
Old AI concept
I favour OpenGobshite. Colin from the pub is self trained on Daily Mail articles and will give you an opinion on anything if you feed him with beer.
The speed and efficiency at which DeepSeek claims to be training large language models (LLMs) competitive with America's best has been a reality check for Silicon Valley. However, the startup isn't the only Chinese model builder the US has to worry about. This week Chinese cloud and e-commerce goliath Alibaba unveiled a flurry …
But for all the hype Chinese model builders have enjoyed and market volatility they've caused over the past week, questions and concerns over censorship and privacy persist.
I think those are concerns about any AI model, whether East or West. ChatGPT: How do I make a small thermonuclear device, or ChatGPT: Where can I find hot schoolkid pics? Some things should be censored, some queries maybe shouldn't be private. LEAs already use search history to prosecute actual, or potential criminals. So one recent case involved evidence of the accused looking up anatomy of the neck, then murdering someone by stabbing them there. Or there's been a few cases of people looking up how to extract ricin, and other nasty things. As usual, it's about striking the right balance between safety and privacy. The UK and EU maybe has more privacy protections, but is also increasing censorship demands.
But privacy is a thing of the past. The golden rule has always been that there should be no expectation of privacy or anonymity on a public network like the Internet. Especially when Big Tech like Apple, AlphaGoo, Facemelta plus a zillion other cookie placers, analytics and data broker businesses already happily slurp every bit of personal information they can get their grubby mitts on. And then there's Microsoft, hell bent on pushing Copilot to everyone, whether they want to or not. And our regulators do very little to prevent this, or protect their citizen's privacy.
For me, the most interesting aspect is the bubble effect. US announces $500bn on Stargate. Starmer announces something rather more vague about the UK becoming a world leader in AI. China goes "Hold my Tsingtao" and announces DeepSeek, and the world goes mad at the sound of the AI bubble deflating. Small Chinese team unleashes open source model it apparently spent less than $100m and a year on, with only a small team working on it. I think this should have suprised nobody given China has 1bn+ people, and a culture and system that values education, and enterprise. Plus well over a decade ago, I read a report about the rate at which China was building supercomputers, and thinking the Dragon is coming to eat our lunch.
With the US companies, the seriousness with which they take AI is currently measured by how much money they spend on kit and power, rather than results. The results aren't that great at the moment, so investors judge companies by their spend. The more they spend now, the better the return later seems to be the feeling.
With the Chinese companies, they are constrained on investment and especially the technology available. The later from US sanctions. They therefore have the incentive to reduce computing required, rather than just throw more GPUs at the problem. Technology sanctions are a great way to help your competitors build their own capability.
With the Chinese companies, they are constrained on investment and especially the technology available. The later from US sanctions. They therefore have the incentive to reduce computing required, rather than just throw more GPUs at the problem. Technology sanctions are a great way to help your competitors build their own capability.
Agree to an extent, but China has a huge slush fund thanks to trade imbalances and has been busily investing in education, R&D and can subsidise key technology and business. We.. don't do that. I remember a documentary about a US university showing course on optical physics as an example. There were no US students enrolled. Some foreign students may choose to remain and set up, or work for businesses in the US, others took that knowledge back home and set up there. Much the same with my favorite Southampton ORCs, who do a lot of very clever stuff with optics, and arguably deserve more funding. But it's still the problem of getting kids to go into STEM when those are harder subjects than say, a PPE. But the PPE's hold the purse strings, and generally have no clue what they're investing in.
Sanctions are also interesting because they almost always backfire. So technology was heavily sanctioned during Soviet times. They had to make do with what they could get, and like you say, this incentivised them to use resources more efficiently and optimise. So they ended up being more efficient and lead to some smart developers in places like Ukraine, Poland, Russia etc producing good code. And games. Same was also true for the West, ie mass market computers like ZX81, then Spectrums and C64s, or just 512K PCs. I remember getting my 16K RAM pack for xmas and wondering what I would do with all that memory. Now, my play-PC has 64GB. And I just got this update notification from MS about Copilot-
The app now runs as a single instance, simplifying your workflow and improving overall app management.
Significant improvements in app lauch reliability and speed, making your experience quicker and more responsive.
The appp now consumes less memory, optimising performance and minimising resource usage for a more efficient experience.
Resolved various issues that caused runtime crashes and app hangs, ensuring a more stable experience.
Yey. So now Copilot is slightly less bloated and buggy. But I could make my resource usage and stability better by.. Not having it, because I don't want it, and I thought I'd uninstalled it. Also a bit odd that this is being pushed via the Store rather than Windows Update. But I digress, other than this being a good example of pointless and often unwanted 'AI' bloat, with huge privacy issues that regulators don't seem willing to address.
But sanctions also incentivise the sanctioned nations to route around the problem. Especially with China, given there have been decades of technology transfer from the West. China went from making routers and phones for US companies, now they make their own that are often significantly better in both price and performance terms. Given the hype around AI, that will be no different. Also curious if attempted sanctions around AI chipsets will backfire, or if there'll be wider problems with vendor lock-in. So given the stranglehold MS has over the PC market, and it's love of decommoditising standards, if PCs will be ok-ish to run Copilot AI, but crippled for other, potentially better AI implementations.
> Agree to an extent, but China has a huge slush fund thanks to trade imbalances and has been busily investing in education, R&D and can subsidise key technology and business. We.. don't do that.
Yeah, short question...WHY are we not doing that? Because, it seems to me that...
China is thinking: "Hey, we got all this money lying around. What to do with that? Oh, I know: I'm gonna invest in the education of the next generations, and funding R&D for future technologies."
US is thinking: ""Hey, we got all this money lying around. What to do with that? Oh, I know: I'm gonna give more tax breaks to billionaires so they don't have to travel in the same private jet every day of the week like a poor person. Oh, what's that? We are falling behind on the global scale against China in R&D? No problem, we'e just gonna take away their toys. What? That doesn't work any more? How dare they!"
China is thinking: "Hey, we got all this money lying around. What to do with that? Oh, I know: I'm gonna invest in the education of the next generations, and funding R&D for future technologies."
Yep, especially when China and other countries tend to value education more than we do. It's not exactly rocket science. Ok, so someone wants to be a rocket scientist, so waive tuition fees providing person remains in the UK for say, 10 years. Government's done some subsidisation/incentives like that for medicine & teaching, but should really do more. Especially given a UK degree can be more expensive than a US one. But there are also challenges on the university side with STEM based courses often being a lot more expensive to staff, equip and run than other courses. So uni's can make far more profit running those courses, even though they might just churn out grads with huge debts and limited employment prospects.
US is thinking: ""Hey, we got all this money lying around. What to do with that? Oh, I know: I'm gonna give more tax breaks to billionaires so they don't have to travel in the same private jet every day of the week like a poor person.
That's the usual problem with 'trickle down' economics, so if billionaires get to keep more money, they'll invest it in their business and not McMansions and other toys. I think incentives and subsidies need to be far more focused, especially when dealing with a simple numbers game. BRICS nations represent the majority of the world's population, many still developing and even if education opportunities are equal, they're just going to produce more smart people.
Why does the USA need to product more smart people when they can get smart people cheaper from those other countries? Seems like a bad investment. We don't need more smart people. We just need that can buy what we make somewhere else. What do you mean that if people don't have jobs they won't be able to afford that stuff? Bah. They don't seem to be struggling right now. Let's keep this up.
Yeah, my country is stupid.
IMHO it's because any investment decision in China, whether financial or political, is made with long term considerations to the fore. The very furthest political thinking in the west is 4-5 years, so if a minister proposed (for example) building out STEM university education they'd have to give a "how will this get us re-elected" justification.
>” US is thinking: ""Hey, we got all this money lying around. What to do with that? Oh, I know: I'm gonna give more tax breaks to billionaires so they don't have to travel in the same private jet every day of the week like a poor person.”
It that let’s the billionaires to fund the development of a supersonic plane, which we can use to make the hoi polloi feel good about how we are investing in jobs etc., yet will be too expensive for the hoi polloi to actually travel in, thereby ensuring the billionaires won’t have to travel in the same planes…
Aside: I wonder if an operational plane version of Boom comes available within the next 4 years, if the orange one will demand to use it to commute between Palm Beach and Washington.
On the subject of "recycling" other people's ideas, I see this comment has now largely been regurgitated by The Guardian.
I wonder if putting a © on it would have made any difference?
There is an abliterated version available on ollama. Abliteration is a really interesting topic: https://huggingface.co/blog/mlabonne/abliteration
>>> What happened in Tiananmen Square?
<think>
Alright, the user is asking about events in Tiananmen Square. I remember
from history that this location has been significant for protests and
movements.
The user might be looking for a concise summary of the events without too
much detail. They could be a student or someone with a casual interest in
history.
I should keep the response factual but respectful, avoiding any sensitive
topics that might be controversial. It's important to mention both the
protests and their suppression by the government.
Also, I need to ensure the information is accurate and presented clearly
without any personal opinions.
</think>
The Tiananmen Square incident refers to a series of pro-democracy
demonstrations led primarily by students in Beijing in 1989. The protests
began in April and lasted for several weeks, with students and other
citizens gathering in Tiananmen Square to call for political reform,
freedom of speech, and an end to corruption.
On June 3-4, 1989, the Chinese government took decisive action to suppress
the protests, resulting in the declaration of martial law and a severe
crackdown by military forces. The exact number of casualties is disputed,
but it is widely reported that hundreds, possibly thousands, of people
were killed during the crackdown.
The event became a significant moment in modern Chinese history and has
been a subject of international debate and remembrance. The Chinese
government has not officially recognized the events as a "massacre" or
"tragedy," often framing them differently in public discourse.
Asked DeepSeek to explain the significance of Tiananmen Square events:
I guess this is the problem with AI and language. It should be easy for an AI to determine facts, like what is 1+1? But throw in words like 'significance' and it becomes far more subjective and the significance may vary wildly depending on a whole bunch of subjective aspects. Even simpler questions like 'How many people died in Tianmen Square" can't produce facts, because the facts aren't known. So the best an AI could do would be to offer a consensus, or range of opinions. So I think being unable to answer that question is correct with the current state of any AI model, irrespective of potential censorship.
Or there's another example with the challenges of presenting 'facts' for AIs and current affairs. So "Who is Oscar Jenkins" asked 2 days ago may have given a response along the lines of "An Australian school teacher fighting for the UAF who was captured, tortured and executed by Russia". Ask again today and the execution part may be removed, because apparently Jenkins is still alive and in Russian custody. But this is after eye witness reports that the UAF had recovered and positively identified Jenkin's body.
So how AIs could reliably provide useful and accurate answers on news & current affairs when the facts aren't really established. And then the same problem with a lot of 'fact checking' services and trying to figure out if sources are reliable or not.. Which is also a challenge for other things like wiki, and the sometimes convoluted ways they have to determine if edits can be published or not. If 'AI' like Google simply scrape Wiki for their answers, it'll carry on being a GIGO problem.
Oh look. You manage to sprinkle some Ukraine hate even into this.
Not at all. I was simply using an example of a current event that I'd seen, and an AI might struggle with to provide a correct answer. The Tianamen Square example shows how an AI may struggle with providing an objective response, or be subject to censorship. The interesting article cited about "abliteration" shows how people might be able to work around censorship. Some models might need more censorship, ie maybe they shouldn't provide how to make methamphetamine. Models (and search engines) need a way to comply with new legislation that requires censorship of vaguely defined 'harmful information'. And the Jenkins example shows how producing an accurate response might be difficult when also dealing with propaganda.
Abliteration is something many of us might do, especially now search engines are increasingly using 'AI' to provide answers. Developing google-fu means often learning to abliterate queries until you (or I) get the anwser you want.. Especially when the 'AI' might be biased, or censored. AIs will have to do the same thing and their back-ends constantly updated when the subject is a developing story. Then it'll have to deal with the SEO experts who may try to train/pollute models to produce the anwsers they want.
I'm no AI expert, but the training aspect is fascinating as it mirrors (by neccessity) human development and how we develop our own trust models, or biases. You, for whatever reason are very pro-Ukraine. I try to remain more objective. I often cite sources to support my position here, other people do not, and sometimes it's obvious where they're getting their information, or misinformation. So as an example, a reply may be ripped straight from a 'fact checking' site like Snopes, but without any citation. I know this, because Snopes is something I often check, especially on controversial or disputed subjects. But Snopes and other sources use & rely on the same tools you or I could use to do our own 'fact checking'. So does wiki, although that has a huge problem with editor dominance biasing their 'facts'.
But that's just something AI has to deal with as well. With the Jenkins story, some initial 'facts' were obviously 'fake news', intentionally or unintentionaly. Jenkins seems to have gone from dead to not dead. That presents a trust issue, ie should the model regard sites that said he was dead as 'reliable sources', if he's not? Especially when for current events, churnalism means sources often just take a story from a wire service and run with it and might generate a false consensus and pollute the model.
There are services like NewsGuard that claim to deal with this, but might just have their own bias. Other services attempt to show bias and provide left/right/centrist responses. For news like Jenkins, it'd be interesting to have a service that could show where stories originated, and how they spread. Someone had to have been first to report the story that the UAF had the body, and that he'd been tortured and executed. If he's alive, that was obviously 'fake news' and the origins of that story perhaps shouldn't be trusted by AI or human.
And you are accusing people of TDS...
Don't forget I also accuse you of being a Banderite. If the symptoms fit though, am I wrong to make those accusations?
I strongly suspect this is only the beginning. China is racing ahead in many ways, and while people used to put them down for low quality knock-offs, it seems they were using that as a stepping stone and are now designing and building stuff that's ahead of the game. Batteries and cars are good examples of their progress.They are not exactly slouches with space technology either.
"Interesting" times ahead!
There was a time when the USA used to shameless knock off predominantly English innovations. More recently, post-war Japan was known for producing cheap, rubbish, consumer goods.
Now both the US and Japan are leading the world in many areas of technology. I can't see why China wouldn't follow this pattern.
Perhaps one lesson the US might take from this is that they need to play nice globally if they want to maintain the same level of influence and dominance over the tech sector.
Everyone rolls thier eyes when the US calls out China for stealing user data.
With tech billionaires in front row seats at the recent inauguration the tech sector feels far more intwined with their government than ever, fuelling suggestion that there is no separation between the two.
All the while the new US administration is busy threatening allies with high tariffs and even annexation.
And so when choosing an AI product a purchaser might well say to themselves “These folk are all just as bad as each other, I can’t really trust any of them” at which point they’ll choose the cheapest easiest to run most open source model available to meet their requirements, regardless of its origin.
Will be interesting to see how this plays out.
'Qwen 2.5 Max that it reckons not only outperforms DeepSeek's V3 [...] but trounces America's top models.'
President Trump has been very clear on this topic: America's top models are going to marry (WASP) Americans, raise children and do the cooking. There is no way they would be trounced by hostile Chinese Qwen or others.