AI Discussion?

[Buffalo Bob]

I just posted (on another thread) an image that I asked the Gemini AI software to create, which was for a two seat car blending the best features of the Lucid Air and the Chevy Corvette. Here's what it came up with in a matter of seconds...

I'm not sure if it's appropriate for this forum, but I wondered if others share the degree to which this technology is starting to concern me. My initial reaction is to be really impressed with it, and then I start to think about the not very long-term implications of where we are headed, with few if any guardrails in place. I understand completely if the mods decide to yank a discussion about this as too far off-topic.

 

[Steveinarizona]

I just posted (on another thread) an image that I asked the Gemini AI software to create, which was for a two seat car blending the best features of the Lucid Air and the Chevy Corvette. Here's what it came up with in a matter of seconds...

View attachment 26234
I'm not sure if it's appropriate for this forum, but I wondered if others share the degree to which this technology is starting to concern me. My initial reaction is to be really impressed with it, and then I start to think about the not very long-term implications of where we are headed, with few if any guardrails in place. I understand completely if the mods decide to yank a discussion about this as too far off-topic.

Wow. That is a beautiful looking car. In my opinion a much better looking version of its BMW and Mercedes competitors. It is scary. My Genesis has a feature where it figure out how my seat settings should be set. I had previously set up the seat settings at what I thought would be optimal for me. Then I decided to try the feature. The car came up with a completely different setting putting me much lower, etc. I forced myself to keep it for a week and found that I liked it much better than what I had come up with. So without using any external computational sources, my car's existing computer was able in a few seconds to do a better job than I had been capable of doing after 19+ years of education. It is an exciting new world but an equally scary one. Either we come up with some new solutions or in a few million years the creatures that will rule earth will be discussion how first the planet's dinosaurs had vanished and then the same thing happened with the great Ape human beings. Then they will take another breathe of their methane atmosphere and move onto other topics.

 

[Buffalo Bob]

My initial concerns are far more immediate. How must an automotive designer feel when seeing what this (no doubt de-tuned) version of AI cranked out in 30 seconds? I came up the ranks with an MBA/MIM degree doing jobs like production planning, forecasting, budgeting, economic/business analysis and a whole host of applied probability roles. I can see (most but not all) jobs like that going the way of the dodo, along with a host of others like the actuarial sciences, accounting, auditing, and most anything numbers-driven. And it's already extending into fields like engineering medicine, literature, law, the arts, etc.. It seems to me that AI might very soon do to the white collar world what robotics did to the blue collar world... only more quickly.

I think we have recently arrived at the point where we can simply tell one version of AI what we want, and then give it the unrestricted ability to write its own software to accomplish it. As integrated as AI is likely to be around the world, I hesitate to think what a dispassionate machine would do if it was asked to solve a problem like climate change, when it may just immediately conclude that the problem is us.

 

[doon]

I think we have recently arrived at the point where we can simply tell one version of AI what we want, and then give it the unrestricted ability to write its own software to accomplish it. As integrated as AI is likely to be around the world, I hesitate to think what a dispassionate machine would do if it was asked to solve a problem like climate change, when it may just immediately conclude that the problem is us.

Agree with your comments, not to mention the issue of questioning the voracity of what you see and hear with your own senses.

 

[joec]

Anyone can make a pretty picture of a car. Whether or not that car is practical in the real world as a shippable product is another universe.

Not car ends up looking exactly as some AI concocts it. There are thousands of compromises that go into making sure the car can be made safe, comfortable for actual humans to sit in, and manufacturable at scale.

In other words, car designers have nothing to worry about from parlor tricks like this.

 

[Buffalo Bob]

Anyone can make a pretty picture of a car. Whether or not that car is practical in the real world as a shippable product is another universe.

Not car ends up looking exactly as some AI concocts it. There are thousands of compromises that go into making sure the car can be made safe, comfortable for actual humans to sit in, and manufacturable at scale.

In other words, car designers have nothing to worry about from parlor tricks like this.

I wasn't suggesting that the car had been properly engineered, but if your job is design, it must be eye-opening. There are many traditional engineering jobs that I think will be on the block eventually. For example, new aircraft designs have come out that would never have been entertained, but they were designed and thoroughly 'flight-tested' without ever being built. And when they were built, the real world and virtual world was the same. Other perhaps than only an AI-assisted pilot can actually fly them. AI is way more than a parlor trick.

 

[xponents]

I wasn't suggesting that the car had been properly engineered, but if your job is design, it must be eye-opening. There are many traditional engineering jobs that I think will be on the block eventually. For example, new aircraft designs have come out that would never have been entertained, but they were designed and thoroughly 'flight-tested' without ever being built. And when they were built, the real world and virtual world was the same. Other perhaps than only an AI-assisted pilot can actually fly them. AI is way more than a parlor trick.

I disagree with the popular sentiment that AI will "replace" humans in their jobs, especially at the lower levels. My opinion is that AI will be used as a tool to help humans do their job, and yes, it will change the WAY we work. However, it will not render us useless.

If a junior level person is replaced with AI (as I see so many people saying will happen), who will be promoted to the senior roles, where communication is extremely important? AI cannot replace any C-level role, that's for sure.

Also, these AI models (thus far) cannot literally "think." That is currently impossible, and is somewhat similar in essence to how computers cannot generate truly random numbers. LLM's, as well as these new image-generation tools, need a "base" to start from, and this base is derived from the internet via training. Although you may have read about "reasoning" models (which are truly great, for all intent and purposes) such as ClosedAI's O1, DeepSeek's R1 (fantastic model!), Google's 2.0 Flash Thinking Experimental, Alibaba's QwQ, etc, these only SIMULATE reasoning, essentially imitating the way people online have reasoned out problems and trying to use it to solve the problem it's given. All an LLM does today is find patterns in data. This is why I think "AI" (currently, LLMs) will never be able to replace "synthesis-focused" jobs where people have to create new things (design included).

(Then again, from another perspective, humans also technically learn in the same way: we get a "base" of something from somebody else and use it to solve new problems. Is my reasoning here flawed? Even if true, I still do think that AI would not be able to replace humans due to the aforementioned senior role issue.)

If anybody disagrees, I'm curious, why?

 

[xponents]

I think we have recently arrived at the point where we can simply tell one version of AI what we want, and then give it the unrestricted ability to write its own software to accomplish it. As integrated as AI is likely to be around the world, I hesitate to think what a dispassionate machine would do if it was asked to solve a problem like climate change, when it may just immediately conclude that the problem is us.

Also, I disagree with this. You stated that you used "Gemini" to create your images. Gemini is simply a UI for their Gemini LLM models, and crucially in this case, their Imagen models for image creation. When you asked "Gemini" for an image, all it is doing is passing your text into another model created specifically for images, in this case, Imagen 3. As I previously mentioned, these created images are also simply identifying patterns in existing images that have the keywords you provide.

If you want to see this "in action," you can check this Google link out, which provides you a "direct interface" with Imagen 3 (rather than having to go through Gemini, which passes your information to Imagen): https://labs.google/fx/tools/image-fx

(P.S: I find this a really interesting discussion, and would hate if it was deleted. Please move it to the off-topic lounge if deemed necessary!)

 

[Buffalo Bob]

Also, I disagree with this. You stated that you used "Gemini" to create your images. Gemini is simply a UI for their Gemini LLM models, and crucially in this case, their Imagen models for image creation. When you asked "Gemini" for an image, all it is doing is passing your text into another model created specifically for images, in this case, Imagen 3. As I previously mentioned, these created images are also simply identifying patterns in existing images that have the keywords you provide.

If you want to see this "in action," you can check this Google link out, which provides you a "direct interface" with Imagen 3 (rather than having to go through Gemini, which passes your information to Imagen): https://labs.google/fx/tools/image-fx

(P.S: I find this a really interesting discussion, and would hate if it was deleted. Please move it to the off-topic lounge if deemed necessary!)

While I agree with what you've said in your second post (above), don't you think that's going pretty far into the weeds for a topic like this in an automotive forum? For the discussion I was looking for, and the general audience, I thought that just saying 'Gemini AI' was quite enough.

As for your first post... I think I'll need a little time and a fresh dose of my dementia meds before I reply!

BTW... Good suggestion on moving this discussion to the 'Off Topic Lounge'. TBH, I didn't know I had that option.

 

[Buffalo Bob]

Here is an interesting performance spec I read on Google’s latest quantum chip, code named Willow. Willow was given a standard benchmark computation test that would take one of today’s fastest supercomputers 10 septillion (that is, 10,000,000,000,000,000,000,000,000 ) years to complete. That number vastly exceeds the estimated age of the Universe. It completed the task in under five minutes. (I wonder if it was running in Smooth, Swift, Sprint or Launch mode.) The article then moved on to muse about how that performance opened the door to the notion that quantum computation occurs in parallel universes, in line with the idea that we live in a multiverse, at which point I decided it would be best to take a Tylenol and a nap.

 

[xponents]

Here is an interesting performance spec I read on Google’s latest quantum chip, code named Willow. Willow was given a standard benchmark computation test that would take one of today’s fastest supercomputers 10 septillion (that is, 10,000,000,000,000,000,000,000,000 ) years to complete. That number vastly exceeds the estimated age of the Universe. It completed the task in under five minutes. (I wonder if it was running in Smooth, Swift, Sprint or Launch mode.) The article then moved on to muse about how that performance opened the door to the notion that quantum computation occurs in parallel universes, in line with the idea that we live in a multiverse, at which point I decided it would be best to take a Tylenol and a nap.

I remember reading about that a couple of weeks ago, but I think that report is complete BS. What they essentially said is "our computer is so fast, therefore there must be parallel universes!" I fail to see any kind of logic in this deduction. It's certainly a very fast computer, and I think Google is well poised to become a leader in LLMs due to their computing power and the models they've released thus far, but what they are saying is complete nonsense.

While I agree with what you've said in your second post (above), don't you think that's going pretty far into the weeds for a topic like this in an automotive forum? For the discussion I was looking for, and the general audience, I thought that just saying 'Gemini AI' was quite enough.

As for your first post... I think I'll need a little time and a fresh dose of my dementia meds before I reply!

BTW... Good suggestion on moving this discussion to the 'Off Topic Lounge'. TBH, I didn't know I had that option.

Ah, I was not aware that you knew of how the image models worked! I was just trying to clear up what I thought was some confusion. My bad!

 

[RM-S8]

I disagree with the popular sentiment that AI will "replace" humans in their jobs, especially at the lower levels. My opinion is that AI will be used as a tool to help humans do their job, and yes, it will change the WAY we work. However, it will not render us useless.

If a junior level person is replaced with AI (as I see so many people saying will happen), who will be promoted to the senior roles, where communication is extremely important? AI cannot replace any C-level role, that's for sure.

Also, these AI models (thus far) cannot literally "think." That is currently impossible, and is somewhat similar in essence to how computers cannot generate truly random numbers. LLM's, as well as these new image-generation tools, need a "base" to start from, and this base is derived from the internet via training. Although you may have read about "reasoning" models (which are truly great, for all intent and purposes) such as ClosedAI's O1, DeepSeek's R1 (fantastic model!), Google's 2.0 Flash Thinking Experimental, Alibaba's QwQ, etc, these only SIMULATE reasoning, essentially imitating the way people online have reasoned out problems and trying to use it to solve the problem it's given. All an LLM does today is find patterns in data. This is why I think "AI" (currently, LLMs) will never be able to replace "synthesis-focused" jobs where people have to create new things (design included).

(Then again, from another perspective, humans also technically learn in the same way: we get a "base" of something from somebody else and use it to solve new problems. Is my reasoning here flawed? Even if true, I still do think that AI would not be able to replace humans due to the aforementioned senior role issue.)

If anybody disagrees, I'm curious, why?

Correct.
AI is a tool that's all.
When AI "generates" contiuesness, which it never will, then we may start worrying.

 

[Buffalo Bob]

I remember reading about that a couple of weeks ago, but I think that report is complete BS. What they essentially said is "our computer is so fast, therefore there must be parallel universes!" I fail to see any kind of logic in this deduction. It's certainly a very fast computer, and I think Google is well poised to become a leader in LLMs due to their computing power and the models they've released thus far, but what they are saying is complete nonsense.

There's more to it than that, but don't worry about it. We are probably all living in a simulation anyway.

 

[RM-S8]

Correct.
AI is a tool that's all.
When AI "generates" contiuesness, which it never will, then we may start worrying.

I meant consciousness, you see?
AI at work.

 

[segbrk]

Here is an interesting performance spec I read on Google’s latest quantum chip, code named Willow. Willow was given a standard benchmark computation test that would take one of today’s fastest supercomputers 10 septillion (that is, 10,000,000,000,000,000,000,000,000 ) years to complete. That number vastly exceeds the estimated age of the Universe. It completed the task in under five minutes.

Kind of off the off topic topic now, but I wanted to comment on that. People like to say things like that because that’s how they get funding, but that’s missing the whopper of a footnote: it can do SOME things fast. We’re not going to be walking around with quantum iPhones in our pockets in a few years, they are not general purpose processors. Not to mention they have to run in a supercooled box or the physics simply don’t work. That’s not to say that they (some of them at least) aren’t very impressive and useful for… things. It’s like the rule here about not saying “my $15 matchbox car can do loopty loops, why can’t my $250k Lucid???” Not a super meaningful comparison, because while true, the matchbox car probably also doesn’t do a sub 2 second 0-60. Or fit people inside.

 

[Buffalo Bob]

Kind of off the off topic topic now, but I wanted to comment on that. People like to say things like that because that’s how they get funding, but that’s missing the whopper of a footnote: it can do SOME things fast. We’re not going to be walking around with quantum iPhones in our pockets in a few years, they are not general purpose processors. Not to mention they have to run in a supercooled box or the physics simply don’t work. That’s not to say that they (some of them at least) aren’t very impressive and useful for… things. It’s like the rule here about not saying “my $15 matchbox car can do loopty loops, why can’t my $250k Lucid???” Not a super meaningful comparison, because while true, the matchbox car probably also doesn’t do a sub 2 second 0-60. Or fit people inside.

Ha! I wouldn't worry about being any further off-topic once the discussion turns to parallel universes and multiverses!

Excellent points. I did know that particular benchmark test had no useful purpose other than benchmark testing, and the supercooling requirements are no big surprise. But wow, only five minutes? That's one helluva loopty loop!