My Ais are all real though of course

"AI is a glorified parrot." ...gotta love it.

In the 80's/90's everything was Turbo, later everything was HD and now everything is AI 😂

Training AI with AI data is like incest.

This is what I’ve been saying, people don’t realize that the current Ai doesn’t have any real understanding of what it’s talking about, and more training isn’t going to solve it

Wasn’t expecting Miko to be the voice of reason on the subject. This is the first video I’ve ever seen that actually tells it like it is.

This is a tank. This is not. This is a tank. This is not.This is a tank. This is not.This is a tank. This is not.This is a tank. This is not. Suddenly the AI could identify tanks. Brilliant! Except that it turned out what had happened was that the tank pics were taken on a bright day, so all it was was that the AI looked at how dark the pic was.

Noble prize physics winner, Sir. Roger Penrose once said 'intelligence' is not the same as 'computation'.

From what I've seen from consumer AI tools is that they do not make decisions. Every time an AI 'thinks' they are only referencing material previously fed to them. Consumer AI at the moment cannot differentiate, contextualise, choose, interpret or compartmentalise like a proper intelligence (like us). They can only fulfil criteria like a glorified search engine.

AI right now is just a big ass encyclopedia with a fancy search bar.

4:55 the picture of the people in the car is not made using AI its from a music video made by Chris Cunningham for aphex twins track window licker which came out in 1999

more often than not, what people call is more likely ten algorythms in a trench coat.

This new Miko exposing NWO and world banks for all of their corruption is best Miko.

When my son was born, I was told he would never get a driver's license, because all cars would be driverless by today. He just got his permit. Sad thing is, all this hype and lying that Silicon Valley does about their products have real life consequences. For example, going back to my driverless car example, we have a huge shortage of truck drivers right now because no one is getting a truck driving license, which is one of the things, though not the only issue, pushing inflation right now.

Come on guys, this is Elon Musk. He's a great promoter of modern technologies but his prediction are always too optimistic. Remember when 5 years ago he was promising that poeple will travel to Mars in 2-5 years?

he says next year because he has investors lined up

Elon also said we'll have full self driving in 2015 😂😂😂

I am more hype about the progress of neuro-sama than chatGpt

This is starting to remind me of the real housing bubble, instead of home buyers who can no longer pay their debts, there will be businessmen who do not have the profits to cover all the expenses incurred by AI. But their debts will surely be forgiven and they will pay even less taxes. But the banks...let's cross our fingers that another 2008 doesn't come back.

telling AI to synthesize data for another AI to learn is basically a blind mouse leading another blind mouse through an alley, but they both don't know they're blind.

Wow, your avatar has gotten so realistic

Fun part is how multiple companies are spending hundreds of millions to train their AIs to nothing but the same level als their competitors.

Ai and especially AGI is being sold as a human replacement which is why companies are salivating over it. It should be sold as a tool which produces content with questionable quality. However, that wouldn’t generate nearly as much funding compared to a selling it as magic black box of wonders.

We don't have enough energy for everyone to run their air conditioning in the summer at the same time. I don't know how we're supposed to all have electric cars and ai

But elon never lied in his entire life? 🥺

Love the mix of absolute insanity and knowledge bombs you've been dropping on us Miko!

AI can't drive a car as well as a 7 year old farm kid can drive a 1946 Cockshutt tractor down the road, to pull grandpa out of the rhubarb he got his 1986 chevy stuck in, day drinking.

I have never seen Miko so sane for 8 straight minutes.

Ai telling us AI is not real .... mad

It's the tech bubble all over again.

As much as I hate the hype train that is AI rn, I have to say, I've never had an experience equivalent to talking to gpt-4o, its makes obtaining basic information a lot quicker and easier in a decent amount of cases than a search engine, and i like that it is "open mic" so i can leave it on on the side and just ask questions like I wish siri and alexa did to begin with

thanks for the video. i've been trying to deliver the same massage across for ages. people seem don't really understand how it works, therefore don't understand what problems can it solve and what not. we need more videos like that from you or either other widely reach people like you. also, will take some symbols to say that you're awesome in what you do, keep it up 0:)

Describing it as a fancy parrot is a great example.

you dont have to be a software engineer to know that elon can't code sh*t

I introduced my dad to AI and he came to me a few weeks later suggesting we could be social media influencers. He told the AI some of his philosophies about existence, and it told him he has a really unique perspective and what he wrote is very interesting. It was copilot/bing, I had to break it to him that it's been programmed to tell everyone the same thing. Never heard him more deflated. He doesn't quite get that it's not AI as we know it from movies and tv shows.

AI is like the new sourdough trend - you can't avoid it.

elon's predictions are always very optimistic because they increase his stocks by making ppl think he will be the first to reach true AGI. he is so optimistic he named his line of robots optumus.

2:21 This is a tweet made for investors to get hyped

If ai needs real world data to train, wouldn't the solution be robots? Let them roam around, record and experience the world (let them see all those elephants from all the angles), then train on that data? Still the costs and power are not enough to do this, plus the human element, but a cool idea I think

"and you are made to believe you think what they are" Me being a computer engineer: 🧐

this has to be the most intelligent code miko video I've seen. I really love her typical brand of comedy, but this was refreshing.

This is a very nuanced articulation of the problems with the PUBLIC's perception of "AI".

I subscribed on this channel because I was fascinated about shifthing the border between entertainment and technology. And I'm so glad that Miko started to do not only entertainment video but some explainers. I hope it will be more about her technical side also.

"the hours of folly are counted twice daily, but ain't no wisdom o'clock" William Blake (paraphrasing)

Great video. I encounter so many people who are programming illiterate and think the language model AI we have are general AI that just need to be developed more and will become exponentially better. They don't understand that there are hard limits to this type of large language model AI and "more time" isn't going to magically bypass those limits. Then these computer illiterate laymen call you an "AI hater" or something when you try to explain the realities to them.

yo don't tell the venture capitalists

thanks miko, love this type of videos, didn't know you started this format, but I enjoyed listening you speak and teach interesting things

Ai right now is the intelligence of a human baby bruteforced with a ton of memory and computing power to appear more intelligent than it is

What's funny is that AI wasn't a bubble in 2022 and early 2023 as generative AI is actually valuable, but then they started pouring billions of dollars into it and now they've turned it into one.

I wonder if there's a sci-fi movie yet where AI machines rise up and conquer human civilization, then keep people in colonies to farm their creativity and collect new data.

4:54 I don’t want to get everyone too exited, but the image in the right is not AI. It is something far worse however.

at the 4:15 mark that's when AI scientists essentially discover inbreeding lmao

Someone's unleashed the inner Tech !

I like this format, please do more, thanks

Videos like this are great, you make it easy to understand

The pizza glue comment was actually a joke reply from Reddit that Google's AI grabbed! It's hilarious because it can't distinguish those types of posts from real posts when, as a human, it's super obvious. It's given me BAD data as well such as telling me I did NOT need to update my Bios to go from an old Ryzen 5 2600 to my new Ryzen 5 5600. However, the 5600 was NOT supported originally and did need a BIOS update, leading to so much frustration until I realized "oh, this is a Google AI answer, and it's terrible!" Upon REAL research, I found helpful info. So even training on data, the fact it can't figure out real vs fake vs bad data... and will make stuff up it doesn't understand (rather than just tell you it doesn't know) is VERY dangerous to society and especially when health is brought in to the mix, can be deadly. Google AI is by far the trashiest example of AI being forced at the top, and it's wild how much it's being pushed. AI is useful as you state, in the right environment/situation. People are abusing it for attention and money though.

It is not just where do we get more data than the internet: where do you get quality data? The internet is full of crap, and there is that old adage "garbage in, garbage out."

lmao 4:55 one of the examples of bad ai is actually a picture of Aphex Twin haha Richard D James would be proud

This was really interesting to listen to! Thanks Miko! 😁

First time watching your video. And the AI incest remark made me subscribe 😂

Robert Penrose suggested neural nets was the key to AI because they offer an approach to learning that resembles the human brain. Neural nets were not feasible until we created gpus. Large language models showed emergent behavior when you feed a Neural net 'a lot of data'. To know what a zebra is, it had to learn what an animal was, what mammal is, etc. It's very different than simply echoing what is in a dataset. The main difference in the human brain vs GPUs, is the number of neurons.

Good point on A.I. A few lawyers learnwd the hard way, they used A.I. to pull from existing cases to use in court, the A.I. made up some cases that never occured. Also lots of the Art A.I. get free reign to use any creations and people still use them. I say it is better to learn to draw, program or do research oneself.

in before she realises she was wrong after robots powered by ai take over the world

Currently, there is no true AI - just a cut & paste machine learning experience. Even the wonderful Mikoverse characters essentially use learnt responses and prompts when engaging in conversation. It's clever, don't get me wrong, but it isn't true "sentient" AI, but it will come in the next 5 to 20 years

4:46 That's also the problem, I ran into when I tried to make an AI detect a Tetris game field. Everything was ok-ish for the average game field, but when something new or unexpected happened like a miss-drop, it just couldn't "see" it properly, returning garbage. There were no gaps expected, because that's not how a Tetris field looks in most cases, by a decent player and it's optimizing for the majority of cases. And when I tried to teach it with actually randomized fields that also resulted in problems. Truth to be told, I probably would have been faster, writing my own detection algorithm without any AI involvement. But I still think it's a great assisting tool, it just can't work unsupervised.

love the new video format

"Miko" is some of the best entertainment on line; but Wisdom from Technician is gold & we need more of it！💖

great and informative. loving this new content of yours

love these videos Miko 💙

A photocopy of a photocopy ends up in a blank sheet of paper. AI can't generate quality intelligence. You would need quality control programming based on a series of well defined variables, or you'll end up with Black George Washington.

love this style of videos

No one is buying iphone's or PC's, so big tech needed something new to raise their stock price. That's all AI is.

Thank you for saying it. Current AI is just fancy pattern recognition using a lot of data and a lot of calculations. It does not understand what anything means or is. Current LLMs are just existing machine learning models taken to the extreme with huge model sizes and large computing resources. LLMs are predictive text scaled up massively and nothing more. We are no closer to AGI than we were 10 or 20 years ago and I would think that it is impossible for current machine learning techniques to be used for an AGI due to inherent limitations in the models. To make an AGI we need to come up with a different way of doing things rather than just expanding the models and throwing more compute power at it. We currently don’t even really understand how our brains work either or what in needed for actual intelligence so trying to make an AGI without knowing that seems like building something with a non existent foundation. Even a human baby is much more complex and impressive than any current AI, sure development is slow but they are relatively quick and easy to teach and can learn and problem solve on their own. Other animals are good examples too.

I'm always impressed with your work, so much to learn from you. I'm glad we have you to drag us kicking and screaming into the technological future. Many may not know that you get regular lessons on how to develop your vision with AI tools.

Ai is in it’s toddler phase and as time goes by it will only mature exponentially into a full grown adult where you can’t distinguish something that is real or Ai generated

>implying AI training can't get more power efficient over time

A pocket calculator is smarter than humans, when it comes to arithmetic or remembering a large number of big numbers, if it has a memory function. Defining what is intelligent is pretty difficult. It's always "we just need to figure out that one more thing, then we have AGI" and then it turns out that thing is way not enough.

I didn't expect such a interesting video today, will share, thank you very much.

It feels like a solution looking for a problem, ai that can halluciate is less reliable than most encyclopedias.

Finally a video that properly talks about AI without either demonizing it or putting it on a pedestal. As you say, AI is just a bunch of statistical/probabilistic models that without data can't do much. I started out college as a CS major but eventually switched to applied maths, yet I still remember what my professor on one of my first classes said: "Computers are stupid". She is the reason why when I use things like chatgpt, regardless of the use, I first provide it with a lot of context and data before I go on to give it the task. Cuz I know without it it'll just spew nonsense. Just recently I asked it to help me write a function in python and it came up with this weird mathematical formula that I had never seen before which already rang a few bells in my mind since it seemed so simple and, according to it, was "an appliance of basic arithmetic properties". So I go on to try and prove it, not even 10s in I find a counterexample. It really infurates me when people talk about Ai as if it is anything more than a tool. And one that is barely in diapers still. What's even worse is that no amount of training will make a model better than what it is. There is a limit. However I do think AI can be improved a lot, we just need better mathematical models!

I like this kind of content. Keep it coming Miko.

@CodeMiko, there was at one point a craze about the "blockchain", to the point that one ice tea brand renamed themselves to Arizona Blockchain Ice-Tea (or something similar with the word blockchain in there) just because their stock was bumped up because of that word. You are most likely correct when you say that AI is just another marketing buzzword. I mean, when ChatGPT got into the mainstream, all the companies started using "AI" or :Artificial Intelligence" in their commercials, slogans, product pages, and on and on... Great video. Informative, to the point, with an opinion that doesn't go both ways.

Finally, someone who understands that what we call AI these days is just an overhyped marketing term.

Saying VC capitalist is like saying ATM machine

Not gonna lie! This was super informative would watch more videos like this!

the problem is we dont fully understand what "intelligence" is, but its definitely not just regurgitating data.

"Ai will bust because it's not good at making new images from nothing." A year ago AI was eating noodles in such a horrifying manner and now it looks like it's eating food, in a manner that is hard to tell. I doubt we will get self learning, how ever it will take most jobs, Mental before Physical. In fact it already have taken jobs away.

Almost all new tech follows a very similar S curve with how advancement occurs. There is a slow build up in the research phase when no real world (monetizable) applications are shown and this phase can last decades. When the technology is shown to have provable real world applications and market investment starts there is a very very rapid take off period with seeming unlimited growth. This cannot last, once all of the "easy" innovations have been exhausted you reach the point where it becomes increasingly more and more expensive for the next big breakthrough to occur. This is where funding stops as it no longer proves profitible to strive for more, it is much easier to settle with small incremental improvements. There is always a point with a new technology where it looks like there is infinite potential to it but the realities of business always step in when companies will start losing money by pusing for innovation.

Just a slight correction, the time Gemini answered stuff like using glue to stop cheese falling off pizza and telling people to eat rocks as part of their dietary mineral requirements wasn't because it was missing data its because it couldn't tell when something was a joke or a "bad" source of data

"AI is like a glorified parrot". The perfect woman doesn't exis.....

I laughed way too hard at AI incest. It definitely does not turn out well.

Really love these videos because you're an amazing and intelligent person! Not gonna lie the vtuber stuff/Miko verse can be cool at times but these are important videos, and I really prefer them from someone with your level of expertise in these subjects.

4:54 This is probably a meme image. On the right you can see an image from the Windowlicker album from Aphex Twin, where the Richard D. James' face is superimposed on the faces of the people.

I don’t know why these companies think they just need more data. If the entire internet isn’t enough data for your model to learn from then it’s the model that needs to change not the amount of data.

I can only imagine that AI feeding on AI will at the very least lead to everything looking samey and at worst some sort of exponential growth of its mistakes as wrong data keeps being shoveled in and processed into more mistakes.

Every time I see a headline about AI, I always think it's referring to a dude's name.

AI is good enough to replace humans with, especially positions relating to western art looking forward to actual artists coming out, making original content, now that the greedy occidental ones (who let social media/profit dictate their entire output) will lose their jobs

There's a lot grifters riding on the hype, but the potential actually is there for people that actually know what they're doing to indeed deliver on the hype; problem is, investors don't really care about results, they just wanna grab growing bags and drop them on idiots before they shrink. You should look deeper into the research and actual latest news on the field instead of the mainstream bullshit and xit posts of people that throw logic out of the window because they're freaking out about losing their jobs. edit: Fixed sleep deprivation typo

The thing I love about AI is stuff like Perplexity. That's where AI will be amazing. AI will never be Sky Net, not with the technology people are using today, and not with an upgraded version of it.

There is so many misconceptions in this video. But I get you, when I started with AI couple years back, my understanding was very similar to yours. You are talking about AI as it is all the same, but machine learning like reinforcement learning are very different algorithms than transformers and diffusions and RNN networks, etc. I had this realization when I first worked with base models of GPT3, which wasn't trained with RLHF methods like ChatGPT is.. and what I realized back then were so called emergent properties, like the ability to create new words that phonetically make sense (but don't exist). Or when the AI had zero or very limited language data set (just english), but still was able to converse in other languages with limitations.. or when it is able to generate SVG images and with each generation it gets more precise. These are things that improve not just based on the amount of the data, but more likely on the amount of training and parameter size. Nowdays we call it AI features, but basically it's learned concepts.. yes LLMs are just next token predictors in some sense, but also partially feature learning machines which is the base for smart general AI system that should be able to improve infinitely. So you can train on a very limited subset of the world, but get almost the full representation of the world by filling in the knowledge gaps, which AI does during the training. We already do that as GPT4 wasn't trained on that much data as you think (I believe they've talked about barely 10% of the internet was scraped - I don't know where did you get your numbers - it's not documented anywhere afaik). It's like CISC vs RISC .. you don't need full instruction set (a lot of data) to achieve higher complexities. And this is not all.. something I've realized from my experiments is that teaching the AI via RLHF and creating the dataset will be fully AI automated soon from synthetic outputs and no it won't degenerate over time, because it can be setup to both increase and decrease the entropy of the system at any time and self-improve. Needed computation and energy will be offset by algorithm and optimization improvements and most importantly by new chips and science of computation like thermodynamic chips or optical quantum chips. Also, I know the study you mention about the AI progress not being exponential, not even linear, but worse.. well that's actually a false information imo, because intelligence can't be measured the same way as any other algorithm.. I mean, someone with an IQ of 150 like Einstein should be only 1.5x smarter than an average human by some means, but we all know that doesn't add up and it's much more complex than that. So each percent gained on AI benchmark is more valuable the closer to 100% it is imo, but I am not a mathematician, though I can see a logical falacy when I see one. And imo AGI is truly near.. I think it depends on the definition of it though, but it's likely that we will have first virtual AGI this year that will be able to do at least 80% of digital things we can do and the next year we will probably have an AGI both virtual and in robots that will do about 50% of what humans can do in real-life and in 2027-2029 it will mature and be perfected and by 2040 almost every person from a developed nation will likely have a robot at their home that will be able to do 98% of stuff we do. And by 2045 we might reach singularity and start living in half-simulation, half-reality type of AR/VR existence.