DeepSeek, every mention
177 scenes, the whole family · ← back to DeepSeek
tap a year for its mentions
every year anyone John Coogan 186Jordi Hays 40Jordan Schneider 7Will Brown 4Tae Kim 4Jordy (Geordie) 4Dwarkesh Patel 4Bill Bishop 3Sholto Douglas 2Marc Andreessen 2
Verbatim, from the transcripts: the passages where DeepSeek comes up
How Does GROK Compare? (Full Analysis)
- ▶ 33:22 John Coogan The top OpenAI thinking models, O-one Pro get it too, but all of DeepSeek R-one, Gemini, two point O-flash thinking and Claude do not. 2 times in the scene
Elon Musk vs. Sam Altman Continues (WHO’S WINNING?)
- ▶ 3:13 unnamed speaker It's like, okay, deep seek launches. 2 times in the scene
The BEST Super Bowl Commercial (Saquon Barkley)
- ▶ 37:42 John Coogan That's of course, deep seek the free to use model was built at a fraction of the cost. 4 times in the scene
- ▶ 45:07 Jordy (Geordie) They may anthropic made a clear bet around AI safety and a lot of other firms ignored that deep seek one of them. 4 times in the scene
What is PALANTIR? (And Why it's RIPPING!)
The Truth About Chinese Manufacturing (De Minimis)
- ▶ 14:44 Jordi Hays So over the week, over the last week, it's been, if you talk about deep seek, you're gonna get a lot of, um, views.
The Future of Artificial Intelligence
- ▶ 0:00 unnamed speaker So, uh, Dylan Patel writes, the Deep Seek narrative takes the world by storm. 6 times in the scene
- ▶ 1:24 unnamed speaker Is that deep seek is so efficient that we don't need more compute. 6 times in the scene
- ▶ 4:15 unnamed speaker uh H 800 has the same compute as the H H 100 but less network bandwidth uh the misleading six million figure was widely circulated as the cost of v three but that doesn't include R and D hardware depreciation uh repeated experiments actual… 3 times in the scene
- ▶ 5:30 unnamed speaker And so in terms of model performance, DeepSeq V 3 times in the scene
- ▶ 5:32 unnamed speaker R three is the base model while R one is the reasoning model. 2 times in the scene
- ▶ 9:40 unnamed speaker So, you know, V three surpasses GPT four Oh, but GPT four
- ▶ 10:19 unnamed speaker And so let's, let's break down the direct comparison between R one and O one, uh, deep seeks model versus chat GPT. 7 times in the scene
- ▶ 10:19 unnamed speaker And so let's, let's break down the direct comparison between R one and O one, uh, deep seeks model versus chat GPT. 13 times in the scene
- ▶ 15:54 unnamed speaker And so, uh, well, there was a frenzy of hype for R one, a 2.5 trillion dollar us company released a reasoning model a month before for cheaper. 6 times in the scene
- ▶ 18:57 unnamed speaker Uh, none of this distracts from DeepSeek's remarkable achievements. 4 times in the scene
- ▶ 19:50 unnamed speaker These training pre and post V three utilizes multi-token prediction at a scale not seen before. 2 times in the scene
- ▶ 20:43 unnamed speaker People just went through the deep seek paper and everything that they were doing, they assumed was novel. 5 times in the scene
- ▶ 22:21 unnamed speaker Uh, in terms of R-one, it benefited immensely from having a robust base model, V-three. 2 times in the scene
- ▶ 27:44 unnamed speaker And so, a hundred and forty billion subsidy, AI subsidy, after meeting DeepSeek's founder. 3 times in the scene
- ▶ 29:57 unnamed speaker DeepSeq lags top labs, but closes gaps swiftly with new paradigm reasoning. 3 times in the scene
- ▶ 30:21 unnamed speaker Do something to bring this to the masses so you can cut, stem the bleeding of people installing the Deep Seek app.
What Open AI Got WRONG
- ▶ 0:04 unnamed speaker Um, for those that don't know, this is an American company that competes with, um, a Chinese company called deep seek. 5 times in the scene
- ▶ 7:58 unnamed speaker What's interesting is like, so he goes on to talk about the O-one chain of thought and how they, when they launched O-one in, in comparison to R-one, R-one tells you everything it's thinking and just dumps it out and it's open source. 5 times in the scene
- ▶ 9:23 unnamed speaker That was the, the, the only UI change that people cared about or talked about with the deep seek long launch was that exposed chain of thought.
- ▶ 14:29 unnamed speaker With DeepSeek, if I recall correctly, a lab in Berkeley read their paper and duplicated the claimed results on a small scale within a day. 4 times in the scene
- ▶ 19:13 unnamed speaker One of the big R one takeaways is that you can infuse reasoning into normal LLMs. 3 times in the scene
The Character AI Controversy (Role Play)
- ▶ 4:17 unnamed speaker We see that with the deep seek thing where like the models are very mature, very robust, certainly good enough to have a conversation with you as a romantic partner, or even if they just want to role play as George Washington, right?
- ▶ 17:10 unnamed speaker Like, people are gonna be able to fine-tune DeepSeek on this stuff very easily and create just, like, magnet links out there on the internet where you just download it, run it locally, and
DeepSeek Update, Market Crash, Timeline in Turmoil, Is VC Cooked, Zero Cope Policy
- ▶ 0:04 John Coogan We are staying on Deep Seek.
- ▶ 1:02 John Coogan This went out on January 25th and, uh, was a very, very large deep dive on, um, how DeepSeq and their R-One model might change the demand for GPUs, specifically Nvidia GPUs, and, uh, we have a summary article here, but we'll take you…
- ▶ 1:02 John Coogan This went out on January 25th and, uh, was a very, very large deep dive on, um, how DeepSeq and their R-One model might change the demand for GPUs, specifically Nvidia GPUs, and, uh, we have a summary article here, but we'll take you…
- ▶ 12:24 John Coogan He calls it COT models, reasoning models introduced in the past year, most notably in open AI's flagship O one model, but very recently in deep seeks new R one model, which we will talk about later in much more detail. 2 times in the scene
- ▶ 12:24 John Coogan He calls it COT models, reasoning models introduced in the past year, most notably in open AI's flagship O one model, but very recently in deep seeks new R one model, which we will talk about later in much more detail. 2 times in the scene
- ▶ 21:25 John Coogan But, uh, like, when I saw the reaction to R-one, uh, the DeepSeq model, I tried both of them, and I put the same prompts into O-one and, or O-one Pro and R-one, and I was getting reliably better results with O-one, with the ChatGPT… 2 times in the scene
- ▶ 21:25 John Coogan But, uh, like, when I saw the reaction to R-one, uh, the DeepSeq model, I tried both of them, and I put the same prompts into O-one and, or O-one Pro and R-one, and I was getting reliably better results with O-one, with the ChatGPT… 5 times in the scene
- ▶ 36:43 John Coogan And so, uh, the last quarter of the article talks about the seismic waves rocking the industry right now caused by DeepSeq V-III and R-I.
- ▶ 36:43 John Coogan And so, uh, the last quarter of the article talks about the seismic waves rocking the industry right now caused by DeepSeq V-III and R-I. 2 times in the scene
- ▶ 37:00 John Coogan R one represents another huge breakthrough in efficiency, both for training and inference. 5 times in the scene
- ▶ 37:04 John Coogan The deep seek R one API is currently 27%, 27 times cheaper than open AI's O one for a similar level of quality. 3 times in the scene
- ▶ 46:20 John Coogan They developed a clever system that breaks numbers into small tiles for activations and blocks for weights, so instead of just using, like, a single word for a token, they used multiple blocks, um, they also cracked, but then, this is the… 3 times in the scene
- ▶ 46:20 John Coogan They developed a clever system that breaks numbers into small tiles for activations and blocks for weights, so instead of just using, like, a single word for a token, they used multiple blocks, um, they also cracked, but then, this is the… 2 times in the scene
- ▶ 52:12 Jordi Hays And the result is that they have something like 13 individuals working on the Lama stuff who each individually earn more per year in total compensation than the combined training cost for deep seek V three models, which outperform it.
- ▶ 53:57 John Coogan So at the bottom, yes, you need a robust model and that's what their deep seek V three is.
- ▶ 54:19 John Coogan Like if, if, uh, if deep seek trained on GPT four output tokens, then all the data has already been kind of cleaned because it's only training on like, of course it's going to sound like GPT four because it doesn't have any junk in there. 10 times in the scene
- ▶ 54:48 John Coogan We know O one O three and R one all seem to be pretty good on top of the base model.
- ▶ 1:03:31 John Coogan Four, DeepSeq trained on outputs of American models, which we've discussed. 7 times in the scene
- ▶ 1:10:26 John Coogan And, uh, Wall Street kind of picked up on it today with the R-One release.
- ▶ 1:18:08 John Coogan So, uh, Dylan Patel says deep seek V three and R one discourse boils down to this.
- ▶ 1:18:08 John Coogan So, uh, Dylan Patel says deep seek V three and R one discourse boils down to this.
- ▶ 1:18:43 John Coogan Cheaper AGI will drive even more GPU demand, and the midwit says deep seek efficiency will reduce GPU demand. 3 times in the scene
- ▶ 1:27:45 John Coogan And yeah, like, let's assume this is twenty-forty-eight, like DeepSeq has twenty-forty-eight GPUs. 6 times in the scene
- ▶ 1:33:54 John Coogan And, and with this one, like, no one's saying, like, oh, it's dangerous that R-one is out there.
- ▶ 1:36:39 John Coogan Daniel says, love the deep seek app. 6 times in the scene
- ▶ 1:46:06 John Coogan So Solana says, uh, think it's probably important to adopt a zero cope policy in light of DeepSeq's achievements. 4 times in the scene
- ▶ 1:49:31 John Coogan He says, deep seek R one shows that the AI race will be very competitive and that president Trump was right to rescind the Biden executive order, which hamstrung American AI companies without asking whether China would do the same. 2 times in the scene
- ▶ 2:00:14 John Coogan Justine Moore, the venture twins over at Andreessen says, deep seek censorship is no match for the jailbreakers of Reddit. 3 times in the scene
- ▶ 2:03:51 John Coogan Uh, deep seek could be an extinction level event for venture capital firms. 3 times in the scene
DeepSeek, Stargate Update, Bees Do Enjoy Honey, Golden Retriever Maxxing, Bring on the Trilly
- ▶ 0:09 John Coogan Today, we're doing a deep dive on Deep Seek, the new AI model coming out of China. 3 times in the scene
- ▶ 6:55 John Coogan Um, but he highlights DeepSeq, which launched in May of 2024, 8 times in the scene
- ▶ 23:18 John Coogan Um, so let's move on to, uh, the, I don't know if it was an interview or just, uh, some coverage of China's new face of AI deep seek founder, Liang Wenfang.
- ▶ 24:09 John Coogan So, uh, DeepSeek is in the news again because they launched R-one, which is their reasoning model that competes with O-one. 10 times in the scene
- ▶ 24:09 John Coogan So, uh, DeepSeek is in the news again because they launched R-one, which is their reasoning model that competes with O-one.
- ▶ 24:49 John Coogan And so last December they made waves in the global AI industry after benchmark tests showed that it's DeepSeq version three LLM, which again is not the reasoning model.
- ▶ 42:41 Jordi Hays Yeah, I guarantee if there's a Chinese media, a Chinese media company, and they're like, hey, we're kind of annoyed that DeepSeek is just like harvesting all our data. 9 times in the scene
- ▶ 43:56 John Coogan Give you some, some, some up-to-date information about the R-One launch and how it's been received. 6 times in the scene
- ▶ 44:06 John Coogan Uh, Mark says, DeepSeek R-One is one of the most amazing and impressive breakthroughs I've ever seen. 18 times in the scene
- ▶ 59:28 Jordi Hays He says, I've made over 200,000 requests to the DeepSeq API in the last few hours. 10 times in the scene
- ▶ 1:16:58 John Coogan Well, that about wraps up our deep dive on Deep Seek.
- ▶ 1:38:32 John Coogan He says, uh, Deep Seek R-One has an existential crisis.
- ▶ 1:39:13 John Coogan We'll get more in the coming days of people playing deep seek.
- ▶ 2:17:22 John Coogan I, I saw people got Cursor working with, uh, with DeepSeq under the hood.
Project Stargate, The T Word, Ramp Treasury, Audience Ad Reads, See You at Davos
- ▶ 39:12 John Coogan Um, if they can get away with running a highly efficient distilled model, like DeepSeq demonstrated, this is what I was just mentioning, like DeepSeq distilled the model down, so it's higher margin with a lower price to serve, then that's… 2 times in the scene
The Businesses Making $$$ off Disaster Relief
- ▶ 2:09:54 Jordi Hays Yeah, deep, deep sink, I mean. 4 times in the scene
The END of Ski Season? (E26)
- ▶ 2:16:48 John Coogan He has a long post here about Deep Seek, the Chinese artificial intelligence company, making it look easy today with their open weights release of a frontier grade LLM tuned on a joke of a budget. 3 times in the scene
- ▶ 2:17:14 John Coogan While DeepSeqs VIII looks to be a stronger model at only 2.8 million GPU hours, a hundred, uh, 10 X less compute or 11 X less compute.
← previous page 2 of 2 · 100 scenes per page · newest episode first