Major AI breakthrough.

[meanders]

The BBC just did an section on it. They asked questions like 'what happened in Tiananmen Square' and it didn't know. It seems as though it has been got at by the censors for sensitive subjects from China's position.

 

[Bill-H]

I am stunned that they released the whole thing as open source.

I'm not.
With Trump threatening the Chinese with tarrifs, making it open source does rather cause panic and confusion to the Americans - hence the fall in the value of their stocks - trillions wiped off and no doubt many going bankrupt.
Why would you spend a fortune buying the American version when you can get a Chinese one for free?

 

[The Wino]

I'm not.
With Trump threatening the Chinese with tarrifs, making it open source does rather cause panic and confusion to the Americans - hence the fall in the value of their stocks - trillions wiped off and no doubt many going bankrupt.
Why would you spend a fortune buying the American version when you can get a Chinese one for free?

I bet they recover 2/3 of the loss today. I think waving the big stick on trade wars for either side risks smacking yourself firmly in the face. It's a global economy that no one totally controls.

 

[AndyPK]

As meanders said, the Chinese version is pretty suspect wrt ‘sensitive’ subjects - ask it about Taiwan or Uyghur human rights abuses and see what it says………

 

[CRD]

Interestingly we were chatting to someone at the weekend who said using AI has transformed the way they design systems for their customers. They can survey their customers needs and analyse the results way more effectively then tailor responses. I didn't totally get what he was on about other than it's a pretty big improvement over what they did before

I’ve used ChatGPT to develop some advanced PHP programming code that was beyond my capabilities as a professional programmer.

I used my experience to guide ChatGPT when it created bugs, or didn’t quite understand my requirements, but it got there in the end with very impressive results.

The fact that DeepSeek is open source will be reassuring as there will be many hIghly skilled programmers throughout the world that will thoroughly examine the source code for suspicious spying or data retrieval functionalities.

You can also run it on your own hardware as Gromett suggested and analyse all the port communications in real time to see if the software is “calling home” to any destinations such as China.

^{_{Subscribers  do not see these advertisements}}

 

[Bill-H]

I bet they recover 2/3 of the loss today. I think waving the big stick on trade wars for either side risks smacking yourself firmly in the face. It's a global economy that no one totally controls.

2/3 of the loss is still a drop in value of 33%, when traders work on margins of a sliver of one percent.
The stock market is just gambling - not really any different to betting on horse racing at the end of the day, some win and some lose

 

[The Wino]

2/3 of the loss is still a drop in value of 33%, when traders work on margins of a sliver of one percent

Only if the original drop was 100%!. I don't think they dropped that much!

 

[suavecarve]

I'm not.
With Trump threatening the Chinese with tarrifs, making it open source does rather cause panic and confusion to the Americans - hence the fall in the value of their stocks - trillions wiped off and no doubt many going bankrupt.
Why would you spend a fortune buying the American version when you can get a Chinese one for free?

How much tariff will Trump put onto the Chinese app ?

The timing of a free app, which apparently, has caused loss to Trumps friends, seems like a bit of a warning shot, to me.

 

[Bill-H]

Only if the original drop was 100%!. I don't think they dropped that much!

maybe I didn't word my reply very well, a recovery of 2/3 of the loss is still a loss

 

[Podney]

I have no idea what the tech stuff of this thread means, but it is very interesting.

Me neither,not a scooby! But then I didn’t understand the relevance of binary when we did it in maths at school in the seventies!

^{_{Subscribers  do not see these advertisements}}

 

[Gromett]

I got you

Amazon.co.uk

That is a nice cheap one.

 

[Gromett]

The BBC just did an section on it. They asked questions like 'what happened in Tiananmen Square' and it didn't know. It seems as though it has been got at by the censors for sensitive subjects from China's position.

That is the Chinese hosted version. The filter is after the AI has produced the output as I read it.

However, as we have the model weights, we can always retrain with unfiltered data if necessary.

Watch the Dave Plumber video above where he asks about the famous picture of a man in front of a tank.

 

[Barti Ddu]

Me neither,not a scooby! But then I didn’t understand the relevance of binary when we did it in maths at school in the seventies!

 

[BouncyBoatFolk]

Me neither,not a scooby! But then I didn’t understand the relevance of binary when we did it in maths at school in the seventies!

Well, it's nice to have a choice .
Mike.

 

[OldAgeTravellers]

This guy actually runs it and shows how it does its thinking. I haven’t watched the whole thing but really fascinating:

^{_{Subscribers  do not see these advertisements}}

 

[GJH]

Why should anyone be surprised that a company has developed a far more efficient (AI) model than previous ones? For far too many years major companies have relied on faster and faster hardware to make up for the deficiencies of the inefficient code which their systems use.
The demand for bigger and bigger data centres (which waste massive amounts of electricity) is not purely driven by the requirement to store pure data but by the requirement to store inefficient code, simply because those in charge find it easier than doing the job properly.

I am not surprised, either, by the restriction on "sensitive" questions in the Chinese model. It seems to me that many of the so-called "AI" models in use are written with a bias towards directing the user to what the company wants to sell rather that what the user wishes to buy.

 

[cyberyacht]

Where's John Connor when we need him?

 

[Googlebot]

Where's John Connor when we need him?

He’ll be back.

 

[L' Hobo]

My firm are moving into AI big time.

I don’t really understand it but it’s looking like if you don’t get with it you will go down the pan.

You don't really understand it, neither do they.
The Accountant realise it saves them money and, as the controlling person in most companies over the past few years, they will accept it and to hell with the consequences!

 

[jongood]

CCP know that knowledge = power and they're quite happy to play the long game.

^{_{Subscribers  do not see these advertisements}}

 

[jongood]

Just Noticed Dave Plumber has released a video on the subject. He is brilliant at explaining complex subjects to normal people.

WELL worth a watch.

Thanks Gromett, well worth a watch.

"The cheaper AI systems are more prone to hallucinations and can generate confident but incorrect answers" I've seen some of this, indeed it wouldn't totally surprise me if we have a few on this forum.

The other point is that, in the wrrong hands it could be quite dangerous.

 

[GJH]

"The cheaper AI systems are more prone to hallucinations and can generate confident but incorrect answers"

The fact that some companies (Google was the first I noticed) use the term "hallucinations" speaks volumes for the con which is the basis of the current AI hype.
What they really mean is "We haven't a clue whether the answers our programs produce are correct or not because we haven't put enough care into design, build and (especially) testing, but we daren't admit that".

 

[L' Hobo]

The fact that some companies (Google was the first I noticed) use the term "hallucinations" speaks volumes for the con which is the basis of the current AI hype.
What they really mean is "We haven't a clue whether the answers our programs produce are correct or not because we haven't put enough care into design, build and (especially) testing, but we daren't admit that".

IF the rumours are true and AI computers can teach other AI computers (and vice versa) how can they test it except by asking another computer?

 

[The Wino]

The fact that some companies (Google was the first I noticed) use the term "hallucinations" speaks volumes for the con which is the basis of the current AI hype.
What they really mean is "We haven't a clue whether the answers our programs produce are correct or not because we haven't put enough care into design, build and (especially) testing, but we daren't admit that".

Perhaps it's like driverless cars. Some say they are no good unless 100% safe in every situation the reality is that no human driver is (although some think they are!). For a computer replacing a task previously done by a person if they are more accurate and reliable than the average of the humans previously doing that task it's an improvement. If as I suspect they are able to do a task way better than the average it's a big improvement.

 

[Onemanandhisdog]

Working for a very large I.T. company we are being forced down the AI training path, mostly to tick boxes for having x many AI accredited employees. On a monthly call with my manager last week we chatted about it, he is a bit younger than me and said he had to create AI solutions for his degree 25 years ago, AI isn't new it is just becoming easier to 'use' with modern computers, the latest IBM Mainframes have dedicated AI processors.

If you google how long has AI been around ....
"Artificial intelligence (AI) has been around for thousands of years, with its roots in ancient philosophy. The term "artificial intelligence" was coined in the 1950s, and AI research has advanced rapidly in recent decades."

^{_{Subscribers  do not see these advertisements}}

 

[BouncyBoatFolk]

I've used the AI function on Ebay for listing stuff, but you have to check the carp it's produced in it's normally flowery verbiage as it just parrot fashion replicates from other stuff available, faults as well. So if you're not careful your add just ends up as more fodder for it's inaccuracies.
Mike.

 

[L' Hobo]

Perhaps it's like driverless cars. Some say they are no good unless 100% safe in every situation the reality is that no human driver is (although some think they are!). For a computer replacing a task previously done by a person if they are more accurate and reliable than the average of the humans previously doing that task it's an improvement. If as I suspect they are able to do a task way better than the average it's a big improvement.

But sometimes, as Scotland & other places have found this winter, if there's a power cut (which can go on for days) the human might be able to keep working doing other work.
The computer cannot!

 

[The Wino]

But sometimes, as Scotland & other places have found this winter, if there's a power cut (which can go on for days) the human might be able to keep working doing other work.
The computer cannot!

But does that mean we employ lots of people just in case? It's like when we get severe disruption due to snow do we spend enough on equipment and people who are in reserve just in case? If we did that for every eventuality could our economy cope?
I think it's the case that if we are going forwards as a country we need to move with the times it's been the same throughout the whole of the industrial revolution particularly with the rise of CNC manufacturing

 

[jongood]

Perhaps it's like driverless cars. Some say they are no good unless 100% safe in every situation the reality is that no human driver is (although some think they are!). For a computer replacing a task previously done by a person if they are more accurate and reliable than the average of the humans previously doing that task it's an improvement. If as I suspect they are able to do a task way better than the average it's a big improvement.

From what I understand so far, it could be a liability in an autonomous car. Decisions at a lower level could be that - the balance of probability is that there is not any traffic approaching this junction so I can pull out without stopping.

You're right, it does sound like some humans

 

[GJH]

Working for a very large I.T. company we are being forced down the AI training path, mostly to tick boxes for having x many AI accredited employees. On a monthly call with my manager last week we chatted about it, he is a bit younger than me and said he had to create AI solutions for his degree 25 years ago, AI isn't new it is just becoming easier to 'use' with modern computers, the latest IBM Mainframes have dedicated AI processors.

If you google how long has AI been around ....
"Artificial intelligence (AI) has been around for thousands of years, with its roots in ancient philosophy. The term "artificial intelligence" was coined in the 1950s, and AI research has advanced rapidly in recent decades."

I remember use of the term decades ago, when working in IT. The difference now is that it is used inaccurately, like so many other terms, to con people into buying what is for sale when it cannot otherwise be justified.

^{_{Subscribers  do not see these advertisements}}