Deepseek - China’s New AI Breakthrough

[Ero]

How many have actually read the paper:

https://arxiv.org/pdf/2501.12948

[zazen]

2 hours ago, PurpleTree said:

Remember how Jack Ma disappeared for a while?

Checking excessive power that can challenge, destabilise and undermine the state. Something we are now battling with in the West with oligarchs and corporatist influence.

Theres a saying that goes: in the West capital controls the state, in China the state controls capital. But it’s more so that it directs and checks it when it gets out of hand.  Not entirely controls it like some micromanaging OCD tyrant - which is the type of central planning that causes typical communist states to fail.

Jack Ma was making the same remarks libertarian capitalists like Elon and Techno optimists like Marc Andreeson were making - criticising state regulation saying they stifle innovation, which is euphemism for “get off my back and let me consolidate power”

2 hours ago, zazen said:

There’s a subtle distinction that can help understand how China operates.

They aren’t a centrally planned state like prior communist states - they centrally direct. They don’t control businesses as much as they do direct them and then check the excesses gained from it.

Politically centralised, economically decentralised - that political centralisation allows them to check excesses - something Russia hasn’t don’t well.

In other words: China is centrally directed and actively checks the excesses of the gains made from central direction, ensuring that wealth is reinvested into national development and benefits a broader population. In contrast, Russia is centrally directed but fails to check those gains, allowing them to concentrate in oligarchic hands, leading to major inequalities and a lack of widespread development.

 

US got deepseeked (I know what you thought - you dirty) by chatCCP and is coping. The best part is that this isn’t just a gain for China but due to being open source - is a gain for the entire industry. Every player and lab in US is going to start implementing and speedrunning AI development - which is equally terrifying as it is exciting.

Techno optimists will say tech advancement will uplift us all - but if not given a parachute in the form of some compensatory UBI or safety net - then it isn’t a plot twist in humanity but more of a plot hole that tech elites will build over while we shout from the ditch.

Edited Tuesday at 02:56 PM by zazen

[Thought Art]

4 hours ago, PurpleTree said:

Remember how Jack Ma disappeared for a while?

So? Is that the same as deepseek being the Chinese govnerment? No…

[PurpleTree]

27 minutes ago, Thought Art said:

So? Is that the same as deepseek being the Chinese govnerment? No…

Well i also think you’re a bit naive.

Imo China is basically a semi fascist authoritarian one party dictatorship. The CCP holds the power ultimately.  Absolutely i think the government decides about deepseek.

[Thought Art]

@PurpleTree  Maybe, I don’t know. Jack Ma, isn’t proof. 
 

I tend not to jump straight to government conspiracy. Especially as they are a private hedge fund which is allowed in China. 
 

You can think what you want. I won’t jump to a final conclusion with 0 evidence. If that’s being naive to you. I don’t care. 

Edited Tuesday at 05:26 PM by Thought Art

[Bobby_2021]

3 hours ago, Ero said:

How many have actually read the paper:

https://arxiv.org/pdf/2501.12948

Thanks for posting! Going to take a look. 

[something_else]

14 hours ago, kray said:

Just if anyone is curious, the LLMs that these AGIs are trained on are made up of terabytes (many billions worth) of parameters, which are fed data. So technically the big data that is fed and used to train these models are actually data from humans. Training models with data from the internet requires constant up to date scraping of all the sources on the World Wide Web, which is why it’s very limited in terms of real time data. The reason I’m making this point is because while deepsink has raised the standards for efficient and reliable gen AI, it still works with the Chinese government, and the data that it trains its models on is deeply biased. Just food for thought 🤷🏽‍♂️

Not entirely. In fact, in many ways it's the opposite. Yes, the model used in the Web UI for Deepseek is heavily censored by the Chinese government, but the open source models they have released have no censoring at all. OpenAI's offerings have probably have more bias in the form of human intervention than Deepseek's open source models.

[zazen]

3 hours ago, PurpleTree said:

Well i also think you’re a bit naive.

Imo China is basically a semi fascist authoritarian one party dictatorship. The CCP holds the power ultimately.  Absolutely i think the government decides about deepseek.

If the US were to crack down on corporate or oligarchic power, should we call it tyranny? Should we let capital control the state?

 

Check out America getting adjusted to the reality of a multipolar world:

 

Edited Tuesday at 08:45 PM by zazen

[zazen]

On 27/01/2025 at 5:38 PM, zazen said:

There's a convergence of advancements happening that will completely change how we live. Generalizing here, but if we think of us humans as three components: body, mind, spirit. Map that onto the advancements currently under way in : automation (body), AI (intelligence/mind), energy (spirit). 

Automation scales human labor, AI scales human intelligence - and abundant clean energy is the input that animates the whole thing the same way spirit animates life. Human labor is expensive, our intelligence can be limited, and the energy needed to fuel civilization is either dirty, non-reliable or non transferable. This is the technological trinity - Jezus be damned. We could be entering a post-capitalist, post-scarcity world. What does money even mean? 

China's advancing in the energy department: https://www.livescience.com/planet-earth/nuclear-energy/chinas-artificial-sun-shatters-nuclear-fusion-record-by-generating-steady-loop-of-plasma-for-1-000-seconds

https://www.techno-science.net/en/news/kilometer-long-solar-power-plant-in-space-this-chinese-project-is-monumental-N26400.html - ''A solar power plant, floating 22,370 miles (36,000 km) above Earth, might seem like science fiction. Yet, China is actively working to bring this project to life, which could transform our energy supply.

This project, comparable in scale to the Three Gorges Dam, aims to harness solar energy in space, where sunlight is ten times more intense than on Earth. The goal is to produce, in one year, an amount of energy equivalent to the world's oil reserves. Indeed, the announced production would be 100 billion kilowatt-hours annually.''

The tech trinity above in full swing. First we had AI, then a new record in the domain of nuclear fusion energy, now they showing off robots for Chinese new year:

When MAGA was talking of a golden age..

Edited Tuesday at 09:05 PM by zazen

[zazen]

Perplexity already incorporating DeepSeek.

 

[Leo Gura]

The question arises, will there be a market for AI which is not top of the line? For example, if OpenAI makes a super-intelligent AI that is costly, will using something like DeepSeek still make any sense?

Is there a market for Walmart AI?

If super-intelligence exists, why use anything lower since you will get inferior results?

Edited 14 hours ago by Leo Gura

[Eskilon]

47 minutes ago, Leo Gura said:

The question arises, will there be a market for AI which is not top of the line? For example, if OpenAI makes a super-intelligent AI that is costly, will using something like DeepSeek still make any sense?

Is there a market for Walmart AI?

If super-intelligence exists, why use anything lower since you will get inferior results?

Costly to whom？ Will this super-intelligent AI be costly to the general public? Using today's AI prices as a reference, how much more will the average citizen will need to pay to get this AI? If it is really costly then Walmart AI will not go away since most people and especially 3 World country people will not be able to afford it . And what if the use of this super-intelligent AI is really niche and specified? It could be the case that if someone invent this AI initially it will be most useful in an industrial setting rather than home setting, like supercomputers and computers.

Edited 14 hours ago by Eskilon

[Leo Gura]

@Eskilon Seems to me that even the cost of super-intelligent AI should drive towards pennies over time.

I don't see why AI should be expensive. Running human brains is not expensive.

[Hyena]

Karma ?

[Bobby_2021]

On 1/28/2025 at 7:41 PM, zazen said:

Techno optimists will say tech advancement will uplift us all - but if not given a parachute in the form of some compensatory UBI or safety net - then it isn’t a plot twist in humanity but more of a plot hole that tech elites will build over while we shout from the ditch.

Imagine the kind of stuff we could accomplish if the west were not hell bend on bullying the rest of the world with whatever crumbs they have got and instead worked with them. 

The people should vote against this nonsense. 

15 minutes ago, Hyena said:

Karma ?

Everyone hates Altman at this point. Karma indeed.

[Jayson G]

6 hours ago, Leo Gura said:

The question arises, will there be a market for AI which is not top of the line? For example, if OpenAI makes a super-intelligent AI that is costly, will using something like DeepSeek still make any sense?

Is there a market for Walmart AI?

If super-intelligence exists, why use anything lower since you will get inferior results?

@Leo Gura you will be surprised, most people are essentially using walmart AI right now. I'm developing AI apps so I'm deeply immersed in this psychology. 

If the way a model is structured is ..

User <---> LLM .. notice the LLM is smart, but the user doesn't apply there intelligence, so how much value can users really extract? Not much. The LLM is only as useful depending on how deep users go, which they just don't. There's enough data that shows most users use it too shallowly. Therefore most users don't need much intelligent AI.

There's also an emerging AI Agency market, and most of these people promote using cheaper LLMs in their services so they make more money. Thats the world we live in.

[Jayson G]

"Thieves, look how these thieves stole from us" - Thieves

[Jayson G]

Whatever you think of Jordan Peterson, I don't listen to him much, but this interview, mainly with this guy Brian is one of my favorite discussions on ChatGPT. 

The biggest lesson I got from this is that any one of these LLMs are like continents, unexplored continents of value, tons of valuable insights waiting to be mined and extracted, and with the right prompts, in many cases "super prompts", highly iterated prompts, you can unlock almost any value from these LLMs. 

It's less about how intelligent these LLMs are, which is of course important, but when are we going to get more intelligent with our input, people don't realize just how deep and intelligent our inputs can be. 

[Davino]

5 hours ago, Leo Gura said:

The question arises, will there be a market for AI which is not top of the line? For example, if OpenAI makes a super-intelligent AI that is costly, will using something like DeepSeek still make any sense?

I think that AI will become specialized over the years to come. We'll have the best finance AI, the best engineering AI, the best image generator, the best video generator, the best coding Ai, the best AGI...

There'll be market for everyone. I think it is very unlikely to make a System which is of immense scope as well as immense specialization. There has to be some more revolutions on the tech field for that. So every company will find its niche, invest heavily and monopolize it. It's very hard to keep this level of tech breakthroughs continuously, as history shows with the internet and mobile phones at some point it will stabilise and each company will satisfy particular tech needs. That's my guess.

5 hours ago, Leo Gura said:

If super-intelligence exists, why use anything lower since you will get inferior results?

In that regard, the best eats the market till another one comes and does the same. It may become a very competitive market with high risk/reward.