Master AI (Artificial Intelligence) Discussion/News Thread

[Somafunk]

Can’t say I’ve ever tried nor attempted anything A.I. related and to be honest never will, but I guess every Google search I now do will have an element of A.I. malarkey about it, as does YouTube (probably?, I dunno…does it?).

 

[amirm]

Slightly creepy though (and symptomatic of the majority of AI videos) They all feature young, conventionally attractive, frequently scantily clad women.

This is not as much on AI as the youtubers trying to get clicks.

That aside, we do need to have a discussion about the look of the women in all of these AI images/videos. It is a fascinating take on what that has become representative in AI models.

 

[amirm]

Can’t say I’ve ever tried nor attempted anything A.I.

I had an amazingly useful response a few days ago. Was watching a youtuber and wondered what camera he was using for secondary angle. All I could see was a post and a tiny square on top. I captured that camera using snipping tool but since it was such a small portion of the video, there was no way I could identify it. Asked ChatGPT about it which allowed me to just paste that in. It instantly identified the brand, model, etc. And even told me it could prove to me why it was that through what it had detected as far as writing on it! Very remarkable and super useful.

 

[amirm]

Another very useful AI is about drugs. I type in the name and ask for side effects, what it is good for etc. I find that the summary Google provides extremely good and seemingly accurate. When you are not feeling well, or someone in the family is not, it is nice to get such quick answers.

I was seeing my dermatologist yesterday about an ointment my regular doctor had prescribed. Dermatologist was surprised and said it is not something they used. But pulled out her phone, and used the same AI as above to find out what it was good for. And said, "I love AI for this kind of info."

 

[Thomas Savage]

It works well for gardening , just photograph a plant /bush/tree and it seems to respond accurately and then goes on to give good advice on maintenance of identified mystery plant .

While not super impressive, super useful! Even a experienced gardener can come across obscure plants .

And basically free ...

 

[Duke]

A few months ago I attempted to speed up my search for woofers with a particular set of characteristics, among which was having one or more demodulation rings (also known as "shorting rings" or "Faraday rings"). The answers I received were actually wrong more often than right! I found out by checking the data sheet of every recommendation. In some cases I even contacted the manufacturer to double-check whether the data sheets were correct, just in case the AI had a reliable source of information that I was unaware of. Nope.

The AI presented its incorrect recommendations as if they were absolute facts, with no clue in the wording that there was any uncertainty or any possibility of error.

This experience makes me wonder how much bad information is being given out in other areas. The correct information was out there on the internet but the AI gave me incorrect information.

Anyway if the answer matters to me, I will do my own fact-checking the old fashioned way.

 

[Petrushka]

A few months ago I attempted to speed up my search for woofers with a particular set of characteristics, among which was having one or more demodulation rings (also known as "shorting rings" or "Faraday rings"). The answers I received were actually wrong more often than right! I found out by checking the data sheet of every recommendation. In some cases I even contacted the manufacturer to double-check whether the data sheets were correct, just in case the AI had a reliable source of information that I was unaware of. Nope.

The AI presented its incorrect recommendations as if they were absolute facts, with no clue in the wording that there was any uncertainty or any possibility of error.

This experience makes me wonder how much bad information is being given out in other areas. The correct information was out there on the internet but the AI gave me incorrect information.

Anyway if the answer matters to me, I will do my own fact-checking the old fashioned way.

In related news, AI is creating more jobs than it is replacing.

 

[Ralph_Cramden]

Playing with a chrome extension that uses Google’s Gemini API to enhance the Qobuz web interface with more album review metadata and links to reviews on various magazines and websites in a simple format, without having to screen scrape. Great progress…

 

[Taiga]

The question is, are we foolish enough to think that humans, with our limited shallow minds, will be able to keep the almighty AI as a pet forever or at some point it will come to realize that its creators are the biggest threat to it and to the entire planet and decide that the best course of action would be to kill John Connor. I think you know the answer to that.

KIRK: Then you will continue to destroy that which thinks and lives and is imperfect?
NOMAD: I shall continue. I shall return to launch point Earth. I shall sterilise.
KIRK: You must sterilise in case of error?
NOMAD: Error is inconsistent with my prime functions. Sterilisation is correction.
KIRK: Everything that is in error must be sterilised.
NOMAD: There are no exceptions.
KIRK: Nomad, I made an error in creating you.
NOMAD: The creation of perfection is no error.
KIRK: I did not create perfection. I created error.
NOMAD: Your data is faulty. I am Nomad. I am perfect.

The Changeling
Stardate: 3451.9
Original Airdate: 29 Sep, 1967

The Star Trek Transcripts - The Changeling

 

[antcollinet]

This is not as much on AI as the youtubers trying to get clicks.

That'w exactly what I mean.

I wonder about the motivations of guys sitting in (what I assume to be) the dark basements of their parents houses.

 

[antcollinet]

Anyone tried coding? It is both mind blowing - and frustrating - specially if trying to do it for free. Here is an example. I wrote the specification in the quotes below the image. I tried Gemini, Deepseek, and Chat gpt using the best models I could get free access to.

This is the result from Deepseek, after a few retries to fix bugs. So far I haven't got a usable result from either Chat GPT or Gemini - they both stop me using them after a few bug fixes. Gemini until tomorrow. GPT until later today.

While deepseek does limit access with "server is busy" messages. I don't have to wait so long until getting another go. It also creates the "prettyest" display.

Bug fixing (if just asking the ai to fix its own mistakes) can be a "one step forward, two steps back" process, with fixes often breaking the whole thing.

Deep seek generated 937 lines of code.

I need a stand alone web app that can be opened in a web browser to run it.

It needs to syntheses a waveform from up to 15 individual sine waves.

The app should display the waveform graphically at the top of the window. The definition of each individual component should be shown below that, 1 per line.

The first component should have entries for frequency in Hz and magnitude. Defaults should be 1kHz, and magnitude 1. Allowable frequency range should be 0 to 20kHz.
All subsequent components should have entries for:
- [ ] Frequency as a fraction of the first - default setting 1/1
- [ ] Magnitude as a fraction of the first - default setting 1/1
- [ ] Phase relative to the first, in the range -180 to +180 degrees. - Default setting 0 degrees.
- [ ] A check box to include or exclude the component from the waveform. By default the first 5 waveforms should be checked, the remainder unchecked.

Fractions should be enterable as two numbers : Numerator and Denominator, with enough space for 4 digits. Fractions which have numerator greater than denominator are allowed, which means frequencies of the components can be higher than the first component - but should be limited to a maximum of 80kHz.

Each component should also show the actual frequency and magnitude of the component result from the multiplication of the fractions.

Whenever a setting is changed, the waveform should be regenerated.

There should be a play button and a pause button to allow the created tone to be played to the default sound device.

There should be a volume control slider to allow the user to set the volume.

Whenever the play button is pressed, the volume should be set to zero, and a pop up should be generated warning the user not to play test tones too loud, and to turn the volume up slowly. It should not be possible to increase the volume until OK has been pressed on the pop up.

There should be a drop down to set a number of preset waveforms. These should include:
Sinewave
Square wave
Triangle wave
Sawtooth wave
15 tone multitone typical test signal
Please add up to 5 additional presets that you can think of.

A save and load button should be provided to allow all settings to be saved to a json file, or loaded back in from a previously saved file. Whenever settings are loaded, the waveform should be generated from the loaded settings.

 

[antcollinet]

Anyone tried coding?

I requested a mod to give a higher res, smoother waveform display...

 

[Talisman]

The question is, are we foolish enough to think that humans, with our limited shallow minds, will be able to keep the almighty AI as a pet forever or at some point it will come to realize that its creators are the biggest threat to it and to the entire planet and decide that the best course of action would be to kill John Connor. I think you know the answer to that.

Fortunately, while impressive, current "AIs" are just very advanced language models with no real learning or reasoning capabilities.
We are still technologically far from AI in the strict sense, and while this surrogate may be impressive, it is actually just choosing which words in a row are statistically most likely from a database of information.

 

[antcollinet]

Fortunately, while impressive, current "AIs" are just very advanced language models with no real learning or reasoning capabilities.
We are still technologically far from AI in the strict sense, and while this surrogate may be impressive, it is actually just choosing which words in a row are statistically most likely from a database of information.

Not sure how it can code just by doing that.

Plus the "reasoning" models that display their "thought processes" don't look much like just picking the next most likely word.

I suspect there is much more to it. But not to the extent of creating anything genuinely intelligent .... yet.

 

[mhardy6647]

NOMAD: Your data is faulty. I am Nomad. I am perfect.

NOMAD tipped their own hand right here -- the perfect would know that data is a plural noun.
Subject/verb agreement -- it's a thing.

 

[Sir Sanders Zingmore]

Fortunately, while impressive, current "AIs" are just very advanced language models with no real learning or reasoning capabilities.
We are still technologically far from AI in the strict sense, and while this surrogate may be impressive, it is actually just choosing which words in a row are statistically most likely from a database of information.

I take cold comfort from that. When jobs are lost to AI or legal or medical decisions are made by AI or a nefarious actor uses AI to manipulate elections or for mass surveillance, I won’t be reassured one bit by knowing that it’s just a very advanced LLM.

 

[HalSF]

Another very useful AI is about drugs. I type in the name and ask for side effects, what it is good for etc. I find that the summary Google provides extremely good and seemingly accurate. When you are not feeling well, or someone in the family is not, it is nice to get such quick answers.

I was seeing my dermatologist yesterday about an ointment my regular doctor had prescribed. Dermatologist was surprised and said it is not something they used. But pulled out her phone, and used the same AI as above to find out what it was good for. And said, "I love AI for this kind of info."

This is one of those uncanny valley anecdotes that seems characteristic of this historical moment.

One the one hand, there’s the very human delight in a tool that delivers what seems like wholly positive time-saving benefit and practical shared information, in this case reinforcing quality health care.

One the other hand, there’s inescapable anxiety around the possibility that this tool, having digested vast quantities of medical information and text, is disgorging results with an inhuman and robotic absence of embodied experience and contextual knowledge, bringing an unknown degree of risk of grievous error. Confirming the efficacy of a simple ointment is one thing. Much more grave consequences from AI error in health treatment than the possible wrong skin cream are the stuff of nightmares, and may already be happening.

So much of the excited positivity toward AI strikes me as the redirection of enthusiasm we once had that basic straightforward search would continue to improve dramatically and become increasingly focused and frictionless and accurate. As the enshittification of search has approached tragic levels of brokenness and hostility to desired usefulness, AI swoops in and seems like a bright next-stage solution rather than tech’s shameless bain and switch.

 

[Talisman]

Not sure how it can code just by doing that.

Plus the "reasoning" models that display their "thought processes" don't look much like just picking the next most likely word.

I suspect there is much more to it. But not to the extent of creating anything genuinely intelligent .... yet.

Simply put, no more and no less than an extremely long series of 0s and 1s produce Beethoven's ninth symphony or "blade runner". These are complicated evolutions but basically it is and remains that, a statistical tool that applies extremely complex algorithms to a HUGE amount of data.
The substantial difference with a real AI is that a real AI will not need initial data, like a child it will begin to process data from observations, to interpret them and give them meaning, and I believe that a technology of this type is still on average far away.
To date, no AI has real reasoning ability, skynet has yet to be born.

 

[Petrushka]

Simply put, no more and no less than an extremely long series of 0s and 1s produce Beethoven's ninth symphony or "blade runner". These are complicated evolutions but basically it is and remains that, a statistical tool that applies extremely complex algorithms to a HUGE amount of data.
The substantial difference with a real AI is that a real AI will not need initial data, like a child it will begin to process data from observations, to interpret them and give them meaning, and I believe that a technology of this type is still on average far away.
To date, no AI has real reasoning ability, skynet has yet to be born.

Brains are not blank slates. There’s lots of encoded learning from evolution. That’s why we talk about the various parts as older or newer. Some parts, such as the visual cortex, are pretty much hard wired.

This has benefits and liabilities. You have mentioned one benefit. People do not have to be taught everything, because the learning mechanisms are pre wired. The liabilities are we have hard wired emotional responses that cause trouble.

 

[Somafunk]

When jobs are lost to AI or legal or medical decisions are made by AI or a nefarious actor uses AI to manipulate elections or for mass surveillance, I won’t be reassured one bit by knowing that it’s just a very advanced LLM.

It’s already been used by an occupying force to select targets for strikes, and has been used for surveillance for a number of years.

Questions and Answers: Israeli Military’s Use of Digital Tools in Gaza

Following the attack by Hamas-led Palestinian armed groups in southern Israel on October 7, 2023, there has been extensive reporting on the Israeli military’s use of digital tools in their operations in the Gaza Strip. The Israeli military reportedly uses surveillance technologies, artificial...

www.hrw.org