Text to Image AI Generation

[antcollinet]

it just copies humans. Even Amir imagined future space suite with sexy curves and titties lol

I mean, real AI would generate a suit that would look more like a simple overall. cause that is the tendency....less and less material needed

Yes it does. It mimics the biases inherent in the image sets it is trained on. They need to do better than that though, otherwise it is just perpetuation.

 

[antcollinet]

Everything is better with titties. AI understands that.

Im pretty certain you are not. I know for a fact i am not.

 

[amirm]

Edge browser got updated by itself and it pestered me to create an image. It gave part of a sentence: "create a futuristic image of" and then blank. I put in "audio amplifier." It created the most bizarre, weird looking image of a tube amp (I think) with blue lights and a camera lens in the middle! Before I could capture it, I lost it so tried again. Now it keeps saying this:

So now no one can try "Create an image of a futuristic audio amplifier!" Seems like such a kludge.

 

[Jmudrick]

Meta AI version:

 

[amirm]

That is better but it clearly shows the models have no intelligence to think what would be available 100 years from now. They are simply drawing from the past with some superficial touches like lights to indicate futuristic. In a century we sure would have computing all of such a suit and may not even need a suit of these types.

 

[MCH]

That is better but it clearly shows the models have no intelligence to think what would be available 100 years from now. They are simply drawing from the past with some superficial touches like lights to indicate futuristic. In a century we sure would have computing all of such a suit and may not even need a suit of these types.

True but maybe it has to do with the definition of "futuristic". I am not a native English speaker but in my language, the equivalent word has some fantasy implications, that is, something futuristic does not imply realistic, more the opposite. Maybe in English as well? I am copying here the definition from the Cambridge dictionary:

"strange and very modern, or intended or seeming to come from some imagined time in the future:"

 

[Jeromeof]

That is better but it clearly shows the models have no intelligence to think what would be available 100 years from now. They are simply drawing from the past with some superficial touches like lights to indicate futuristic. In a century we sure would have computing all of such a suit and may not even need a suit of these types.

Yes, so I tried the same futuristic amp on chatGPT and first go this horrendous looking thing:

The adjusted the prompt to ask for a minimalist amp (and I got something that looks like a Bluesound node or the new WiiM Amp

Still at least this looks like something that might sell.

 

[amirm]

Seems like if you boil down the collective postings of humans as represented on the Internet, "futuristic" translates into purple and pink hues! Seriously, they all have the same look on this front. I see youtubers with this color scheme. To me it says current fad, not future.

 

[Killingbeans]

I blame the game Cyberpunk 2077

 

[dasdoing]

the models have no intelligence

noone ever said they have intelligence. they don't. they are algorithms.

 

[amirm]

noone ever said they have intelligence. they don't. they are algorithms.

Everyone says they have intelligence. It is in the name (AI). An algorithm is an implementation and specific solution to a problem. It may or may not impart intelligence.

 

[mhardy6647]

Everyone says they have intelligence.

That is never the case!

 

[amirm]

I think we are on a path for the technology to pass the Turing Test in many areas. Language comprehension is already there now with the latest developments. My comment earlier was that it fails that test when it comes to predicting the future.

 

[GM3]

My comment earlier was that it fails that test when it comes to predicting the future.

Yeah, latest ChatGPT and others can somewhat recognize words & sentences and can spout out responses, but there's no real comprehension to speak of, in terms of 'human' comprehension. Simple analogy, I think current chatbots or text2img AIs are more like dictionaries, where you give an input, receive output, but there's no 'real' intelligence between, just a very large dataset and some randomness. No thinking, just an extremely large index of data, which it can somewhat reliably contextualize to an input.

It can do incredible things though; GothamChess played a game of chess against ChatGPT 4, and it did it did exceptionally well, until it didn't. But some of its answers are nothing short of incredible. It somehow played a good game of chess, without actually knowing how to play chess, which is pretty darn impressive.

So asking current 'AI' to draw something 'futuristic', is like asking a calculator to correct grammar. The 'best' current 'AI' can do is plagiarize from its dataset, which is largely populated by ... Why not let an AI tell us where it gets its data from!

where do current ai get their data for answers from

Answer
Artificial intelligence (AI) algorithms are fed with large volumes of data from various sources to make predictions and improve decision-making processes. These data sources can include surveys, observations, questionnaires, experiments, personal interviews, historical datasets, real-time data, internal data from organizations, and external data sources like the internet1
. Companies can leverage different types of data to implement AI, such as internal and external data, to extract valuable insights and improve the decision-making process1
. Additionally, there are online repositories, public datasets, web scraping, APIs, surveys, and partnerships that can be used to find data for AI projects5
. It is essential to consider the quality, quantity, relevance, diversity, and complexity of the data, as well as the legal and ethical implications of using it when selecting data sources for AI5

But I must admit, it can even somehow 'comprehend' very complex input, which is kinda unbelievable... Just the fact that it can solve math problems, or write programs to solve very complex problems...... Hard to understand how it's able to achieve this. Guess in a way, current AI is a bit like quantum mechanics, albeit likely lesser level: "if you think you understand quantum mechanics, then you don’t".

And the pace is amazing. Hard to imagine how it might look like in 10-20 years. Already self driving cars, AI painting, drawing, composing music, etc, etc., AI is truly the future

 

[IAtaman]

it fails that test when it comes to predicting the future.

That is a pass as far as I am concerned - it is indistinguishable from our human inability to successfully predict the future.

 

[dasdoing]

It is in the name (AI)

they would call a washing machine intelligent when it could turn from washing to centrifuging without user interference.
these algorithms are very far from being intelligent. they are just copying and joining information.
feed it for example with text of apples falling from a tree and how the earth is round. they can't conclude that there is a force attracting the apple to the center unless someone has already concluded it.

 

[Trouble Maker]

... it fails that test when it comes to predicting the future.

To be fair, there are probably exceedingly few humans that can do that.

 

[Killingbeans]

they would call a washing machine intelligent when it could turn from washing to centrifuging without user interference.

One of my pet peeves. The word "intelligent" is probably one of the most misused ones in marketing.

If my electric toothbrush doesn't have self-awareness, it's not bloody intelligent

 

[eddantes]

I think we are on a path for the technology to pass the Turing Test in many areas. Language comprehension is already there now with the latest developments. My comment earlier was that it fails that test when it comes to predicting the future.

Because current generative AI does not actually do "calcualtions", or "projections" - they are merely reflections of the consensus in their training data as it pertains to the concept that you asked for. So if you are looking for a synthesis of existing conceptualizations - it might produce that. Something novel - I very much doubt it.

 

[-Matt-]

It is quite easy to demonstrate the current lack of comprehension.

Using this prompt:

Create an image showing the line drawing on
a page of paper that would result from the
following sequence of actions:

1) Press the pencil tip down onto page.

2) Move the tip of the pencil right by 5cm.

3) Move the pencil tip down the page by 5cm.

4) Move the pencil tip left by 5cm.

5) Move the pencil tip up the page by 5cm.

6) Lift the pencil away from the page.

Microsoft Bing Image Creator (Dall.E 3) creates this: