The Effect of Deepseek on Australian Stocks

[Garpal Gumnut]

Deepseek is a Chinese AI developer which has stolen a mark on the US AI industry by developing a program similar to ChatGPT for a fraction of the cost. Up to now the future of AI involved expensive semiconductor chips and data centres due to the complexity and need for energy on the one hand and heat dissipation on the other.

Already Deepseek has caused a major fall in US IT stocks. Our IT stocks may be affected but more importantly our material sector may as well. It supplies Copper, Uranium and rare metals which are essential for data centre development.

Thus far the effects of Deepseek are not quantifiable but it needs to be a consideration in choosing investment in stocks on the ASX.

gg

 

[Dona Ferentes]

content => AI, AUS, missing stocks

The case for Australian AI​

• 29 January 2021

Australia needs a sovereign artificial intelligence (AI) capability. It must be developed in Australia and built on Australian data. It must be AI for Australian questions and Australian problems. It needs to embody Australia’s values, geography, and economy. Downloading a foreign model and fine-tuning it undermines our economic future, because it doesn’t build Australian capability. If Australia is to control its own destiny in an AI-enabled future, it must build its own infrastructure, not rent it from overseas. Creating an Australian AI capability is the first critical step in the long process of building Australia’s AI economy. Having Australian capability will develop exactly the skills, experience, and capability in AI that Australia needs to drive its transition to an AI-enabled economy and set us up to build a better one.

Background​

AI is the technology of our time. It has changed the global economy permanently, yet its primary impact is yet to come. Businesses that engage in the transformation will improve their productivity and out-compete those that don’t. The larger opportunity that AI offers, however, is to develop entirely new business models.

Various economic reports put the potential value of AI to the Australian economy over the next decade at more than $300 billion. AI is not an emerging technology, or about to descend through the downside of the hype cycle. It is creating far too much economic value right now for that.

Uber, Google, Facebook and TikTok used AI to build global business models that have changed the Australian economy permanently. AI-enabled global businesses will continue to outcompete existing industries over the coming decades. The Australian tax base will shrink, and Australian productivity will continue to decline, unless we compete. This comes at a time when our economic complexity is shrinking, and our population is ageing. We need Australian businesses that use AI to address new global markets if we are to maintain our GDP per capita, let alone grow it.

Large Language Models like ChatGPT are a critical tool for existing companies and startups that want to develop AI-enabled business models. As a result, they have become critical infrastructure for nations wanting to make the transition to the AI-enabled economy. Australia needs AI that reflects its culture, data, and values if it wants to retain economic and cultural sovereignty. The countries we compare ourselves against have already made this step.

AI and global markets​

The five largest companies in the world are AI companies. The revenue of the smallest of the five (Amazon) would see it placed at number 25 in the list of nations ranked by GDP. This puts it above 152 countries including Ireland (population 5 million), Norway (population 5 million) and Austria (population 9 million). Amazon has 1.2 million employees, slightly smaller than the population of Adelaide. AI is driving unprecedented value creation globally and will continue to do so.

The founders of Google didn’t inherit a small Internet search engine and make incremental improvements. Larry Page and Sergey Brin were doing PhDs at Stanford and realised that Internet search could be framed as a matrix inversion problem. They started an Internet search engine on this basis, and it was better than its competitors. This meant they attracted more traffic, which gave them more data, which allowed them to improve their algorithm further. Before long they had an insurmountable advantage in search, which they leveraged into online advertising. It is critical to their model that each additional search customer, and each additional advertisement, have almost zero marginal cost to the business. The accuracy and scalability of the model undermined the viability of classified advertising globally, and thus the business model of most newspapers. It had a similar impact on television.

The founders of Uber realised that there was a misalignment of interests between taxi drivers and passengers, and a lot of unused capacity in privately-owned vehicles. They used AI to enable drivers and passengers to connect and undermined the taxi industry’s business model as a result. The taxi drivers were protected by legislation, and by the physically and geographically focussed nature of their business. The value of a taxi licence is now less than 10% of what it was pre-Uber. Uber make 30% of every transaction and have almost zero marginal cost.

The challenge for many incumbent businesses is whether they want to be Uber, or Uber drivers.

Value proposition​

The value in AI is not chatbots. The value is in the fact that AI enables existing business problems to be solved with far less training data, and far more quickly, than has been possible previously. It has thus removed the moat that many existing businesses depend upon. The truly disruptive value in AI is that it enables solving new problems that would have been considered impossible previously. Some of these new solutions enable global businesses. This is a great opportunity for Australia to transition to a more complex, productive, and modern economy.

Australians need to be able to use AI without sending data to foreign countries or companies, and without leaking IP. More than this, building Australia’s sovereign AI capability is the first step towards joining the modern global AI-enabled economy. If Australia does not develop its own capability, it will perpetually need to download this critical infrastructure from overseas. In developing our own infrastructure, we build the skills and experience required to create AI-enabled global businesses in Australia.

Professor Anton van den Hengel is the founding Director of The Australian Institute for Machine Learning, a Chief Investigator of the Australian Centre of Excellence in Robotic Vision, and a Professor of Computer Science at the University of Adelaide

 

[Craton]

Excellent info @Dona Ferentes

On a IT related forum, I posted that Quantum Computing and AI, known as Quantum AI will probably be the "next big thing".

Forbes 8 Oct 2024 article and reads in part:

The Road Ahead: Preparing For A Quantum Future​

As we stand on the brink of this Quantum AI revolution, it's crucial for individuals, businesses, and policymakers to start preparing. While fully realized Quantum AI systems are still years away, the groundwork is being laid now. Major tech companies and startups alike are investing heavily in this field, and governments around the world are launching initiatives to support quantum research and development.

For businesses, now is the time to start thinking about how Quantum AI might impact your industry and to consider how you can position yourself to take advantage of this technology when it matures. For individuals, particularly those in tech-related fields, gaining an understanding of quantum computing principles could be a valuable career move.

For policymakers, the challenge will be to create regulatory frameworks that encourage innovation while also addressing the ethical and societal implications of this powerful technology. This will require a delicate balance and close collaboration between government, industry, and academia.

As we move into this quantum future, one thing is clear: Quantum AI has the potential to be one of the most transformative technologies of our time (my bolds). It promises to unlock new realms of scientific discovery, drive unprecedented economic growth, and help us tackle some of humanity's greatest challenges.

Re. DeepSeek also from Forbes 28 Jan 2025
I've posted the full article without quotes because reading this, should be of great concern to each and every user!

Seek Deeper On DeepSeek For Artificial Integrity Over Intelligence​

What DeepSeek R1’s privacy policy reveals about its AI systems deserves close attention and the utmost caution in its use. It’s not about technological prowess or outperforming OpenAI’s o1 or others on benchmarks related to mathematics, coding, or general knowledge — topics that are already widely discussed.

It’s about its capability to exhibit artificial integrity over intelligence.

Issues With DeepSeek’s Inner Mechanisms​

First, the lack of DeepSeek’s internal mechanisms makes it a greater model for user exploitation rather than user empowerment.

DeepSeek’s privacy policy outlines the types of data it collects but fails to clarify how this data is processed internally. User inputs like chat history and uploaded files are collected to "train and improve services," yet no mention is made of anonymization or safeguards for sensitive data.
There’s no clear documentation on whether user data is directly used to update AI models either. Terms like "hashed emails" and "mobile identifiers" obscure meaningful transparency, leaving users uncertain about the implications of their data being collected.

Overall, DeepSeek collects extensive data (e.g., keystroke patterns, device IDs) but does not justify how these granular details are necessary for providing its service. However, it takes care to indicate that it is retaining user data "as long as necessary," but without specific retention periods or guarantees exposing user data to prolonged vulnerabilities, including misuse, breaches, or unauthorized access.

Its reliance on tracking mechanisms (such as cookies) demonstrates a fundamental trade-off: Users can "disable cookies," but the policy warns that doing so limits functionality, subtly coercing users into sharing data for basic service use. Moreover, by tying essential functions like logins or account continuity to data collection practices, DeepSeek blurs the line between informed consent and forced compliance.

Surprisingly, its policy does not mention any mechanism to prevent biases in how the system processes user inputs or generates responses, while there’s no mention of explainability in how AI outputs are generated, leaving users in the dark about the logic behind decisions or recommendations.

And last, by relying on internal reviews of user inputs to enforce "terms of service," DeepSeek places the onus of ethical behavior on the users, not the system itself.

Issues With DeepSeek’s Outer Mechanisms​

Second, DeepSeek’s promises of innovation should not justify its lapses on critical external matters threatening societal structures.

DeepSeek stores personal information on servers located in the People’s Republic of China, and its privacy policy acknowledges cross-border data transfers.

While it mentions legal compliance, there is no explicit mention of compliance with major global privacy frameworks like Europe’s General Data Protection Regulation or the California Consumer Privacy Act, raising concerns about the legal treatment of user data from jurisdictions with stringent data protections.

Given the regulatory environment in China, where data localization and governmental access are significant concerns, storing sensitive personal data on Chinese servers introduces potential geopolitical vulnerabilities as users from regions with strict data protection laws may unknowingly subject themselves to less protective data regimes, undermining their privacy rights.

DeepSeek openly admits to sharing user data with advertising and analytics partners to monetize its platform, enabling them to target users based on granular data, including activities outside the platform.

And typical of such a model, there is little (if any) transparency about how users are compensated—or even informed. Not to mention that the data collected can be used to perpetuate existing inequalities, such as targeting vulnerable populations with manipulative advertising. Indeed, as algorithms shape what users see and consume, they indirectly influence societal behaviors, values, and trends, often in ways that prioritize profit over well-being.

The privacy policy also allows DeepSeek to share user data during corporate transactions, such as mergers, acquisitions, or sales, leaving user data vulnerable to further potential exploitation to which they sign a blank check.

And it is worth noting the absence of independent audits or external validation, which means users must rely on DeepSeek’s self-regulation—a risky proposition for any AI system.

Issues With DeepSeek’s Interactions​

Third, in failing to address vulnerabilities in relationships, DeepSeek risks turning from a mediator into a predator.

DeepSeek’s policy positions user participation as contingent on significant data sharing.

For instance, while users can disable cookies, they are warned that this will result in diminished functionality, effectively coercing them into sharing data for a "seamless" experience.

Though users can delete their data, the policy offers little clarity on the consequences for long-term service use, creating an imbalance in the relationship between the platform and its users.

Moreover, its handling of user input, such as chat history and uploaded files, raises significant concerns about how the platform mediates human-AI relationships. Indeed, user-provided data is treated as a resource for the platform’s benefit (e.g., model training), without clear opt-out options for individuals who do not want their data used in this way.

While DeepSeek states users can exercise rights like data deletion or access, the process is buried under layers of verification.
Also, the platform’s privacy notice provides no assurances that the AI’s responses or outputs are rooted in integrity-led principles, leaving users uncertain about the trustworthiness of interactions.

Equally concerning, DeepSeek’s handling of dependent relationships, such as minors or emotionally vulnerable users, highlights critical oversights in its intermediation mechanisms.

While the policy acknowledges parental consent for users under 18, it lacks robust safeguards to prevent data misuse or exploitation of younger users. There is no mention of how DeepSeek’s systems detect or handle users in distress, such as those discussing mental health or other sensitive issues, creating a risk of emotional harm.

Finally, regular updates to the privacy policy are mentioned, but there is no clear process for users to track changes that could significantly affect their privacy.

Redefining what we ask of AI—artificial integrity over intelligence—is the guarantee of AI’s performance being directed toward serving what matters most: humanity. Without this, economic value comes at the expense of societal well-being and therefore, individual lives.

AI needs performance that does not come at the expense of excessive energy, water, and terrestrial resources, nor lead to economic concentration in the hands of a few.

AI also needs to be ingrained with integrity, not just from an external standpoint, but primarily in its core functioning. Without this, artificially created intelligence can veer into harmful territory at a societal level, beyond what any developer could manage with a rollback.

On the former, let’s hope the promise of models such as DeepSeek R1 opens groundbreaking avenues; while most importantly, on the latter, ensuring that innovation empowers humans with machines, not the other way around—artificial integrity over intelligence.

 

[frugal.rock]

As soon as I heard about Deepsheet, I had my suspicions. Now reading some of that above makes me wonder;
Why on earth would the US let so many people download it when they are against Chinese ownership of Tik Tok in the US?

 

[divs4ever]

As soon as I heard about Deepsheet, I had my suspicions. Now reading some of that above makes me wonder;
Why on earth would the US let so many people download it when they are against Chinese ownership of Tik Tok in the US?

let ?? look how big Tik Tok got , Ali Baba , answer the US is slow to move

and probably 75% of the downloaders were not US citizens anyway

maybe the US military want to use Deep Seek as well , especially if Trump slashes their budget

 

[Craton]

From my POV, this DeepSeek is a storm in a teacup however, the cynic in me says psyops and from both fronts.

 

[divs4ever]

From my POV, this DeepSeek is a storm in a teacup however, the cynic in me says psyops and from both fronts.

depends on what the current AI( s ) really is

they have been working on computer chess engine for decades ( that is a massive super-computer to beat a lone human in a board game )

is the new AI just a glorified search engine/parameter filter ?

i can tell you one thing for sure , it has been the destruction of a huge amount of money up to date

 

[Smurf1976]

Australia needs a sovereign artificial intelligence (AI) capability.

To not do this would be the equivalent of deciding 30 years ago to not adopt the internet or before that not adopting the motor car, electricity or railways.

Today those things have become absolutely entrenched in the functioning of society, to the point nobody consciously "uses electricity" or "uses the internet" since they're in the background of practically everything. So too AI will end up in that category in due course.

To the issue of stocks specifically, the big question about DeepSeek is how real actually is it?

An awful lot of things have been faked over the years. What you see is a desktop computer 40 years ago or you see one person on stage or whatever but dig a bit deeper, follow the cables, and you find that what you're being told is doing the work isn't actually doing the work. Behind the curtain there's a great big computer that cost a fortune connected to the monitor you're looking at or there's a dozen musicians actually playing the music. Or worse still, the whole thing's just a recording.

Point being until it's proven otherwise, it's at least possible that DeepSeek involves an army of people sitting in front of computers creating the illusion that it's working or that it's really just repackaging some other existing AI tech and achieving the claimed efficiency by simply shifting all the effort onto a different account (put the real computers on a different electricity meter, that's one obvious trick).

That's not to say it is doing that, but so far as I'm aware there's no proof that it isn't so I'm wary. If it sounds too good to be true, tread with caution.