I agree with this. I don't think we are going to all lose our jobs tomorrow, but AI is game changing. I think the development of AI will be seen as similar to the invention of the internet. The technology seems really useful and has a lot of potential
From the perspective of a reader of AI generated content, I have one big beef - the responses it generates go on forever. Can AI be supplemented with a KEY TAKEAWAYS function that does like a reader's digest/cliff notes summation? Great, a machine kicks this crap out with no concern for the human that has to read it. But apply AI to summarizing what I have to wade through and then I"m all in. Until then, it's too much of a one-way street, and it's not going in my direction.
From the perspective of a reader of AI generated content, I have one big beef - the responses it generates go on forever. Can AI be supplemented with a KEY TAKEAWAYS function that does like a reader's digest/cliff notes summation? Great, a machine kicks this crap out with no concern for the human that has to read it. But apply AI to summarizing what I have to wade through and then I"m all in. Until then, it's too much of a one-way street, and it's not going in my direction.
You know you can tell AI how long the response should be, right? It will make it as long or short as you want
come on this is basic stuff. More proof that most of you don’t even know how to use AI, let alone understand it’s capabilities
From the perspective of a reader of AI generated content, I have one big beef - the responses it generates go on forever. Can AI be supplemented with a KEY TAKEAWAYS function that does like a reader's digest/cliff notes summation? Great, a machine kicks this crap out with no concern for the human that has to read it. But apply AI to summarizing what I have to wade through and then I"m all in. Until then, it's too much of a one-way street, and it's not going in my direction.
You know you can tell AI how long the response should be, right? It will make it as long or short as you want
come on this is basic stuff. More proof that most of you don’t even know how to use AI, let alone understand it’s capabilities
I don't use AI in terms of generating it. I use it because I find it on message boards like this one, and it is less than satisfying, to the extent that it is detected as obviously AI generated,
Everyone oohing and aahing over AI are people who have never programmed. You are not seeing anything special or creative.
Those text to picture apps, for example, are just some programmer's idea of how to turn text into a picture. The AI creates nothing. Spend enough time at it (won't take long) and you'll suss out all the algorithms.
I'm not on the AI bandwagon, but this post is incredibly wrong. Suss out the algorithms?
The program is basically giant matrix of trillions of numbers that was evolved over time to give accurate answers. No programmer designed it, and you're never going to figure out everything about it-- it's not just some crummy simple recipe that a person made up.
Why are people upvoting this garbage?
Not going to read your arrogant rant, but I'll bold face your own idiocy for you. What kind of picture is an "accurate answer?"
Computers have parsed text for a long time. From that parsed text, the programmer then decides what the AI will do with it, what to draw, what style of art to pick. There is no question of "accurate," it's a set of instructions.
I'm not on the AI bandwagon, but this post is incredibly wrong. Suss out the algorithms?
The program is basically giant matrix of trillions of numbers that was evolved over time to give accurate answers. No programmer designed it, and you're never going to figure out everything about it-- it's not just some crummy simple recipe that a person made up.
Why are people upvoting this garbage?
Not going to read your arrogant rant, but I'll bold face your own idiocy for you. What kind of picture is an "accurate answer?"
Computers have parsed text for a long time. From that parsed text, the programmer then decides what the AI will do with it, what to draw, what style of art to pick. There is no question of "accurate," it's a set of instructions.
Before anyone answers this post, I think you should tell us how you "sussed out the algorithm" of Open AI's 50 terrabyte LLM models.
Machine learning is only as good as the training data. Conveniently, most consumer facing companies simply ignore the existence of adversarial data. Your car with self driving features sees a stop sign and stops. But if you painted that sign green, your car might drive full speed through the intersection. Because there were no green signs in the training set, it doesn't recognize it as a stop sign.
Algorithms are much more robust when written by humans, because you are able to intuit what *might* happen. You could have an algorithm that has 100% accuracy on the training data, but when you introduce additional data that breaks from the mold, it fails 100% of the time. In the stop sign example, you might give weight to a sign being red, octagonal, saying stop, being on a single stake, not attached to anything. This means that you might be too general to always recognize a stop sign, but that you can perform better against nonstandard signs.
Machine learning is only as good as the training data. Conveniently, most consumer facing companies simply ignore the existence of adversarial data. Your car with self driving features sees a stop sign and stops. But if you painted that sign green, your car might drive full speed through the intersection. Because there were no green signs in the training set, it doesn't recognize it as a stop sign.
Algorithms are much more robust when written by humans, because you are able to intuit what *might* happen. You could have an algorithm that has 100% accuracy on the training data, but when you introduce additional data that breaks from the mold, it fails 100% of the time. In the stop sign example, you might give weight to a sign being red, octagonal, saying stop, being on a single stake, not attached to anything. This means that you might be too general to always recognize a stop sign, but that you can perform better against nonstandard signs.
you gotta have good pure data sets and boost with synthetic data sets and the lastest u-net algorithms to label and segment images well. AI hardware doesn't fire the parameters in a true parallel, simultaneous manner. that's the biggest weakness. even the smallest mammals have 2 billion neurons and today's fastest consumer cuda is many orders of magnitude slower.
I'm not on the AI bandwagon, but this post is incredibly wrong. Suss out the algorithms?
The program is basically giant matrix of trillions of numbers that was evolved over time to give accurate answers. No programmer designed it, and you're never going to figure out everything about it-- it's not just some crummy simple recipe that a person made up.
Why are people upvoting this garbage?
Not going to read your arrogant rant, but I'll bold face your own idiocy for you. What kind of picture is an "accurate answer?"
Computers have parsed text for a long time. From that parsed text, the programmer then decides what the AI will do with it, what to draw, what style of art to pick. There is no question of "accurate," it's a set of instructions.
Dude, go read the wikipedia page on Neural Networks.
If you're specifically wondering about image generators, read about Dall-E and transformers and stable diffusion. There is not a programmer who has decided what to do. As I said before, there are gigantic matrices of numbers that were evolved to match up sentences and images accurately.
And, yes I mean to use that word: there is a training collection of images and descriptions and the matrices are adjusted over time to detect the relationship between descriptions and images-- they are scored on how closely they match them, and iteratively improved.
I'm a hardware engineer who hacks up AI for fun. Transformers came out of a USC paper and is implemented by Pytorch and Tensorflow which is used by virtually every deep learning system i've ever used.
The human brain has 90 Billion Neurons and 120 Trillion Parameters. Commercial AI chips have at least 4 orders of magnitude to go. That's do-able. But by what date?
It can create images? So what. When I tell Gemini to show me a picture of a pope, I don't get why it creates a picture of a black pope instead of just sending me a link to a picture of a real pope.
The other day I asked it to give me the margin of victory in the men's 1500 at the Olympics since 2000. It 'impressively' did it in 5 seconds except it just made stuff up and had Mehdi Baala winning silver in 2000 and 2004 - when he never won any.
So it clearly can't be trusted. It thinks there are black popes and thinks Medhi Baala is a 2-time Olympic silver medallist.
Exactly. Instead of solving world hunger or curing cancer (which it already should have done given how much hype it’s generated), we’re stuck with it doing stuff that no one really asked it to do in the first place.
Generics such as Google Gemini, Chat GPT, Apple Ferret, etc. are provided to the public for widespread use.
If you want a specific system to work for your application, say pickleball, you'll need a team of extreme AI experts to piece the open source together, train and test it for you.
You'll need to regularly repeat the process as algorithms, hardware, neural networks, data sets, etc. evolve.
I agree that certain types of video projects still require a human team to produce - for now.
But look at the rate of advancement in the last year. in 2023, AI-generated videos were hilariously bad. Now, they're scary good. In terms of making AI videos indistinguishable from real videos, we're already 99% there.
Given the way AI models are built + the current rate of advancement, it's very likely that we'll achieve 100% in just a few months.
To your example of testimonials: If I'm a business owner and want to do a video case study on client, and AI can write the script and make the video perfectly, for free - why would I pay a production agency $1000s to do it?
The client gives me permission to use their name and likeness, approves some canned lines, I give the prompt to AI and it generates an inspiring script + video in seconds.
I think you’re missing the big picture as well. No one is comfortable having an AI create a likeness of them. Even in a positive light. It would only reach the level of humor or a joke. but someone who spent years to build a reputation would not want a computer program to risk their reputation.
plus people like talking themselves. Heck most customers are reticent to sign off on a written testimonial. What makes you think they’d be ok with us creating a fake video of them speaking?
that assumes that sora is even good enough to create human scenes. The ones I saw of Nigeria were weird. It got peoples arms all weird. And we’re not even talking about someone’s lips moving.
the human face has hundreds (thousands of muscles) in the face. Animators around the world have tried to recreate a realistic 3D representation of characters speaking and it is so insanely difficult that they have to use real actors with sensors attached to their face in order to get the correct motion.
AI can’t do that without the same data inputs, therefore it’s not really creating any efficiencies.
Look at today’s cartoon and compare them to Disney in the 1950s. Today is cartoon crappy, same thing in the movies. In the 1970s Purdy wrote “Computerized Running Schedules.” Technology driven financial plans were the rage the previous decade. Canned is the best way to describe any of this. The danger is the biases of programmers can influence outcomes. Think about the most fantastic theories, social or otherwise, pushed as a norm. Largely nonsense.
It can create images? So what. When I tell Gemini to show me a picture of a pope, I don't get why it creates a picture of a black pope instead of just sending me a link to a picture of a real pope.
The other day I asked it to give me the margin of victory in the men's 1500 at the Olympics since 2000. It 'impressively' did it in 5 seconds except it just made stuff up and had Mehdi Baala winning silver in 2000 and 2004 - when he never won any.
So it clearly can't be trusted. It thinks there are black popes and thinks Medhi Baala is a 2-time Olympic silver medallist.
One of my college teammates did an AI query on distance training theory. The output on Daniels and Igloi were in many ways superficial at best, misleading, or just plan wrong.
there are gigantic matrices of numbers that were evolved to match up sentences and images accurately.
Oh, NUMBERS. MATRICES. Impressive!
Parsing sentences and matching them accurately to something else - e.g. sentences in another language - was a challenge until 10 or 20 years ago.
You seem proud that throwing a lot of flops at this can now translate words into a picture instead. And oblivious to the fact that how it draws the picture has nothing to do with that. That comes from the programmer's instructions.
Next you'll be saying, "oh yeah the AI spent 10 years analyzing and learning all the styles" but again, the programmer told it how to go about that.
It can create images? So what. When I tell Gemini to show me a picture of a pope, I don't get why it creates a picture of a black pope instead of just sending me a link to a picture of a real pope.
The other day I asked it to give me the margin of victory in the men's 1500 at the Olympics since 2000. It 'impressively' did it in 5 seconds except it just made stuff up and had Mehdi Baala winning silver in 2000 and 2004 - when he never won any.
So it clearly can't be trusted. It thinks there are black popes and thinks Medhi Baala is a 2-time Olympic silver medallist.
But you still don’t see the big problem coming with deep fakes and ‘hallucinations’; as the accuracy and power improves exponentially, and more trust is given, the proportionately small errors could wreck havoc as humans are removed from ‘the loop’.
OK, I did my own test and ChatGPT clearly failed. Here's what I did.
ME: write a paragraph in the style of Letsrun.com's Rojo
ChatGPT Alright, listen up, folks. It's time to cut through the noise and get real about what it takes to succeed in the unforgiving world of competitive running. We're not here to coddle you or pat you on the back for showing up. This is about raw talent, relentless training, and a hunger for victory that burns brighter than the midday sun. We've seen it all - the highs of triumph and the lows of defeat - and let me tell you, there's no shortcut to greatness. So if you want to toe the line with the best of them, you better be willing to put in the work, embrace the pain, and leave it all out on the track. This is Let's Run, where excuses go to die and champions are made.
But you still don’t see the big problem coming with deep fakes and ‘hallucinations’; as the accuracy and power improves exponentially, and more trust is given, the proportionately small errors could wreck havoc as humans are removed from ‘the loop’.
OK, I did my own test and ChatGPT clearly failed. Here's what I did.
ME: write a paragraph in the style of Letsrun.com's Rojo
ChatGPT Alright, listen up, folks. It's time to cut through the noise and get real about what it takes to succeed in the unforgiving world of competitive running. We're not here to coddle you or pat you on the back for showing up. This is about raw talent, relentless training, and a hunger for victory that burns brighter than the midday sun. We've seen it all - the highs of triumph and the lows of defeat - and let me tell you, there's no shortcut to greatness. So if you want to toe the line with the best of them, you better be willing to put in the work, embrace the pain, and leave it all out on the track. This is Let's Run, where excuses go to die and champions are made.
There seem to be two types of people in the "AI doubter" camp:
1. Educated people who don't really understand how AI works or aren't aware of how quickly it is advancing.
2. Uneducated morons who aren't able to reason cause and effect properly or have any sort of grasp at the scale of the changes happening.
I used to be a doubter, but not anymore.
AI is progressing faster than even the evangelists predicted. With OpenAI's latest release (Sora), you can create any kind of movie or video with a simple text prompt. Just whatever absurd premise you can dream up, it generates a video in any style you like. Or just normal things, like drone shots of a city or coastline.
AI basically killed videographers, drone companies, animators, video editors, photographers, graphic designers and a bunch of related careers overnight. There's no future in those careers; they have about 6-12 months left of viability.
AI is getting very close to being able to solve abstract problems without using the material is was trained on. This is the key point. Soon it will longer relies on data inputs from humans, it will be able to reason on its own.
Lawyers, healthcare technicians, data analysts, accountants, finance workers, copywriters, people working in marketing are next. Better start preparing.
If you disagree, you're not paying attention.
Stop panicking. Your hubris to think you know how still will play out is showing. Relax.
Any evidence that it's close to reasoning on its own?
Current iteration of AI is amazing, but they are just very sophisticated pattern-recognition machines, nothing more. ChatGPT is predicting the next most likely word given the prompt, and it does this on a loop. It truly is amazing some of the stuff it can produce, but this path won't ever lead to true reasoning capability.
Help us build the best running shoe review site for a chance to win a LetsRun t-shirt.Help us build the best running shoe review site for a chance to win one of 10 LetsRun t-shirts.