The Weekly Cybers #53
Chinese AI DeepSeek breaks people’s brains, the UK reports on AI safety, Donald Trump dabbles in cryptocurrency, and much more.
Welcome
Sorry this is a day late. Some medical stuff turned out to be more tiring than I’d planned for. Nothing to worry about, thank you for asking.
Once again developments in distant domains have dominated the digital discourse this week, especially DeepSeek, the new Chinese AI chatbot that’s reportedly way more efficient than existing systems.
We stay overseas with the UK’s International AI Safety Report, and a look at Donald Trump’s dabbling in cryptocurrency.
And as usual there’s lots of other things to think about. Enjoy.
DeepSeek’s AI triggers OMFG CHINA response
It’s cheaper, in many ways it’s better, and even more frightening, it’s from China.
You can’t have missed the news of DeepSeek, a new large language model (LLM) which rivals American models such as OpenAI’s ChatGPT, Google’s Gemini, Microsoft’s Phi, and Meta’s Llama.
And, as I say, it’s from China.
Investors immediately dumped tech stocks, with AI chipmaker Nvidia losing US$593 billion of market cap, “a record one-day loss for any company on Wall Street”.
In a longer-term context, however, that’s a relatively small blip at the end of two years of soaring prices.
So what’s the big deal?
DeepSeek is claimed to be far more efficient when it comes to the amount of computer grunt it needs, which means that new models can be developed using smaller and therefore cheaper computer clusters, which in turn means its datacentres use far less electricity and cooling water.
The fossil fuel industry is now worried its revenue might suffer.
DeepSeek has also been published as open-source software, or at least the core data-crunching code has. Anyone, including you or me, is free to download that software, train it with our own data, and commercialise that new model however we want.
No wonder the share market’s AI bubble wobbled.
What we don’t see, however, are the source code for the smartphone and web apps, the dataset which was used to train the model, or any further code which affects the results.
DeepSeek won’t talk about China’s politically sensitive subjects, for example, although the Guardian fooled it into doing so with a simple trick. I bet that won’t last.
In summary, then, rather than throwing ever-bigger computers at the problem, DeepSeek tackled the efficiency of the algorithms at the heart of LLMs — and then gave away their work for anyone to use.
What they can still make money on is services — their knowledge of how to train the models, which is bound to be better than anyone coming to it cold, and building and operating datacentres.
Privacy risks? National security threats?
Australia was first to call out the risks, because of course we were.
The Conversation detailed the potential privacy risks, which are real, but the vast majority of what DeepSeek does with your data is no different from any other AI chatbot — or any other online service for that matter.
“[T]his is all fairly standard practice for modern online services... Much of the cause for concern around DeepSeek comes from the fact the company is based in China, vulnerable to Chinese cyber criminals and subject to Chinese law.”
Security researchers have found security flaws which exposed user data, but these days who hasn’t suffered a data breach?
Meanwhile, authorities in the US and elsewhere are looking at DeepSeek’s potential national security risks and the interesting phenomenon of distillation.
As is so often the case, though, it’s mostly OMFG CHINA!
Even more stories about DeepSeek
- ABC News has a profile of founder Liang Wenfeng. In the photo he looks about 14 years old — that’s probably due to the schoolboy suit and the haircut — but we’re told he’s actually 40.
- There are claims that DeepSeek used computer chips which are banned from export to China, although one imagines that China will soon be making their own advanced chips.
- At The Conversation, an article setting out how DeepSeek’s success challenges assumptions about Chinese tech companies and US-China competition.
- As 404 Media puts it, ChatGPT creator OpenAI is furious that DeepSeek might have stolen all the data that OpenAI stole from us. Diddums.
- There’s a bunch more expert reaction at Scimex.
- The Cyber and Infrastructure Security Centre (CISC) is holding a meeting this Monday 3 February to discuss DeepSeek security risks, though it’s limited to members of the Trusted Information Sharing Network (TISN). Scrub off your security clearances!
UK report details AI safety risks
The UK government has released an independent International AI Safety Report, which they claim is “the world’s first comprehensive synthesis of current literature of the risks and capabilities of advanced AI systems”.
The Guardian has a summary of the main points:
- The impact on jobs will “likely be profound”, although many economists believe this could be offset by the creation of new jobs — and of course many sectors won’t be touched by automation.
- AI’s impact on the environment is a “moderate but rapidly growing contributor”.
“Datacentres and data transmission account for about 1% of energy-related greenhouse gas emissions, says the report, with AI constituting up to 28% of datacentre energy consumption.”
- Some experts fear that an all-powerful AI system could evade human control and even extinguish humanity, though opinions vary.
- AI could provide instructions for making pathogens and toxins for bioweapons.
- Autonomous bots could be used to find vulnerabilities in open-source software, although personally I assume this is already being done.
- And there’s a bunch of examples of deepfakes being used maliciously, “including tricking companies into handing over money and creating pornographic images of people”.
The 298-page report is well worth at least some of your time.
Trump pushes crypto, launches memecoin
One I didn’t cover last week: Donald Trump signed an executive order titled Strengthening American leadership in digital financial technology which aims to “ promote United States leadership in digital assets and financial technology while protecting economic liberty”.
Among other things, it creates a President‘s Working Group on Digital Asset Markets chaired by a Special Advisor for AI and Crypto.
As Investopedia reported, Bitcoin dropped slightly to around $103,000 in the hours following the release of the order. Clearly investors were expecting more.
Trump was also mocked for relating his own branded memecoin just before his inauguration.
“The digital coin called TRUMP appeared on his social media accounts ahead of his inauguration on Monday and quickly became one of the most valuable crypto coins. The value of a single coin shot up to $75 within a day, but since has fallen to $39.”
That BBC story also details some of the frothing from cryptocurrency speculators, with one even claiming: “TRUMP token just signaled to every company, municipality, university & individual brand that crypto can now be used as a capital formation and customer bootstrapping mechanism.”
Also in the news
- The Australian Electoral Commission (AEC) has warned that the election misinformation threat is local. Indeed, ABC News made a deepfake senator to show how easy it is.
- More than 5000 Australians have been warned they may be victims of romance scammers based in the Philippines. It’s part of a joint operation between the Australian Federal Police (AFP) and the National Anti-Scam Centre, and the Philippines’ Presidential Anti-Organized Crime Commission and National Bureau of Investigation.
- Communications minister Michelle Rowland reckons there’s “no way” Starlink could match NBN fibre ($), which must have Elon Musk trembling in his cowboy boots.
- eSafety Commissioner Julie Inman Grant expressed “great sadness” that the UK’s Southport killer Axel Rudakubana had viewed video of the Sydney church stabbing — the video X had fought to keep online — mere minutes before leaving home and murdering three girls in July last year. In her statement she said research shows a “clear link” between online material and real-world violence.
- CSIRO says it’s developed a practical application for quantum machine learning.
- Services Australia is drawing up a 10-year IT architecture strategy. I wonder whether it will include more on top of Microsoft’s $954 million of government deals.
- The Guardian has a good feature on farmers who want the right to repair their own tractors.
- Telstra has paid a $394,000 penalty for NBN migration failures.
IF YOU’VE FOUND THIS NEWSLETTER HELPFUL, PLEASE SUPPORT IT: The Weekly Cybers is currently unfunded. It’d be lovely if you threw a few dollars into the tip jar at stilgherrian.com/tip.
Elsewhere
- Donald Trump says Microsoft is in talks to buy TikTok. Trumps says many things, however.
- A delightful headline: “‘Everything I say leaks’, Zuckerberg says in leaked meeting audio.”
- Research shows that people with less knowledge about AI are actually more open to using the technology. Or to put it another way, the more you know about AI the less you want it.
- Australians who get most of their news from commercial media are more likely to believe in climate conspiracy.
- Siva Vaidhyanathan, professor of media studies at the University of Virginia, has a critique of Jonathan Haidt’s book against social media, The Anxious Generation: How the Great Rewiring of Childhood Is Causing an Epidemic of Mental Illness, the one behind Australia’s push against the socials.
Inquiries of note
Nothing new for us this week.
What’s next?
Parliament returns from its long summer break this coming Tuesday 4 February 2025.
The Senate draft legislation program includes the Oversight Legislation Amendment (Robodebt Royal Commission Response and Other Measures) Bill and the Health Legislation Amendment (Modernising My Health Record—Sharing by Default) Bill.
The House of Reps draft legislation program includes the Scams Prevention Framework Bill.
In both cases the government is likely to suddenly add things on the day, especially when they want to manage the news cycle. Watch what the politicians say over the weekend.
On Wednesday 5 February there’s a couple of hearings for the inquiry into the shutdown of the 3G network.
DOES SOMETHING IN THE EMAIL LOOK WRONG? If there’s ever a factual error, editing mistake, or confusing typo, it’ll be corrected in the web archives.
The Weekly Cybers is a personal look at what the Australian government has been saying and doing in the digital and cyber realms, on various adjacent topics, and whatever else interests me, Stilgherrian, published every Friday afternoon (nearly).
If I’ve missed anything, or if there’s any specific items you’d like me to follow, please let me know.
If you find this newsletter useful, please consider throwing a tip into the tip jar.
This is not specifically a cyber *security* newsletter. For that that I recommend Risky Biz News and Cyber Daily, among others.