Opinions on AI

[R-T-B]

Serious;y? There are many different types of AI. Tesla's FSD v12 has like 30 lines of code. There are no heuristics, it isn't told to do shit. It learns by being fed video. SpaceX uses machine learning to teach the boosters how to land. It's not heuristics.

It's still a glorified series of tables telling it "do this, not this" based on successes and failures. That's all training is. These tables are friggin' massive to accommodate things as complex as art, but that is literally all it is.

If you have some foundation with which to challenge this I'd love to hear it. I do actually have some background in this (mostly just me running my own models for personal use in of all things, an old strategy game), but I'm not infallible as it's certainly not my main field, so I can and will be open to corrections.

Tesla's FSD v12 has like 30 lines of code

I'm sure that's verifiable somewhere besides notoriously exagerated-factoid CEO man. /s.

Even if it isn't, it's possible to write an AI if it has a good dataset using that little code. But a "AI" is useless without it's training data, that's where the real value comes in (those tables I've been harping about).

Which is precisely why I think "AI" is less "AI" and more "machine learning" if even that.

And again, using an AI in a spacecraft would be stupid when a simple state machine is far more reliable and trustworthy, like the one I wrote above.

 

[thesmokingman]

It's still a glorified series of tables telling it "do this, not this" based on successes and failures. That's all training is. These tables are friggin' massive to accommodate things as complex as art, but that is literally all it is.

If you have some foundation with which to challenge this I'd love to hear it. I do actually have some background in this (mostly just me running my own models for personal use in of all things, an old strategy game), but I'm not infallible as it's certainly not my main field, so I can and will be open to corrections.


I'm sure that's verifiable somewhere. /s.

Yeap, typical. Don't have jack for proof to back up your opinion so you revert the typical troll behavior. Great job man. See the thread title, it's called opinion. Stop crapping on others opinions with the lame sarcasm. Proof, the proof on Tesla SpaceX is everywhere lmao. Elon only livestreamed v12 to the world, first time neural network end to end, no tables, no code heuristics, etc etc. And before that Ashok, Tesla head of AI has talked and shown live at the last AI symposium their vision training etc etc.

 

[R-T-B]

Tesla SpaceX

These are separate companies tied only by the CEO. I await your "proof" eagerly. The rest of your post is too ironic to warrant much discussion.

no tables

I really would like to start here. I don't troll.

 

[WhitetailAni]

first time neural network end to end, no tables

A neural network is controlled by a set of tables.

 

[R-T-B]

A neural network is controlled by a set of tables.
View attachment 317217

I tried to explain this, but couldn't think of how to cite such a foundational thing. Thanks.

 

[Assimilator]

That's how AI works. It's brain is essentially a giant spreadsheet of "this worked, this was bad"

That's pretty much how organic brains work too, though.

And again, using an AI in a spacecraft would be stupid when a simple state machine is far more reliable and trustworthy, like the one I wrote above.

Exactly. Far too much of the buzz around "AI" is idiots trying to reinvent the wheel.

Yeap, typical. Don't have jack for proof to back up your opinion so you revert the typical troll behavior. Great job man. See the thread title, it's called opinion. Stop crapping on others opinions with the lame sarcasm. Proof, the proof on Tesla SpaceX is everywhere lmao. Elon only livestreamed v12 to the world, first time neural network end to end, no tables, no code heuristics, etc etc. And before that Ashok, Tesla head of AI has talked and shown live at the last AI symposium their vision training etc etc.

The only troll here is the person who threw a tantrum when called out on their inability to provide any sort of proof for their claims. Do you know what we call those who throw tantrums when they don't get their way? Children. And children have no place in an adult discussion.

 

[AusWolf]

That's pretty much how organic brains work too, though.

Science works that way. Brains? Not really. We have way too many personal preferences, biases, tastes, and new ideas, plus we're capable of thinking outside the box. We can rewrite the spreadsheet with completely new data if we want to. Heck, we can even give it a new name or purpose. I don't think AI can do any of this.

 

[Assimilator]

Science works that way. Brains? Not really. We have way too many personal preferences, biases, tastes, and new ideas, plus we're capable of thinking outside the box. We can rewrite the spreadsheet with completely new data if we want to. Heck, we can even give it a new name or purpose. I don't think AI can do any of this.

Incorrect. Evolution is essentially an incredibly lengthy optimisation process, whereby the organisms that make "this worked" decisions live on and pass the results of those decisions to their offspring, whereas the organisms that made "this was bad" decisions either perish or fail to procreate - or are simply crowded out by others that made better decisions. We can't see that particular spreadsheet, but it stretches over millions of years and trillions of organisms and is too enormous for us to comprehend. And we remain incredibly tightly bound to and by it, as much as we like to pretend that we aren't.

Why do spiders instinctively make webs? Because it allows them to more easily catch prey, which is a massive evolutionary advantage.
Why do humans instinctively avoid spiders? Because our ancestors who avoided spiders survived, while the others got bitten by spiders and died.
Some people can overcome that aversion, but it will remain the default mode of operation for any and all humans for thousands of years, until evolution catches up to the fact that it's no longer as useful an advantage as other aspects.

Ultimately, while evolution is the reason we are here and sapient, it's become a cage that traps us. It moves too slowly to accommodate the changes occurring in our world, and the changes that need to occur.

 

[AusWolf]

Incorrect. Evolution is essentially an incredibly lengthy optimisation process, whereby the organisms that make "this worked" decisions live on and pass the results of those decisions to their offspring, whereas the organisms that made "this was bad" decisions either perish or fail to procreate - or are simply crowded out by others that made better decisions. We can't see that particular spreadsheet, but it stretches over millions of years and trillions of organisms and is too enormous for us to comprehend. And we remain incredibly tightly bound to and by it, as much as we like to pretend that we aren't.

Why do spiders instinctively make webs? Because it allows them to more easily catch prey, which is a massive evolutionary advantage.
Why do humans instinctively avoid spiders? Because our ancestors who avoided spiders survived, while the others got bitten by spiders and died.
Some people can overcome that aversion, but it will remain the default mode of operation for any and all humans for thousands of years, until evolution catches up to the fact that it's no longer as useful an advantage as other aspects.

Ultimately, while evolution is the reason we are here and sapient, it's become a cage that traps us. It moves too slowly to accommodate the changes occurring in our world, and the changes that need to occur.

You're right about evolution - I just thought you were talking about humans as individuals. These are two entirely different subjects, and work entirely differently, I believe.

 

[thesmokingman]

These are separate companies tied only by the CEO. I await your "proof" eagerly. The rest of your post is too ironic to warrant much discussion.


I really would like to start here. I don't troll.

This just shows you have limited knowledge of AI and are stuck on defining it by code and sadly not much different than deniers. Also, no one said there isn't code in my reference to v12, there's just very little of it especially compared to v11.

Proof of what? That Tesla and SpaceX share resources? Elon has shared many times how the companies are synergistic. It's clear even if the proof smacked you in the face you wouldn't believe it. And the joke is that here I'm talking about Tesla/SpaceX and you come back ChatGPT a text/digit based NN where Tesla/SpaceX is VISION.

 

[R-T-B]

This just shows you have limited knowledge of AI and are stuck on defining it by code and sadly not much different than deniers. Also, no one said there isn't code in my reference to v12, there's just very little of it especially compared to v11.

Proof of what? That Tesla and SpaceX share resources? Elon has shared many times how the companies are synergistic. It's clear even if the proof smacked you in the face you wouldn't believe it. And the joke is that here I'm talking about Tesla/SpaceX and you come back ChatGPT a text/digit based NN where Tesla/SpaceX is VISION.

So you got nothing and are insisting I don't know anything about this while bringing nothing to the table, while I provided literal custom coded examples of what AI is poor at and others provided direct links supporting my claims on how it works.

Cool. Now I see the troll. If you seriously want to proceed I was asking for proof of any AI in present tech that does not use tables for training.

PS: Btw, Dall-E works using the same tables and is visual, check the source.

 

[thesmokingman]

So you got nothing and are insisting I don't know anything about this while bringing nothing to the table, while I provided literal custom coded examples of what AI is poor at and others provided direct links supporting my claims on how it works.

Cool. Now I see the troll. If you seriously want to proceed I was asking for proof of any AI in present tech that does not use tables for training.

You actually haven't shown proof of anything instead just your opinion that all AI is, is simple tables and code. You keep using troll tactics of ignoring everything and pin pointing portions that you think you have a basis to attack, like out of my whole post you single "no tables" instead of the point of the post. I have linked and suggested where you can find the data I refer to and you don't look it up cuz if you did you wouldn't keep on this tactic.. I figured someone whose interested in the topic would at least google what I referred to but nope.

https://twitter.com/i/web/status/1695247110030119054

^^original feed

^^rehosted

 

[mechtech]

Give Siri ai and my voice so it can have conversations with my wife so I can do other things.

 

[Count von Schwalbe]

Give Siri ai and my voice so it can have conversations with my wife so I can do other things.

Just train a GPT and a voice AI model, and feed one into the other.

That would actually be a relatively low effort project for someone who is experienced in writing such code.

This has a lot to do with why I voted "Worse".

 

[R-T-B]

You actually haven't shown proof of anything instead just your opinion that all AI is, is simple tables and code.

Others have posted evidence supporting my claims, I provided literal examples, what more do you want? I have a hard time citing this because it's so basic to understanding how AI works dude, no one even assumes someone would argue it. It's essentially like contesting the order of letters in the alphabet, or the way numbers work in addition, how on earth am I supposed to provide "proof" of that? It just is. Others have done better than that but frankly, you are just not going to believe this for whatever reason, so yeah, I'm not going to put a huge amount of effort into it.

There may someday be an AI that is not based on an LLM and tables but we aren't there yet.

PS: I want something mentioning a lack of tables specifically, none of what you linked claims that.

 

[thesmokingman]

Others have posted evidence supporting my claims, what more do you want? I have a hard time citing this because it's so basic to understanding how AI works dude, no one even assumes someone would argue it. It's essentially like contesting the order of letters in the alphabet, or the way numbers work in addition, how on earth am I supposed to provide "proof" of that? It just is. Others have done better than that but frankly, you are just not going to believe this for whatever reason, so yeah, I'm not going to put a huge amount of effort into it.

There may someday be an AI that is not based on an LLM and tables but we aren't there yet.

They haven't posted proof of anything but the same lack of knowledge in the space. And no surprise you again ignore the evidence in front of you.

 

[R-T-B]

They haven't posted proof of anything but the same lack of knowledge in the space. And no surprise you again ignore the evidence in front of you.

Dude, you honestly think anyone is going to look at this convo and believe that?

You haven't posted evidence, you have posted PR videos from youtube and X.

Quit making a fool of yourself. I'm out.

 

[A Computer Guy]

I selected other. Like other advanced technologies it depends on who is using them and for what purpose. For example governments can build AI tools to improve traffic patterns or city layouts or they can use AI depending on how they feel this week or the next to suppress or eliminate political enemies standing in the way of their financial or advancement in power structures. Once the genie is out of the bottle it's up to humanity to quickly come to terms with it and hopefully make it count for the most good and least harmful impact to society.

 

[R0H1T]

Evolution is essentially an incredibly lengthy optimisation process, whereby the organisms that make "this worked" decisions live on and pass the results of those decisions to their offspring, whereas the organisms that made "this was bad" decisions either perish or fail to procreate - or are simply crowded out by others that made better decisions. We can't see that particular spreadsheet, but it stretches over millions of years and trillions of organisms and is too enormous for us to comprehend. And we remain incredibly tightly bound to and by it, as much as we like to pretend that we aren't.

And you can have human "evolution" within a single lifetime. I;m talking about how an individual works & behaves, not necessarily walking on 2 feet from 4 probably over a period of thousands of years.

Also if passing on the results is true why are we still living in an era of endless conflict, yes let's forget what Israel's been doing in Gaza for decades

 

[A Computer Guy]

Also if passing on the results is true why are we still living in an era of endless conflict, *****************************************

Human motivations (logical or irrational) can run contrary to ideas of morals or ethics that might otherwise block paths of conflict.
If AI is to become truly intelligent then it needs to be able to construct personal and meaningful imperatives to itself as an entity, however then can you really say that is still artificial once it can do that?

 

[R0H1T]

On the contrary, I think human motivations have been consistent across millennia ~ power & the pursuit of it. Evolution on the other hand, probably wrong(?) on that, to me would suggest that the collective is more important & that one individual's interests shouldn't trump that of the society we live in yet here we are!

 

[natr0n]

Ai is good for games/graphics/arts and such.

I don't give much care for it other than that.

 

[lilhasselhoffer]

So you got nothing and are insisting I don't know anything about this while bringing nothing to the table, while I provided literal custom coded examples of what AI is poor at and others provided direct links supporting my claims on how it works.

Cool. Now I see the troll. If you seriously want to proceed I was asking for proof of any AI in present tech that does not use tables for training.

PS: Btw, Dall-E works using the same tables and is visual, check the source.

So...I really don't want to spark this off again, but I need to say something.

Let's assume for just a moment that Tesla has made an AI. Let's also let that AI be their self driving mechanism. I'll start by qualifying some things:
1) The AI cannot account for variation in the roads due to construction...so it fails the test of intelligence there.
2) The AI has been documented to just go silly occasionally...implying that the learning structures can be side tracked by many things.
3) The AI is legally required to have someone watching it...so it's dumber than most people who learn to drive a semi professionally.


What could account for such an AI? Well, it's not an AI. Instead of learning it's a complicated series of lookups, which has been "trained" to weight certain responses against other responses. For instance, your forward ultrasonic emitter detects an obstruction. It doesn't know what it is, but because it's relatively maintaining distance and you are moving at speed, it can allow you to follow that sensor reading without applying brakes. Cool. It's now something that humans have had since the early days of mining.
But that's not all....it can slow you down. Yes...if the input for the sensor is decreasing by a certain value it knows something is in front of it...but only that it's decreasing in value so the brake should be applied. Cool...your average 2 year old knows this when the start walking and discover that running into a wall hurts and thus should not be repeated.
But...it's more complicated than that. It must be, because Musk isn't completely full of crap....right? Well, no. Let me go with the crap example. Your toilet bowl is about as intelligent as your car...just mechanically based rather than electronic. It's got a float, that in air is heavier than the surrounding medium. As such, it pulls open a valve and allows water to fill into the tank. Once the water fills, the float is less dense, and instead of applying a gravitational force the bouyancy actually applies an upward force. This closes the valve when the tank fills...making sure that the tank doesn't overfill. Is your toilet an AI...or is it just an I/O system designed to take in data and respond to that very narrow range of values?


Alternatively, let me ask you something simple. An AI by its nature can learn like a human. It is an artificial intelligence...but intelligence. Take your AI for the rocket, and ask it to process how to make a peanut butter and jelly sandwich...without significant rules. Now...what do you get? Nothing. Funny that...your average five year old can make a sandwich...but because your "AI" isn't strictly inside a situation with pass/fail conditions and measurable outcomes it cannot do the thing. Read up on a dead simple version of it here: PB&J
This is why I'm only going to be afraid of AI if it gets to the point of being able to make me a sandwich without comparing a thousand different good examples and iterating until it gets something viable. It's why I laugh when I see people create an "AI" for racing games...that takes 10 runs to not simply drive off the track...and a thousand more to understand that slamming into a wall is not a good strategy. That type of "AI" is not intelligent, but iterative and scaled responses to stimulus. That's not something to be afraid of...as long as you can introduce any stimulus that it hasn't been trained on yet. Being fair, Star Trek knew this in TOS era. They proved it when their "thinking machines" were destroyed by a simple logical paradox...where a true intelligence is barely even slowed by it. To quote a famous saying, "that does not compute." This basic ability to do more than process and weight responses is what people often forget...and I think it's what you're trying to communicate.

That said, if people believe that all thought is that it's easy to pretend that AI is already here...because why not. If you believe this though then there's no such thing as free will...because our outputs are always defined by the inputs...and with enough understanding of a person you could adequately predict them 100% of the time. That is the goal of AI though....right? Always do the "right" thing with certain data....ignoring that the beauty of the I in AI is being able to respond to stimuli that haven't been trained for.

 

[mouacyk]

it's a giant mutating decision tree

 

[lilhasselhoffer]

After spending some more time on this, let me ask everyone who believes that AI is a simple look-up table how they drive a car. Let me walk through this...so you can understand that a look-up table is really not an entire picture.

You start the car. You listen, and everything sound fine. Cool. It's on, and you're ready to go.
You don't know your speed, your surroundings, or anything about the terrain. You activate your sensors, and approximate a volume of space around you.
You identify obstacles, road markers, and conditions.
You then engage the car, pressing down on the brake pedal and/or clutch to engage your car into drive.
You approximate the things around you in a volume, guessing at relative velocity and path. You then engage sensors constantly to monitor this.
You review the road conditions in front of you, constantly modifying your foot position to accelerate and match a desired condition.
You indicate any proposed changes in your current system, assuming you aren't a jerk. Think turn signals.

What kind of calculations are going on?
Relative distance visually, and change over time, to approximate velocity and position. This is everything surrounding the car....
Processing local rules and laws.
Processing motor control to get the desired outcome, based on feedback from external gauging.
Literally anything else. Listen to music, talk on the phone, etc...


The problem is look-up tables are a brute force method to deal with this. It's why in the 1950's we could describe the control systems required to have a robot car....but in 2023 they're still a joke. Heck, even the Tesla cars in the Vega loop are done by a driver...despite being a glorified train loop that would have been infinitely safer and cheaper if they'd installed a train track....faster too if the current limited speeds are still accurate to the release speeds. AI has to be able to do the look-up, process different kinds of data, control feedback systems, and do all of this as reliably as a human being. This is why what we have now is Large Language Model....because we've gotten it to do one thing well.
Show me an AI that can "Make me a ham sandwich, a PB&J, and grab me a juice box" as competently as the average five year old (synthesis of request, construction of incomplete data bits, determination of goal, and creation of a plan to get there) and I'll be scared and impressed. Right now given any known AI, a pair of 6-axis arms, all of the required tools (and any it asks for), and a kitchen stocked with the required items I'm not aware of a single AI that could beat a child...let alone beat them without first running those arms into literally everything and tearing the kitchen apart before it could iterate enough to kind of do the job. (note I did add more to this...because the sandwich making robots in the food industry that do exactly one thing exist...so it had to be complicated enough to not be done with a single complex specialized tool....despite a child using non-complex tools)