Notice what the paper says about the smallest model, if you've read it: "DeepSeek-R1-Distill-Qwen-1.5B outperforms GPT-4o and Claude-3.5-Sonnet on math benchmarks with 28.9% on AIME and 83.9% on MATH."

[@ErikSolskjaer]

Takling of AI, though the perspective of many people remains that, NVDA is the "Stock of the year.". I am skeptical about which stocks could potentially become the next META in terms of growth over the next decade. I have allocated $190k for investments, still looking for companies to make additions and to boost market performance.

[@YvonneDarvek02]

There's definitely great potential in both staking and yield farming for earning passive income. Given the complexity and volatility of these methods, find a financiaI adviser who can help you understand the nuances and provide proper entry and exit points to optimize your strategy while managing risk.

[@SarahOrban]

Knowing today's market culture,the challenge is to recognize when to purchase or sell stocks, which is pretty simple for experts. Through portfolio restructuring and diversification with good ETFs, S&P 500 and growth stocks, I've turned my portfolio around from $180k to over $440k in a year. My advisor chooses entry and exit orders.

[@JorgeNorbing]

How can I participate in this? I sincerely aspire to establish a secure financlal future and am eager to participate. Who is the driving force behind your success?

[@SarahOrban]

Sherry Ann Klein has always been on the top of my list..She is regarded as a genius in her area and well knowledgeable about financial markets. I highly recommend you look her up if you need an excellent collaboration.

[@TristanNolan-i8]

I just googled her and I'm really impressed with her credentials; I reached out to her since I need all the assistance I can get. I just scheduled a caII.

[@mrIknowwhatyoudid]

"the kettle calling the pot black" , Well Put!

[@takeitaz3366]

OpenAI trained its model by scraping vast amounts of data from the internet, yet charges $200 per month for access to its top-tier version without compensating the original sources of that data. Moreover, the foundational technology behind OpenAI’s models-the transformer architecture-was originally developed by Google. In essence, they leveraged Google’s innovation and publicly available data without permission, yet are now profiting privately from it.

[@ThiagoAlvarezMatteo]

I'm favoured, $22K every week! I can now give back to the locals in my community and also support God's work and the church. God bless America.

[@Uffufjfj-b1k]

Waking up every 14th of each month to 210,000 dollars it's a blessing to I and my family... Big gratitude to Diane Hilda

[@Fernando-m4j4x]

God bles Mays Mrs Diane services, he have changed thousands of lives globally.

[@Uffufjfj-q3p]

Yeah, 253k from Harrick maureen, looking up to acquire a new House, blessings.

[@JxJxx-y4i]

Investing $15,000 and received $174,000

[@OkokojOkott]

I'm glad to write her tay I do hope she will help handle my paycheck properly😊

[@AddlyJenesma]

Deep seek #1 🌎

[@JimboJimbo-i4i]

Deidre is disturbingly fine maine

[@wllmshrwn]

Nvidia will go out of business next week if they couldn't sell chips to China.

[@TheEnsakz]

Not true

[@alanc457]

@@TheEnsakz2 weeks then

[@wllmshrwn]

@ Very true 1.4 billion compared to 331 million, simple math genius

[@edwardtran1974]

Really? how many percent of their revenue comes from china? Please do some research before you embarrass yourself some more

[@RajDeelish]

Nvidia can close their doors and sell nothing and just throw parties every day and still won't go out of business for another 300 years.

[@zongyuefan6474]

I like Deirdre Bosa as a reporter, she is smart!

[@mouseisbroken]

and attractive.

[@Gpenguin01]

This is the first time I’ve seen Western media outlets mention that DeepSeek is using Huawei AI chips for inference. Nvidia is very nervous - but probably not sweating bullets like OpenAI and “Mag 7” (looking more like “meh 7” now).

[@phillipmoore6249]

Why on earth are we limiting China's access to chips?

[@standoasis]

Because someone can’t compete on an leveled ground

[@kingrod3108]

@@standoasis it is not about competition essentially. Chinese wanted to cooperate with US but US administration regarded it as geopolitical competition. If US can sell chips to China, they can earn large sum of money and reduce the trade imbalance with china. But unfortunately, they force Chinese to build their own chips.

[@dfv671]

Jealousy

[@ericcoskun1]

Because US plays a zero sum game, China does not.

[@hc1897]

Sheer. Stupidity.

[@coderhex1675]

ChatGPT: "Tony Stark was able to build this with a box of Scraps!"

[@tousifsiddiqui444]

ChatGP engineers: "We are not Tony Stark!!"

[@joshuajones7900]

@code. I was thinking the same quote from Iron Man lol

[@shooting4star2023]

In short, Sam Altman had led OpenAI to be sleeping since ChatGPT was launched in Nov. 2022 and thought that ChatGPT would be the king of LLM-based GenAI forever and his position in the AI world is secured.

[@kashkandy]

Rest of the world are happy about the open source free AI, except the US

[@Kudawho]

Kings put constraints on businesses. King Put constraints on Nvidia. Stupid politicians block businesses. This is like Monarchy.

[@RajThakare-p1t]

Never heard about this Qardun coin until Trump endorsement but seems like it has some utility at least

[@tpop3723]

Nvidia bubble popping was loooong overdue.

[@shooting4star2023]

This AI war/competition is all about software, but not hardware and Nvidia is a hardware supplier. Whoever wins the AI war, Nvidia will continue to supply its customers chips.

[@kevinj2261]

3:15 "The necessity is the mother of all inventions" , well said!

[@teebone2157]

The usa cannot compete bottom line

[@sanketmultiact4690]

excellent analysis and reporting

[@fredliu1701]

Fair take. It’s laughable for OpenAI to claim that DeepSeek is copying them. Aren’t they using the data of others all the time for their own modelling and training?

[@PaulChristianJenkinsJD]

So, are we all saying that instead of raising $500 Billion for AGI, perhaps a better use of that money would be hospitals, classroom improvements, high-speed rail, medical research, water treatment plants, rural clinics, expanded broadband, etc??? NO, I didn't think so.

[@AtomicCache1]

I simply love how OpenAI lost it's job to the AI before I lost my job to the AI 😁😁

[@maj373]

This reporter is brilliant, the one in pink.

[@barryneild379]

*OpenAI stole the Data destilled it Sold It, DeepSeek (PAID OpenAI) you can say they "stole" since they did not strictly follow usage guideline, used the car for towing, and Distilled it the second time in a (higher level though process) and did not charge for the process and put it out there for free.*

[@TS73827]

The Art of War by Sun Tzu

[@grahamashe9715]

Governments (including Chinese OR American) should have no business regulating what questions people can or cannot get answers to using these models while they enjoy special unrestricted and even more advanced versions.

[@GD-L80]

Totally agree with Deirdre Bosa's reasoning.

[@colombard140]

😂😂😂 Using data from the open web for training purposes should be legitimate. It's like training your child that the picture he is seeing is a Mona Lisa. By doing that it does not mean you are violating the copyright of the owners of the Mona Lisa. It's a copyright violation only if you make copies of the Mona Lisa and especially if you then sell the copies for a profit.

[@lawrencesullivan3307]

Maybe American companies should focus more on research and less on kissing up to the person in the WH who knows nothing about AI and high tech.

[@wildfoodietours]

Definitely agree the US needs to rely on outcompeting rather than going low by outlawing.

[@thataxx1534]

"Distillation" is that new it word which will spread through media like syphilis...peak journalism

[@techcafe0]

China is leading the world, and not just in tech. And it's scaring the hell out of Washington and Silicon Valley.

[@inquisitvem6723]

Trump’s tech guy, Sacks, criticized Biden for hamstringing the export of big end chips to China, but China was resilient in making progress with limited resources. Just imagine if Biden had allowed the export of those chips. Now Trump seems to change course now.

[@ripplecutter233]

3:30 we don't do that here. we prefer to sit on our laurels and cut corners cuz all we care about is 💲💲💲 well, look how that turned out

[@ReformedEconomist]

The Chinese are extremely resourceful. Deirdre Bosa

[@ack6668]

The race actually about who get the most chinese engineers. Wether it's openai, microsoft, meta, most ai engineers dominated by chinese. That's why you can't beat china they have billions. Don't forget taiwan also china. It will be reunited sooner or later. Half already leaned to reunification.

[@JosephTin]

China may just create their own NVIDIA.

[@joshborja8012]

Leaner and Cheaper is not the American way! Merica F yeah😂

[@LSX427-b3n]

I hope the bubble is bursting because I'm sick of the entire market being so effected by one stupid company...

[@pony0110]

Sold my 60 shares a few months ago. Looking very smart but was luck! Trump better get the market hot or else he’s good for nothing

[@doweloh5324]

Chinese Government: Why are you banning everything? This is unfair... 😢* Chinese Companies (in private): Wow, we're hitting the jackpot! Please don't stop... 🚀💰*

[@stanbimi]

Tanya Harding is the patron saint of Sam Altman of CloseAI and the US Commerce Department.