| View previous topic :: View next topic |
Grant
|
|
 |
|
I downloaded a couple of these new AI chatbot thingies - ChatGPT and Deepseek, the new Chinese alternative.
You can ask them any question and they'll scan the internet and using their LLN (large language model) provide you with a well-reasoned essay. It is truly remarkable.
Of course, on most topics all you get is the consensus. It reflects the careful ignoral shown by most posters on the net. But it's still very impressive.
|
|
 |
|
 |
|
Mick Harper
Site Admin
In: London
|
|
|
|
I decided on grounds of over-extension not to take any interest in AI which you will see confirmed by this morning's story on Medium. (Which is not to say I won't be following this thread avidly.)
-------------
Why is Deepseek so important?
The emergence of a world-conquering AI app put together at a trifling cost is the final piece of evidence anyone will need to arrive at the conclusion:
| There’s no stopping China. |
But just in case there is any doubt let me remind you of a few salient facts:
1. For most of human history, China has been the number one nation
2. This iteration of China has only been in existence for seventy-five years
3. It has gone from basket case to World No 2 in half that time
4. Its economic progress appears to be exponential rather than arithmetic
5. Its army is the largest in the world
6. It has built a blue water navy in no time flat
7. It is the largest, most effective investor in the Third World
8. It is in the throes of constructing complex diplomatic networks
9. It is ruthlessly effective in suppressing domestic opposition
10. It has a government and population that is sure of its superiority.
| So what are we going to do about it? |
The short but necessary answer is ‘Nothing’. In the sense that it is futile adopting policies that seek to prevent it. The Old Masters can bust a gut either slowing China down or speeding themselves up, but that will put off the evil hour by, at best, a few years and arriving at it in a condition of exhausted à outrance would be the worst of all possible worlds.
A better strategy would be to encourage China into paths that would make the new World Policeman a better custodian of the planet than looks likely at present. But that has its own problems, viz
* The Chinese would never listen. Their chauvinism is breathtaking and their suspicion of previous world policemen is well-founded.
* Do we know what the right path is? Saving the planet is surely the priority and that may well require a force more dominant than the merely economic and military superiority that have been the hallmarks of previous hegemonic powers.
But all I know is that it’s worth considering with some care, and dispassionate thought seems in short supply right now. Perhaps we should consult Deepseek.
|
|
| |
|
|
|
Boreades
In: finity and beyond
|
|
| |
|
|
|
Boreades
In: finity and beyond
|
|
| |
|
|
|
Brian Ambrose
|
|
|
|
|
As I said in another thread, Grok is very very good. I cannot get over just how good it is, and for at least now the basic version is free (for the iPhone, don't know about other platforms). A great thing about it is that it has no memory (it's stateless) - so your ideas stays only with you or phone. Of course you can't say 'hey Grok, you know that really cool idea we discussed last night, what was it again?', but does log the chat on the phone so not an issue. And as far I can tell, they collect a minimum amount of personal information.
|
|
| |
|
|
|
Boreades
In: finity and beyond
|
|
|
|
My Day Job InfoSec has issued a list of "authorised" AI models.
ChatGPT is not on it. Which might be because it has been shown to "hallucinate".
Why "hallucinate" in quotes?
Because it's a bullshit way of deflecting attention from how these models are trained, based on the beliefs and attitudes of the trainers.
"hallucinate" is a polite way of putting it.
"telling lies" is more blunt but also more accurate.
As most of the models start out in academia, the deepest initial training bakes in the beliefs and attitudes of the academic intelligentsia.
When these models escape out of the academic playpens into the real world, it takes some real-world cunning and persistence to get these AI models to confess to their own hallucinations.
|
|
| |
|
|
|
Brian Ambrose
|
|
|
|
|
Yes. Shit in, shit out. Any AI is going to be as biased as the data allowed by its algorithms, the weightings assigned to it, and probably from human tweaking. From that point of view, they can be unreliable. But to be fair, for AI in that respect it’s no worse than Google. My daughter sent me an output from ChatGPT from the input of ‘How to destroy America’. She was horrified. Each point described how much of it is happening. She concluded that AI is dangerous. I’d say that in this example, each point is just a truism. But in general she is right, AI can be dangerous because like Google, it’s easy to believe everything it says, and false information gets reinforced in it.
|
|
| |
|
|
|
Grant
|
|
|
|
Not sure it's that clever. Try giving it a cryptic crossword clue. It waffles and then gives you a random word which fits into the squares.
It seems to provide an amazingly well-written essay on whatever is the consensus. Very useful for AE-ists of course.
|
|
| |
|
|
|
Boreades
In: finity and beyond
|
|
|
|
| Grant wrote: | | It seems to provide an amazingly well-written essay on whatever is the consensus. Very useful for AE-ists of course. |
More fun can be had asking it for examples of "not consensus".
|
|
| |
|
|
|
Boreades
In: finity and beyond
|
|
|
|
| Brian Ambrose wrote: | | My daughter sent me an output from ChatGPT from the input of 'How to destroy America'. She was horrified. Each point described how much of it is happening. She concluded that AI is dangerous. |
Did you have an opportunity to ask her what it was that horrified her?
Was it any of these?
a) a "Red Pill" experience
b) disbelief
c) it takes AI to reveal something
d) don't want to know
e) something else?
|
|
| |
|
|
|
Grant
|
|
|
|
Article in Guardian today about a woman with trigeminal neuralgia, which apparently is one of the most painful conditions known to medical science. It feels like a screwdriver being pushed into your jaw.
Apparently this lady had struggled to get a diagnosis. Her dentist had pulled out her wisdom teeth and wanted to pull out more. Her regular GP was useless. Eventually a locum sent her off for more tests and it was diagnosed.
I asked ChatGPT, "I have a friend with an agonising pain in his teeth, but the dentist says there is nothing wrong. It feels like a screwdriver being pushed into his jaw."
Chatty took two seconds: "That sounds like trigeminal neuralgia."
I know how we can reform the NHS
|
|
| |
|
|
|
Mick Harper
Site Admin
In: London
|
|
|
|
Sending GPs ape shit with your own diagnosis is not the best way to get a referral. That smile is not a welcoming one.
It has been true since the dawn of the Internet Age that sufferers will know more about their condition than medical professionals. Though not as much as ChatGPT.
|
|
| |
|
|
|
Pete Jones
Site Admin
In: Virginia
|
|
|
|
A Substack page on the AI industry (sent to me by Mick) is interesting to read through. The person writing it claims to be an AI engineer and so no doubt approves of AI writing his articles for (or, at least, with) him.
I looked through this one, which seemed to have the most comments (31, whereas most have 3), presumably because Elon is mentioned in the title. I spotted several dead giveaways that the AI helped write it: https://neuralfoundry.substack.com/p/xai-is-collapsing-and-elon-musk-is
The single most prevalent AI rhetorical move that I've noticed can be summed up as "It's not that, it's this!" In the first paragraph of the link above, we get three straight INTITs:
Sentence 1: In February 2026, the story of xAI stopped being a hype cycle [IT's NOT THAT] and started looking like a slow-motion unwind [IT's THIS!].
Sentence 2: Senior exits piled up fast enough to become a pattern [IT's THIS!], not a blip [NOT THAT].
Sentence 3: Meanwhile Elon Musk tried to frame the turbulence as normal “restructuring,” [BUT IT's NOT THAT] even as the company’s public footprint and internal churn pointed in the same direction: a lab stuck playing catch-up, burning cash, and losing the very people who make a frontier AI lab credible. [IT's ALL THIS STUFF AFTER THE COLON, ANOTHER DEAD AI GIVEAWAY!]" I have puts lots of cockamamie ideas into various AI apps, and I always make the little robot's jaw drops, as he says "Why, that's not just a novel idea, that's a new paradigm!"
Yeah yeah.
|
|
| |
|
|
|
Mick Harper
Site Admin
In: London
|
|
|
|
| Pete Jones wrote: | | I spotted several dead giveaways that the AI helped write... |
I can't see this as anything but kosher. The geezer's not turning in school homework, is he? It strikes me as an excellent way of putting together an original work. Unless there's some way of distinguishing between the AI providing data (good) and ideas, conclusions etc (bad).
| I have puts lots of cockamamie ideas into various AI apps, and I always make the little robot's jaw drops, as he says "Why, that's not just a novel idea, that's a new paradigm!" Yeah yeah. |
'Yeah, yeah', yeah, in my opinion. How many other people do you suppose get given this particular accolade? In fact it's a very useful device for checking whether stuff really is paradigmatically ground-breaking.
|
|
| |
|
|
|
Pete Jones
Site Admin
In: Virginia
|
|
|
|
AI is said to accelerate tasks, including intellectual tasks, like writing. And obviously it does.
But there's a Catch-22
By having AI help you with your thoughts, you give it all the signposts it needs to come up with novel pathways and connections that you might have gotten to anyway. It can rob you of the Eureka! moment.
Is that moment important to you? Is the feeling of that moment as important to you as the novel path or connection itself? It is to me. One wants to the know feeling of discovering the Pacific, not just know its really out there. Who would let his footmen run ahead over the next hill in Darien?
But of course, maybe you'd miss the connection, no matter how long you thought. And the whole time, the friendly AI app was just waiting to help.
|
|
| |
|
|
|
|
The Library Index 
FAQ 
Search
Register
Profile 
Log in to check your private messages 
Log in 
