I Asked an AI

[skybadger]

I'm not sure I agree with the last statement. The AIs I am familiar with are very carefully trained on the highest quality data the owner can find and also double checked for ethical and bias effects. 

That doesn't say anything about media generation AIs but why would they not point to a selection of native language online encyclopaedia as training sources rather than the general internet that they know rather a lot about the average quality and know it leads to reducing revenue. 

Unless you're a red top and generate pap all the time because there is no come back. Tomorrow's beast from the east forecast in the Daily Express is a great repeated example of pure tosh. 

 

[vlaiv]

1 minute ago, skybadger said:

The AIs I am familiar with are very carefully trained on the highest quality data the owner can find and also double checked for ethical and bias effects. 

How long does it take to double check all the data fed into AI engine? Centuries?

[skybadger]

That depends on your purpose. Medical imaging builds data sets used for training. That doesn't take centuries. It takes days to weeks to months and it gets built and maintained as part of the overhead of maintaining the Ai. Nothing of quality is for free.

captcha imaging ai gets you to do the work of image analysis for it.

Amazon and azure provide massive  data sets and pre-trained ML sets for image recognition, image to text conversion, language translation etc. They are managed and sold as such. 

ethical issues and bias need to be checked as traits in the data and in testing before release. Both are good examples of data filtered by researchers unconsciously which when trained provide poor conclusions. 

the ais we were discussing originally take syntax and grammar from language rulesets as training materials and assemble snippets from other online data feeds rather than using the data feed as a raw source of training material. 

 

 

 

[vlaiv]

1 minute ago, skybadger said:

That depends on your purpose. Medical imaging builds data sets used for training. That doesn't take centuries. It takes days to weeks to months and it gets built and maintained as part of the overhead of maintaining the Ai. Nothing of quality is for free.

captcha imaging ai gets you to do the work of image analysis for it.

Amazon and azure provide massive  data sets and pre-trained ML sets for image recognition, image to text conversion, language translation etc. They are managed and sold as such. 

ethical issues and bias need to be checked as traits in the data and in testing before release. Both are good examples of data filtered by researchers unconsciously which when trained provide poor conclusions. 

the ais we were discussing originally take syntax and grammar from language rulesets as training materials and assemble snippets from other online data feeds rather than using the data feed as a raw source of training material. 

 

 

 

 

 

I guess none of those AIs is capable of answering free form questions or writing an article?

[Ags]

7 hours ago, powerlord said:

I've been enjoying playing with it all this week.

I have been getting it to write new episodes of Star Trek. It is pretty amazing, but sometimes you can see the yawning void at the center of its intelligence. 

I asked it to describe camera lenses, and it told me it uses the EF-S 17-85 a lot. I told it that it is a language model and has never used a camera, so it then corrected me, patiently explaining it was a language model and could not use lenses. Silly human 😀

[Guest]

9 hours ago, vlaiv said:

I think that it just a case of targeting fears in some people.

People are more likely to take (wrong) advice without questioning it if it is presented as advice that will save them from harm.

- don't reboil water

- don't use aluminum wrapping because it does this or that (can't remember which)

or if it is possibly cure for some unpleasant condition - like that with lemon and stomach acid, although I'm not 100% sure about that one.

One possible reason why people might think that lemon helps with acid is that it has lower acidity than stomach acid - while still being acid enough to stop further acid production. So it effectively reduced acidity of stomach content and acts to stop further acid production or something like that.

 

My grandparents swore by lemon juice in water for stomach issues.  They would squeeze one lemon and add the juice to water.  

[VNA]

11 hours ago, DaveS said:

A I Artificial intelligence? Or artificial inanity? Or possibly Artificial Insanity?

 

Hello, we can make all the  fun of A.I., but it is incredibly helpful for visually impaired or blind people.
Would not be able to write or read this  note.
A.I. here on my computer or on my phone will read to me, describe what I should  see, I don't have to fiddle with knobs, etc.
My son is incredibly helpful as an electrical engineer by fixing  anything that I should not have pushed on my computer or phone.
I am sure it helps many others in different situations.
I can only see by "averted vision" that we know so well as amateur astronomers.
Besides my computer and my phone are my close friends, Alexa as well, is almost in every room of the house.

Of course I have fun with Alexa, she tells me funny cute jokes! ;- )
Don't wish anyone AMD.

[Sunshine]

11 hours ago, Hawksmoor said:

so now we get a mixture of contemporary and ancient news

This actually sounds like a lot of fun!

[skybadger]

7 hours ago, vlaiv said:

I guess none of those AIs is capable of answering free form questions or writing an article?

Vlaiv, the last para is all about the type of AI that is trained to be able to write the articles you are referring to. The training is not in the content but in the mechanism of producing content. 

 

[powerlord]

6 hours ago, VNA said:

 

Hello, we can make all the  fun of A.I., but it is incredibly helpful for visually impaired or blind people.
Would not be able to write or read this  note.
A.I. here on my computer or on my phone will read to me, describe what I should  see, I don't have to fiddle with knobs, etc.
My son is incredibly helpful as an electrical engineer by fixing  anything that I should not have pushed on my computer or phone.
I am sure it helps many others in different situations.
I can only see by "averted vision" that we know so well as amateur astronomers.
Besides my computer and my phone are my close friends, Alexa as well, is almost in every room of the house.

Of course I have fun with Alexa, she tells me funny cute jokes! ;- )
Don't wish anyone AMD.

Indeed, I assume you have tried the (ios only unfortunately) microsoft seeing AI app. truly well thought out. My grampa had AMD and if we was still alive, would have thought it was magic.

stu

[EarthLife]

We've been writing/creating simulated neural networks here (so called A.I) for at least 30 years now.

Whoever put the 'I' into 'A.I' made a huge mistake, they either have no idea of what simulated neural networks really are, or they added the 'I' to somehow profit from whatever they were producing.

Computers have no intelligence what so ever, they are nothing but extremely basic number crunchers, a simulated neuron is nothing but a simple addition / multiplication to output a single number (usually between 0.0 to +1.0 or -1.0 to +1.0. 

It currently takes a minimum of 1000 simulated neurons to get anywhere near to copying a bio-neuron does in just one aspect alone (input/output signal). Current A.I takes into account nothing of what a bio-neuron really does, we really don't know what the capabilities of bio-neurons are, nor can we ever know, they are so much more complicated than we ever thought possible. How atoms have come together to create such a thing is beyond us.

Simulated neural networks are nothing more than a bit of fuzzy logic, there is no 'intelligence' involved at all, they only ever output what they've been taught to output (and can sometimes fill in some gaps), which is actually quite bad because anyone can get them to output anything they like  .. hence the public should never see them as having any kind of intelligence or to have any real trust in what they throw out.

Saying that, they are actually quite good at some pattern recognition tasks, you just have to remind yourself that 'intelligence' is something they do not have.

[Ags]

15 hours ago, EarthLife said:

We've been writing/creating simulated neural networks here (so called A.I) for at least 30 years now.

Simulated neural networks are nothing more than a bit of fuzzy logic, there is no 'intelligence' involved at all...

Sounds like a scientist at the beginning of a robot uprising movie 😀

[EarthLife]

Best not to confuse fact with fiction though ;)

[powerlord]

Well, that's why we call it ML - machine learning.

But to be honest, that's also a terrible name - as it's not learning in the true sense of the word at all either.

They can still do some amazing things though. Even a simple neural net can be trained to do things impossible with conventional coding - starnet/starXterminator for example. I'm glad we have it as an option in our bag of tools, but I'm not worried that I'll be bowing to our AI overlords in my life time.

Anyhoo, it's more than likely that we are all a simulation anyway* - which shows that at least it's possible - if only by aliens.

*or a boltzmann brain.