Girls Tech Talk Epsiode 2 - Will DeepSeek surpass ChatGPT?
In our second podcast, we discuss the threat of DeepSeek to OpenAI's ChatGPT
2/12/2025
1. Why is DeepSeek Creating Such a Buzz Now?
In our previous "GIRLS TECH TALK #1" (Watch on YouTube), we discussed the topic of whether AI will take away engineers' jobs.
This time, we're diving deeper into "DeepSeek," exploring its impact on businesses and the international community.
DeepSeek is an AI known for achieving ChatGPT-like LLM (Large Language Model) capabilities at a lower cost. Developed by a Chinese company, it's generating discussion worldwide.
GIRLS TECH TALK #2: DeepSeek
2. What is DeepSeek? The Low-Cost, High-Performance Approach
DeepSeek is a China-originated LLM that offers ChatGPT-level natural language response capabilities while being developable and operable at a lower cost without expensive GPUs.
- Implementable on a Small Budget: Makes AI adoption accessible not just for large corporations but also for small and medium-sized businesses
- ChatGPT-Level Response Capabilities: Returns rich insights to questions through extensive text learning
3. Intensifying International Competition: US-China AI Battle and Deepening Discussions
-
NVIDIA Chip Export Restrictions
- While the US restricts the export of high-performance NVIDIA GPUs to China, China developed DeepSeek using its own technology
- Mashable, The Japan Times and others report on the "US-China technology hegemony battle"
-
Open Source Impact
- DeepSeek has been open-sourced, making it available to many developers
- Reddit and Forbes raise concerns about information censorship and restrictions by Chinese AI
4. Business Perspective: Cost Reduction & New Market Creation
The "affordable yet reasonably high-performance" aspect is shaking up the existing AI market.
-
Lower Implementation Barriers
- While enormous operational costs were previously a bottleneck, they're significantly reduced with DeepSeek
- Technical details are discussed in UncoverAlpha (technical explanation) and arXiv (advanced paper)
-
Impact on Existing Players
- Could pose a threat to OpenAI and Google, who have made substantial investments
- Some suggest US AI dominance might be wavering, intensifying competition
5. Summary: The New AI Era of Mixed Opportunities and Risks
DeepSeek brings two aspects: wider AI adoption through cost reduction and concerns about Chinese companies' AI technology.
When considering implementation, companies need to consider not just cost benefits but also information security and risk management in operations.
In the next GIRLS TECH TALK, we'll discuss employee dissatisfaction at Meta (formerly Facebook). Stay tuned!
Reference Links
-
Public Media:
-
Criticism & Censorship:
-
Tech Explanation:
-
First Episode (GIRLS TECH TALK #1):
YouTube Transcript (Translated from Japanese)
0:04: Welcome to Manft Radio. In this radio, we have female engineers working at tech companies and
0:10: Designers will explain trending technology and AI news
0:15: I'm Mari, a designer, and I'm Sawako, who is involved in development
0:21: Yes, so, thank you for your hard work. Thank you for your hard work.
0:31: Last time, we talked about learning, but now it's every week.
0:40: I have a lot to talk about, so I think it's okay, but yes, every week.
0:48: Thank you for your help. Thank you for your help. Yes, how are you doing lately?
0:56: Recently, today is the first snow, isn't it? The second time.
1:02: The second snow this year is really heavy. Yes, there were big balls of snow falling.
1:09: Oh, is that all over Japan? Kansai, oh, quite a bit in Kansai
1:15: I think it was raining everywhere, oh, yes, yes, Hokkaido is really cold now, oh, it's been snowing since last week
1:21: Yes, yes, yes, yes, it's cold, right, right, it's cold, but it's nice, I'm going to draw
1:29: Are you getting cold, oh, it's not cold, it's just raining a little today, so it probably won't pile up
1:36: I think so, but it's already February, isn't it, but I think February is the coldest month, right? Right now
1:45: It's getting colder as usual, but I have hay fever
1:51: I think it's just right, right, it's about time it starts, right, right, I have severe hay fever
1:58: So you said, right, right, when we met, right, right, it's already February
2:06: I wonder if it came from there. Yes, yes, I was in the middle of February last year.
2:11: It was around that time, and this year it was the second half. Is it still this year? I think it's the second half. It's just starting to get better.
2:18: Yes, that's right, and recently it's been a hot topic, and now you see deep seeks everywhere you go.
2:24: I guess so. Wherever you go, wherever you watch TV, you see them almost every day.
2:31: The news is amazing, isn't it? It was around when Chat GPT came out.
2:37: The response may have been even more. At that time, AI was really like AI.
2:42: I think there were probably a lot of people, but now the awareness of AI is increasing, so I think so.
2:49: Yes, surprisingly, more people may know about it. So today, I would like to talk about this.
2:57: I would like to talk about the current topic of deep SE. Yes, thank you.
3:02: Please. First of all, that deep SE. When did you explain what seek is?
3:08: I would like to ask Professor Sawa to explain it. Yes, that professor has a lot of knowledge.
3:14: I don't have much knowledge, but please excuse my nasal voice today.
3:21: I caught a cold last week, so I talked about DEEPseek last week.
3:27: But Chat GPT, Gymna, and other things like that are called LLM.
3:35: It's just one part of it, and LLM refers to a large-scale language model.
3:42: To explain briefly, it's a model that can learn from a large amount of text data, and it can answer questions asked by humans from what it has learned.
3:59: It refers to the function of AI, and DEEPseek is also a GPT.
4:06: So, an app like this where users can ask questions and get answers back
4:13: Well, they've developed an app, and I think it's the top download on Apple right now
4:21: Yes, it's a very popular app
4:26: First of all, thank you very much, this is the very popular Deep Seek
4:32: But first of all, why is there so much fuss about it?
4:38: I want to ask, why is there so much fuss about it? Why is it so
4:43: The biggest reason why it's so fussed about is because of cost reduction
4:51: But as I said earlier, chat GPT, Gemini, tape
4:58: seek all have the same function, right? Users ask questions and get answers back
5:03: That's the kind of function it has. Chat GPT and its gems are
5:09: It was made with a lot of money and is still being developed, but this Deep
5:16: Seek is so cheap, I wonder how much it was, 700 million?
5:25: It's said to be 9.9 billion yen in Japanese yen, or 6 million yen, and if you convert it to 600 million dollars, it's dollars
5:32: It's 600 million dollars, but the yen is weak right now, so it was made for about 900 million yen, and chat JPT
5:39: It's already made for about twice that, maybe more than twice that
5:46: It's said to be so expensive, and maintenance and stuff like that requires a big data center
5:51: It needs a lot of power, and that GPU and stuff, and that's what they say Chat
5:58: GPT was already the standard, but this Deep Seek has broken that wall It's a very cheap price and
6:08: It's not a bit of an outdated technology, but
6:15: It was achieved using a lower standard machine than what you would normally think
6:23: So it's become a big topic, isn't it? I see, but I've heard
6:28: And if it's cheaper than the chit GP and the results are good, that's great
6:34: It sounds like a lot, but why is it so rough? There are so many controversies
6:40: It's not that, but it's rough, isn't it? It ... If it's cheap and the results are good, then it's fine.
7:05: It's a good thing, but why is there such a debate?
7:12: Well, China and America have been fighting for a long time about various things, such as technology.
7:18: So, they've been fighting a lot, but this time, it's also about AI, and they're fighting over who can make the best thing.
7:23: Well, America is an American company.
7:30: It was mentioned last week, I think it was mentioned last week, NVIDIA is a company that makes this graphic GPU chip.
7:37: But that company's chips are really expensive, but
7:44: By using them, you can process the data, and if you don't have the chips to process the data, you can't process it quickly.
7:52: You know that, America is already in China.
8:00: They don't want to be overtaken in AI technology, so they're taking a stance like fortune-telling.
8:06: Yes, yes, China is China, and we want it, but we can't get it, we can't buy it, so we'll just make it ourselves.
8:14: It's like we have to do something, but we really want it to be sold, and there's a conversation like that.
8:20: There's a lot going on, and that's the first debate, and the second one is
8:26: The cost of Chat GPT is very high for this small company,
8:32: It's huge, but Deep Seek provides its service at a very low cost
8:39: Now that it's possible, what will they do with the US stock market?
8:46: What will the CEO of Chat GPT do?
8:52: That debate has spread again, and the third one is China's
9:00: Deep Seek was created by a Chinese company, and Deep Seek is open source
9:08: I think it's a little difficult to understand what open source is unless you're in a technical job
9:13: I think so, but yes, well, well, you know, you have meetings at the company, right?
9:20: Yes, you have meetings, right? You have a lot of meetings, right? There's the president and the bigwigs in that meeting
9:28: There's a group, and then there's various assistants and employees, and everyone talks like that
9:35: Isn't that right? Yes, if that's a normal meeting, then I wonder if it's not like that in Japan
9:41: Meetings in front of the water server, in the toilet, or at the water server
9:47: They say they do it, it's tiny, what do you call tiny? It's not a small meeting
9:54: But yes, yes, there are, yes, there are, yes, there are, like that kind of meeting
9:59: It's like people from all departments gather together and talk about the same
10:05: It's like talking about fun topics, or the contents of work, and saying, "This idea is good," and creating a place where you can easily talk like this
10:14: There's something called an open source project, and that's A lot of companies like software companies
10:23: Ordinary people, or people who want to make something like this, can make it by themselves, but
10:28: When they can't make it, they make it open source and people from all over the world, all countries, all levels of engineers can join
10:35: But anyway, let's all join and work on this project together, like that
10:41: All of the projects that we have are open source projects or open source
10:46: It's called software, but it's a meeting that anyone can join, so anyone can really join
10:51: No, you write the code, and if you give an idea, like "What do you think of this code?", if someone says it's good, you test it as it is
10:58: It's really used and used by people, so it's like that, from anywhere
11:03: It's a form of software that people can participate in, and it's made into a mobile phone, and that's probably the biggest thing
11:12: I think it's probably the biggest thing I see, so even if you're not actually working at GPC, you can use it normally.
11:20: So that means that ordinary people can see those actions, or use them? Yes, I see, so that's it.
11:27: One more thing, well, in addition to the three points I just mentioned, I'll just say it.
11:35: Maybe those small companies can use it because the costs have gone down, and they can beat big companies like Google's gym chat or Microsoft.
11:51: It's very likely that they can develop software and develop AI software.
11:58: It's a hot topic. Ah, I see, yes, I wonder why GPT came out.
12:04: Everyone can use it, but it was like they could only make really open AI.
12:10: I think so, but GPC is like that. It's like the hurdle has been lowered.
12:15: No, I think it's been lowered a lot. Ah, I see. Yes, that chat.
12:20: The point is that you can dig deeper into development than just making pots or something like that.
12:27: I'm just curious.
12:33: Yes, then, if we work hard, our company can also
12:39: Can you make something like this? For example, can you make something like DeepSeek using open source?
12:46: Of course I think you can. Yes, it expands your dreams. We're still doing things like this.
12:52: Chatbots like GPT are being used in our company now.
12:58: Well, just in many ways.
13:04: It makes it easier to work as an engineer. Or rather, this is a black box.
13:09: I'm saying that among engineers, when you say black box, you mean that you develop things that you can't see with a certain assumption.
13:17: It's like that black box has become a little more of a skeleton.
13:22: I see, so it's like this here, so I wonder if I can do it like this.
13:29: So you feel like you can understand it? Yes, yeah, yeah, yeah, it's amazing. It's really amazing when you hear it like that.
13:36: I think it's a big step forward. But why is it that Chat GPT and that AI hide that? Ah, how?
13:43: This deep technology hides the fact that it's being made with these technologies.
13:49: Seeker said all of it is like this, so that's right. Because it's normal, but you want to hide it, you know, like your own company.
13:56: It's about making things, yeah, yeah, yeah, that's right. Yes, but this time, I'm going to put it out in the open.
14:03: So everyone please join us, let's make it even better.
14:10: I think it's a great thing, so I think it's a great dream.
14:16: Isn't it spreading? Yes, that's good. Well, let's start with Mako's DEEP
14:22: seekI'd like to introduce two articles about the hot topic of deep seek.
14:28: Yes, let me introduce the first article.
14:34: Ma.com, an article about Chinese AI company DEEPseek and American Open
14:42: This article begins with the idea that AI can be used by both parties.
14:48: DEEPseek is an open source, highly customizable, low-cost AI
14:53: It provides models and allows the market for chit GPT. Until now, American companies such as Open AI, NVIDIA, Google, and Microsoft have always been considered leaders in the AI industry. However, with the emergence of DEEPseek, this assumption is being overturned. The impact of DEEPseek is already being felt in history.
15:18: The DEEPseek app was on the top charts of the Apple Store and Google Play in just one day.
15:25: Furthermore, the appearance of DEEPseek has a huge impact on NVIDIA.
15:31: DEEPseek uses an effective AI model that does not require expensive NVIDIA chips.
15:37: As a result, NVIDIA's stock price plummeted, and CEO Jin Sung Han-sang's assets were estimated at $21 billion in one day, a decrease of about 3 trillion yen.
15:50: Meanwhile, the question is how OpenAI and ChatGPT will compete in the future.
15:56: ChatGPT already has a strong brand and is used by many users.
16:02: However, if a free competitor appears, it is likely that users will abandon it, especially.
16:09: The open source model is not suitable for business use. If it is optimized and available at low cost,
16:15: GPT's advantage will be shaken.
16:22: If it cannot adapt to market changes like AOL did,
16:28: DEEPseek's arrival is not just
16:34: It has the potential to redraw the power map of the entire AI industry, not just the announcement of a new AI model.
16:40: How major American companies such as NVIDIA and AuAI will respond to this will be the focus in the future.
16:49: It's an article, so I'd like to ask Sako-san to introduce it.
16:54: I've talked quite a bit about the content, so I'd like to use only the main points.
17:02: Well, I've read a few, but I'd like to introduce one, Gama Japan.
17:08: The article in the Japan Times is about why the world of AI is now
17:17: It's not the world of AI, it's the world of AI. Why is this world of AI, this world of AI, this world of AI, this world of AI?
17:25: The content of the article was about why China's DeepSeek is shaking everyone up, like, what should we do?
17:32: Of course, as I said earlier, the cost is also easy for everyone to use, and it's open source, and there's also open AI and meta
17:47: What's different from that model? Of course, that
17:53: It was developed in-house, how should I put it, developed efficiently, and it's definitely
18:00: It was thought that without chips, data processing would not be possible quickly, but these people
18:07: The fact that DeepSeek's company achieved it without it is a huge
18:13: So, why is this a big deal in America?
18:19: The reason is that China and the US have been fighting over AI for a long time.
18:28: It's a debate over which country can train and create a better AI first.
18:35: It seems that China is overtaking them now.
18:42: So, this article was written about that point.
18:49: And, why, and, why, and, and,
18:54: If you read the Japan Times, it's really interesting.
19:01: The person who founded DEEPSE, yeah, yeah.
19:06: Well, I think he founded this company with 1.4 million US dollars, and Deepseek
19:17: I think he made 99 million or 96 million. A total of about 9 million
19:24: It's amazing that he's accomplished this, and also, he doesn't seem to be on social networks or anything like that
19:30: It seems like he hasn't been mentioned much, so there aren't many photos of him, like this OpenAI automaton
19:37: He's not really being talked about, but now it's like, who is he?
19:43: That's right, that's what people are saying. I see his logo, but it's true in chat
19:48: When GPT was released, the founder, Sam, was often mentioned, wasn't he? All over the place, right?
19:54: That's right, but he hasn't been mentioned much yet, but from the people around him, it seems like they're wondering what he's like, what kind of person he is, and that's also a bit interesting, and finally Mako's deep seek
20:09: Noma 1 I'm not being negative, but
20:15: Both Chaopen AI and DEEPseek are like that, but the country is behind them, so they don't say anything that might make that country sound bad,
20:30: They don't answer, or they try to deflect the question, or something like that,
20:35: There's a tendency for that to happen, but it's the same for everyone,
20:41: This is a point that's been raised in the Japan Times, so I thought that was interesting, too.
20:48: Please read the Japan Times, thank you
20:54: Based on what we've talked about, I'd like to ask Sawako, the engineer, a few questions
21:01: I've been talking about DEEPseek so far, but what is DEEPseek and originally
21:08: Is there a technical difference between ChatGPT and ChatGP? At our company, we
21:16: We tried some verification, but DEEPseek and ChatGP
21:21: There is a technical difference. We made a chatbot with ChatJPT
21:28: But we verified whether the same results would come out with DeepSeek
21:34: Well, we had the developers at Nameka do it, and the result was
21:41: ChatGPT is not a response result, but before the result, we have to tell you what we did
21:46: Well, what did we do? Sorry, what did we do? Are you open on Sundays?
21:52: Are you taking care of patients with colds? Is there parking? Those are the kinds of phrases that general users use, right?
21:59: And the code, when you write the code, I use chat GPT, etc.
22:05: I don't use all of them, but I'm looking into what methods are available
22:11: It's very useful for doing things, so I'm using it, and after putting all of that together, I found that DeepSeek
22:21: The response time is a little long, but there are many detailed and accurate answers
22:29: I see, so that's what it is, and the code generation is particularly impressive, especially in terms of performance
22:35: The results showed that it was good, and the content is rich, so it's DEEP
22:41: As for seek, I think there's another blog that came up later, so if you'd like, you can take a look at it
22:47: Well, when I was talking later, I realized that it was made by a Chinese company
22:53: It's a Large Language model
22:59: It seems that the response is quicker. I see.
23:05: Also, the good thing about Chat GPT is that it responds quickly and gives concise answers, and it is very stable, so it is a very suitable option for simple interactions.
23:19: The results are in, but in terms of system stability, Deepseek is
23:27: It seems to be a little inferior to Chat GP, especially now.
23:33: It seems that hackers are messing around with Deepseek, so there are a lot of errors, access, and service interruptions.
23:42: And the processing speed is a little slower than Chat GP.
23:47: The results are that it seems to be compared to GPT, and there are negative aspects like that.
23:55: But in terms of cost, Deepseek is the clear winner.
24:02: It's so cheap to use, so it's great for small and medium-sized businesses.
24:09: I think it's a very useful software for companies. I see, the cost of using it for business purposes.
24:17: That's right, yes, it's open for business purposes.
24:22: Since it's source code, the advantage is that anyone can use it. I see, that's right.
24:30: The cost you're talking about is actually chat GPT.
24:37: But you're talking about the cost of the API. The API is the access to access the data to the llm.
24:44: If you access the server for free, the company will go up, right?
24:51: That's the cost of that part. For example, if you chat gbt 10 times.
24:56: If it costs 100 yen to access, you can do deku for 10 yen, so it's really cheap.
25:03: As expected, it is very cheap. Thank you. Well, yes. Yes.
25:09: Yes, the second question is, basically, the content of what we have talked about so far is that
25:17: America and China are competing in AI, right? Which one is it? America
25:23: China has monopolized it up until now, but China has made it easier, or rather, China has entered the market all at once.
25:28: Yes, this competition is intensifying.
25:34: By the way, I really feel that there is something Japan can do now.
25:41: But, for example, if Japan, what about engineers who are good at AI?
25:47: And if the government could provide more support to companies for such people, I think the government should provide more support to companies and improve AI. If we were to get serious about this,
25:58: Is it possible for Japan to join this so-called AI war?
26:06: Of course, I don't think it's impossible. The reason is that it's open source.
26:13: It's become clear to everyone that it's structured like this, so we're not just copying it,
26:23: We're looking at it, incorporating it, and developing new things based on that.
26:30: I think it's amazing that Japan can do things like that, and I think it's great that Japan is great in various fields.
26:38: When I watch it, I think that Japan is quite good at improving things, whether it's cooking or technology.
26:44: They hone their skills, so they're good at making things better.
26:51: There was a story that a great social studies teacher said, when the surrounding countries were at war, Japan was breeding goldfish.
26:58: What do you call it? It seems like there are a lot of different kinds of goldfish.
27:05: They made different kinds of goldfish, made different kinds of beautiful goldfish, made love, and they competed to see who could make the most beautiful goldfish.
27:11: I think that's the good thing about being an island nation, or maybe it's the internet.
27:17: It's connected, so even though it's an island nation, we live in an age where a lot of information comes in, but I think that kind of old-fashioned thing
27:25: I think that Soutake is really good at bringing amazing technology to the top.
27:33: I would love to see that. The kinds of dressings are amazing, too.
27:38: It's crazy, isn't it? I thought a little bit when I went to the supermarket, there were so many
27:45: There are things like that, right? It's true, it's true. American ketchup is also amazing, but the number of dressings is incredible.
27:52: That's right, isn't it like this? There's Japanese style, and of course there's Italian and French, but it's like something in between. I understand.
27:58: I understand, I think that's a really good analogy. It's really amazing, isn't it?
28:05: How should I put it? They're good at polishing things. I mean, it's true, it's true, but it's completely different.
28:10: Even if they don't make original things, they can still use this technology in China.
28:17: They can make American things even better, or rather, there are technologies that can make things better from here.
28:22: That's what I'm saying, so I think there are. I think there are, right? But, just now, Japanese people.
28:27: There are fewer of them, so I think we should work on that by getting more people who can read English.
28:34: Or by increasing the number of people who can speak English, and spreading the skills of engineers to more people.
28:40: I think we can do new technology if we do that. Yes, that's right.
28:45: Thank you. Now, my third question is, AI is quite expensive.
28:52: I think that in the first place, if you want to do it in the general chat business, you have to pay thousands of pages a month.
29:00: I think the cost will continue to rise, but if the cost is low, AI will
29:08: How much will it penetrate our world, or the human world?
29:15: I feel like the day is coming when AI will gradually become a part of our lives.
29:23: Do you think so? Yes, maybe, but I happened to see a YouTube video recently.
29:29: There is a talk show called American News, where former news anchors talk about various things.
29:36: There is a talk show with Tacca Carlson and Piers Morgenstern. There are people who say that, but the awards they receive are
29:43: I see this every now and then, but one of them said that he continues to reject AI
29:52: He's a journalist, so he's not really involved in technology or anything like that, but maybe people who aren't interested in technology in general
30:03: I think that for people who aren't interested in technology in general, the image of AI as something a little scary is still strong
30:10: I feel that way, so no matter how cheap the cost is, the engineers are good at this
30:18: If they don't hide their suspicions that they're using AI, it might be that it will take a while for it to take hold
30:24: I was looking at the features, not the news, and I felt that robots look scary, to be honest
30:31: That's right, no matter how cute they dress Even robots are cold.
30:38: Human emotions get in the way and they don't really penetrate.
30:44: It might be possible to penetrate with software or apps, but
30:49: It's really good to make them into robots and make them really cute.
30:56: Furby, do you know Furby? I was scared of Furby.
31:01: But it's scary. I was scared of Furby too, but it seems like they made Furby a little smaller and cuter.
31:07: If they made it a little smaller, it might be accepted as a child's toy, but I think it's difficult to make it a little smaller.
31:16: I think the impression of taking up more and more things is very strong, so it's difficult to make it a little at a time.
31:23: It spreads, but it doesn't spread all at once. I think so.
31:29: I see, so it's still going to take a while, I guess. Yes, this is just my personal opinion, but I see, thank you.
31:35: Yes, thank you for that. Thank you very much. Thank you very much. Today is already
31:42: I'm glad I was able to hear a lot about deep seeking. It's a fun topic. Now, the center is fun.
31:49: Yes, well, it's a fun cutting-edge topic that will change all at once next week.
31:58: That's right, the Facebook meta is that the employees are angry at the Facebook meta, which is completely different.
32:04: It's a different story, but that's right. I actually
32:11: I've been to the Facebook meta office, and I see, that's right. It's a very free place.
32:19: I have friends who work there, and I'm not sure.
32:24: I feel like there are always a lot of problems coming up, but the topic of self-study is very interesting.
32:31: I thought it was deep, yes, very well. It's quite shocking, isn't it? Now, next week, this time.
32:38: Hamacho, that was really serious. But next week, the Facebook instructor is angry, so are they angry?
32:45: I would like to talk about that. Yes, this is a teaser.
32:53: Well, thank you very much for today. Thank you very much. See you next week.
32:59: Let's do it. Yes, thank you. Thank you. Girls talk.
33:04: It was a radio show. Thank you very much.
