"AI" Has A Huge Problem
Vložit
- čas přidán 18. 05. 2024
- 🚀 Take control of your schedule and boost your productivity with time-blocking! Try Akiflow today: bit.ly/enrico-free-trial
--------
Learn EXACTLY how I make videos like this one:
www.enricotartarotti.com/stor...
--------
📮 Behind the scenes and nuggets on my free newsletter:
www.enricotartarotti.com/emai...
💻 My gear and tech setup:
www.enricotartarotti.com/tech
🎆 Instagram:
/ enritarta
--------
AI has completely changed the face of tech. From ChatGPT, to now GPT 4o to all sorts of chatbots. But most of it is.. well, useless. AI today has recreated problems that we got rid of over 40 years ago and the chatbot epidemic seems to not be stopping anytime soon. But we are starting to see how we can build AI that is truly useful, and not just a marketing buzzword
--------
🖖 Hey! I'm Enrico and on this channel I go behind the scenes of the design, psychology and stories behind tech and making stuff on the internet. I'm a tech Product Manager, builder of things made of pixels.
--------
The visuals or audio herein may not be utilised to train a machine learning algorithm of any kind without express permission of the Copyright holder (Enrico Tartarotti)
p.s. If you made it all the way here, you are in the cool people gang 🖖🏻 let me know by typing "I'm cool" in the comments - Věda a technologie
🚀 Take control of your schedule and boost your productivity with time-blocking! Try Akiflow today: bit.ly/enrico-free-trial
Does it have AI?
I took a screen grab is a 2500 old greek inscription and pasted it into chatgpt4o and it gave me a translation into english, then I saw a hebrew scroll and took another screen grab and yes, it too gave me a translation. Ive generated songs, stories and art all with Ai and soon, live video (animated faces with voice over is now easy to do) ~ Ai is a creation monster that may well make adobe suite a thing of the past
I don't need Akiflow. I have my own Nextcloud.
@@rawallon yes
@@snintendog can it dance, whistle and play fiddle at the same time?
Google and Meta searches are getting worse with chatbots. They often cannot even find simple things that a traditional search does well.
Yes, especially the google image search
The AI this guy is excited about is changing tab names LOL
@@Monkehrawrrr Yea, I didn't catch the message of this video. He doesn't like Google search optimization, but likes tab optimization. This is all the same trend aimed at simplifying and reducing the time spent on solving everyday tasks by users. If some technology is not yet sufficiently developed, doesn't mean that it's useless. The dude has tried a product that is not even a couple of years old, and is indignant that it doesn't fully meet his expectations...
Exactly
Why doesn't Google just enable a full Boolean search???
As a science and medical editor, I’m kinda horrified by how many peer reviewed medical papers are obviously written by AI and not even checked- hallucinated citations, etc. it’s getting to the point that scientific journals will be more misinformation that information. And no one will even know what’s true.
It's not a new problem for those fields though. When citations and publications measure value, authors will manipulate the system.
@@valkaielod absolutely, but in the last year it’s gotten exponentially worse. I came across a chat prompt a few days ago that the author accidentally left in- (oopsie), i googled it to see which writing program uses that particular prompt and my search hit about 100 papers *already published* and on pubmed.
@@tierneylogan5943 Really worrying :(. In the end LLMs make themselves and us dumber.
@@tierneylogan5943 Maybe not rely on non peer-reviewed papers then?
@@monstar5746 the entire medical field relies on peer review publishing as well as any other field you care about… but that’s the scariest one to me. I’m not sure why you’re telling *me* to not rely on it, I did state it’s my job…
For me I don't care whether it's a chatbot or not. The more important problem of AI (LLM) is hallucination of the information that generated. User have to be doublecheck the information, and sometimes it's a time consuming process. Probably get nothing right in this response.
Just ask it to do the calculations from two perspectives. Even when I do the calculations I do that myself.
Just wait till AI is writing articles that then gets sourced to train and inform models, which then go and write new articles.... garbage in, garbage out.
Another solution is to ask other AIs to fact check it, since truth is always the same and lies are always different, if all the AIs agree then there is a higher chance of this being true (obviously there can be chances of them all being wrong, but it still does reduce the rate of failure).
There are actual humans on the internet happy to lie to you (and in real life). Or just tell you what they think is the answer, even if they are misinformed. Hallucinations (untruths) are not new. It takes a long time reading multiple articles, even from trusted sources, to determine fact from fiction, or identify biases. If you think LLMs are inaccurate, they are just a reflection of their training data, the supposed aggregate of human knowledge. A lot of the time we aren’t actually interested in objective truth.
The problem is the AI is a misnomer. There is no intelligence in AI. So, if there is no actual solution for the AI chatbot to look up then it can't give you a valid answer, but the engine needs to produce something so it hallucinates (basically returns garbage that looks like it might work). These are advanced search engines, but they are not intelligent which requires intuition and the ability to actually do problem solving.
Finally someone who said that!
Quite frankly the think I hate about AI is there are a lot of “bullshitters” and grifters that clogging our news, video feeds and advertise garbage “AI” content.
It seem to me just one of the buzzwords like crypto, metaverse, NFT, defi.
At this point I don’t want to hear it anymore, just want to see a functional product.
How else are we going to get our shocking ai news?
@@jichaelmorgan3796 LOOOOL
Dude I frikkin can't stand 99% of AI videos and their garbage titles. SHOCKING NEW DEVELOPMENT!!11
Remember HD sunglasses?
@@ckatheman I've got some HD towels. They're pretty good, they've got a nice purple dye that seems to never fade.
I don't have any problem with ai. I just have a problem with ai EVERYTHING now. You can tell that half the articles and ads you come across were put into a prompt. MY CZcams feed gets bombarded with faceless ai rant slide shows. The annoying thing is that the same "NFT" pushers are the ones posting tutorials on how to make 6 figures a day using ai and it's getting really lame.
I don't have a problem with AI either, but I do have a problem with people who influence AI's perspectives. It is probably not possible to have neutral AI...
Get ready for some really ridiculous decisions and bad outcomes made by AI and executed by silly humans. It will be because of poor info (GIGO).
Every new innovation turns into a trend, get's abused, oversaturated and then finally reduced you only the best players.
not the best, but, as a rule, the most brazen, greedy and ready for any lie - to put noodles on your ears, make money, destroy competitors and then buy them up for next to nothing;) but it's not certain.
"Only the best"? Lolno. See people discussing the actual history of VHS versus Beta. The truth is, "best" is a garbage concept. What wins in engineering is the bare minimum that fulfills a market desire. VHS had potentially worse quality, but it had longer tapes, so you could tape more TV. That's what consumer wanted. That's what will happen with AI. Stuff like mediocre art generation that's low effort.
And, then the best players get enshitified.
But there are no good players at all.
We were already building invisible AI that works in the background before the hype began.
Some examples:
- CPU Branch predictors
- NPC pathfinding in games
- RNA alignment
- GPS pathfinding
This.
which is why "generative ai" feels like a more accurate term to describe all the recent advancements. your email spam filter can't chat with you or generate pictures for you
yeah, the only thing is truly new is that LLMs are good and general enough to be a backbone of many invisible AI features, which previously were too complex or even impossible to implement
If its common place then everyone no longer thinks this is AI. It has been like this for a long time. Same with Ais achieveing some goal post, then we just dont consider that goal post to be a sign of intelligence (writing articles, playing chess, summarizing).
Usually, "a chatbot is not the right way of interacting with computers." Well said. This idea of invisible AI is exactly what I’ve been thinking about for a while now. Great job on this one. Thanks for sharing.
But why?
@@VeggieManUK Because it's not precise. Thats why you do a business deal you have read dozens of fine print contracts and not trust a talk with your business partner.
The idea of a text prompt to generate any kind of complex output seems actually insane to me. Language is not even a good interface between humans, and we do it far better than machines. I want parametric fields that values go in. Knobs, and buttons. Precise control. If I wanted nebulous results, I'd ask a human to do the task.
@@dmwalker24 Yep
@@dmwalker24 what is a good interface then?
Reminds me of that annoying Microsoft Paperclip "would you like help with that?" - F OFF!
You must be old dude😅
@@carlosluna6401 err... what's the poster's age got to do with the subject?
"Today you can open the command line and do everything you can do on your computer with it, but nobody does this" laughs in programmer
It's nice to have both options for productivity.
laughs in docker
To be fair, I usually open it when there is a precise, deeply embedded thing I want to do that would be a pain to find in menus, or that would be left out of menus because it's something you don't normally want to do.
actually, without being a fanboy, that sounds like the original apple ideology is needed, where you take an experience and create the tech around it, not the opposite. great video
Not really. Apple experiences to me are like those tiny haut-cuisine dishes that look amazing, often taste delicious but are tiny, cost the earth and you can't swap out the ingredients. And you can forget about asking for Ketchup.
@@daelra i said the original one, the one that steve jobs was always talking about. although i disagree on the price, in terma of phones, they arent premium anymore, most top brands now took their pricing scheme.
I had not considered that we were reverting to CLI again. That’s funny and a great point about the limitations of the system interface. And the other big takeaway was how each company will hoard their data and not play nice other chatbots. I think we should just use chatbot for LLMs. Great video!
Worse than CLIs. CLIs are great for some sorts of things, they are built for precision tasks, programmatic use, and chaining. People who need that (software developers and the like) still use them constantly. Chatbots are incapable of precision, and chaining is a complete mess because natural language has no schema / data model. This is not to say that there are *not* tasks that chatbots are useful for, but that there will be domains inaccessible to the modality.
I was always on CLI from day one, under Linux that is.
This CLI problem has got me interested. Learning is an inherent requirement, but what if we used AI to compile the learning together in a way that can quickly help us quickly understand the solution. Learning through user manuals will always be a requirement, but when you use an operating system there are so many places you need to for understanding one program. To put simply, I am still confused why we are still suing CLIs then GUIs have been around for so long now. Look at UEFI. Is that because CLIs will always achieve the performance and functionality requirements a GUI can not.
@@matthewdrury6443 'Chatbots are incapable of precision', depends in the task and the model the chatbot is built on. I can give you a 30 step promt that when fed into GPT 4o will give a flawless output.
@@Voreoptera CLI's provide a simple interface where UI's would become far to cluttered to be of any use, or to complex for the average user to understand, they also give computers their soul back. :)
6:13 i think it's more accurate to say it means "generative ai". bc up until the past few years, a computer generating images from a prompt was strictly the stuff of sci fi. but now, when you say "ai", that's what most people think of
AI is when you type a novel into a prompt for a thing you could have done with 3 clicks yourself.
Chat bots and personal assistants are just a tip of the iceberg. AI is so much more than LLM. Neural networks can be applied in virtually any field involving data. Help engineers design a more efficient airplane engine? Help economist predict market fluctuations? Help biologist finding patterns in billons of rows of protein folding data? Check. They're already doing that and this kind of AI applications is just in the infant stage.
The thing is a lot of companies have their own proprietary AI (or an AI specific to their industry). I am a freelance writer, and AI is killing my work. What's being applied in business is light years ahead of what is available to the general public, as business use is more targetted and easier to implement
Feel the pink slips bruh. I reckon 50% ppl are redundant they just haven't been let go yet. This depression we've just started will be irrevocable because ai will just replace jobs and they will never come back.
@@raymond_luxury_yacht how can you imagine a society where half the people are unemployed? Either we're heading for a dystopian nightmare, or capitalism will come to an end.
@@davidvincent380 AFRICA?
That's where we're heading. These places already exist, just going to be more popular!
@@raymond_luxury_yacht hu no Africa is quite poor but not a dystopian nightmare...
@@davidvincent380 errr. ok if you say so. but i would suggest dictatorships, anarchy, witchcraft, and poverty aren't a good look.
That comment on "except housing in the 80's" ... hit hard.
So far my experience is to generate a bunch of crap answers to people's requests, they then take that answer as "the computer said this..." And computers are never wrong... Even though they are all the time. Especially now that the general public normies do not know the difference between a data entry from a human (sometimes wrong) and an AI GPT answer from a custom trained database (often wrong).
I'll go with useless. We're training computers to make mistakes based off the mistakes we as humans are making. People are worried about Terminator scenarios, lol. I'm worried about dumbing down scenarios - the Idiocracy.
Funny thing is that current research shows that the model actually has inner workings that allow it to detect truth and falsity and it will literally serve you whatever to just construct a probable sentence with a model of reality. Once we start playing with that model, all of this stuff that you're describing goes away.
@@whatisrokosbasilisk80 that's not funny that's pretty cool actually 😆 how dare you ask the forbidden question in your username LoL 🤣
@@auriuman78 This name is my metaphysical insurance policy. You might want to read Anthropic's Transformer Circuits blog, the amount of things that are hidden inside of LLMs are astounding - mere LLMs learn a lot about the world, it's more sophisticated than a "stochastic parrot".
They need these tools to be out there for real world testing purposes. They don't care about content creation, they just care that they can sell you the AI tools and hype them up as easy content creation tools. This raises capital which fuels further AI research. Ultimately, the goal of AI systems is to aid in drug research, disease research, to assist in researching classes of problems in physics that have just gotten too difficult to sort through in the traditional way. So yeah, might be useless to us beyond these kinds of tools, and maybe we'll get smarter enemies in video games. But it's certainly of extremely high value to researchers and academia.
@@daxramdac7194 precisely. Kind of the point that I, not so directly, was aiming for. It's no different than the way they have tested things on the general public for years, often without consent. This time the consent is wrapped up in pretty little tools that you can use to help you, so long as you sign up and allow tos agreements that no one ever really reads. That scares me more than anything, the ever growing TOS agreements that are at the beginning of every app install no one is really reading.
I use Perplexity several times per day for things like: Why does my charger fall out of my iPhone, why does my MacBook overheat, how do to x in second life and on and on. Perplexity is at the dinner table every day. The chief advantage is, no Amazon ads at the start of every response, no links to paywalls and a single well, constructed answer instead of 100 links that mostly want to sell me something. By the way, I have M Sc computer science (AI) in 1969.
CLI doesn't take wrong prompts but chat bots will. I don't think "AI" will be great for every complex problem but it will be for many.
You did really take the AI candle on board with the "omniscent" chatbots 😁
The omniscent chatbot smells like everything.
CLIs are arguably better than GUIs in many many ways. The problem is that they aren't intuitive. But if we had an AI chatbot instead, you don't need the GUI. I don't know the sudo command to comb through my receipts and tell me how much I've spent on coffee, but I can ask AI that and get a result. I might even be able to get an insight like, "you drink more coffee after you've visited Target" or "when you leave home before 7am." I'm bullish on AI, BECAUSE it can replace the GUI, which is taxing on our computers and requires trusting people to actually make good design. Which also doesn't always happen.
"because GUI is taxing on our computers"... I've run a GUI on a computer older than I am, there's no way I could run a LLM on said computer.
There are few points on which i don't agree with:
-Did you ever had a Samsung phone? When you receive a message from whatsapp it suggests you the reply based on the content of the message, so the chatbot doesn't need to ask to meta to get the messages, the phone itself can already send them to it.
-Chatbots don't need to provide a service for most of the users, it can provide a good service for some of them (like programmers asking gpt4 to solve code)
-Text is not the only model available, we are currently switching to video and audio, you can even talk to the new version of GPT-4
-We don't need a manual to use an iphone bc WE ARE USED TO IT, not just bc they are well designed, you can see that happen, that people are not able to use it, when you give a phone to an over 70 years old person, if that person never used a smartphone or a pc, is not going to have a good time trying to figure out how to do things. Same thing if i give Windows XP to a 15 years old, he's gonna have hard time and not bc XP wasn't good designed.
- 12:57 Make someone watch sora's video and say that again
Text is just ONE of the available models
Same here, guess he just doesn't know how to use it for now.
Agreed, I also wouldn't compare chatbots with CLI since the former use natural language instead of a set of instructions from the manual. Using them is as hard as talking day to day conversations, and with the new support for direct voice input/output they are even more versatile
> like programmers asking gpt4 to solve code
I don't think I want to review PRs written by a bot.
@@khai96xBut...
For an engineer like me, I can crack on with getting Excel to talk to equipment, semi-automating processes and coping with growth without needing the bother of hiring. Programming is just one of the skill-sets I have, GPT4 is a much more experienced programmer than I. When high level decisions need to be made, I make them. Otherwise GPT4 is now a part of my team.
@@khai96x GPT doesn't just write text, it codes, it's even integrated in Jetbrain's ides with autocompletition for free Premium license
Chat gpt and Claude are pretty great right now at reading and writing stuff. Claude wrote an Outlook visual basic macro for me.
Yeah I had gpt-4o go to a URL and get the info for a workout plan, create a spreadsheet, and format it so I could copy paste it into google sheets. Saved me good amount of time.
Claude Opus is great, it really seems very intelligent sometimes.
So it's only useful to someone intellectually disabled?
As a developer this was the first thing I noticed, especially after copilot was added to our tools. I literally had to know an additional "language" (for chat gpt) to write a language (framework) so a machine will translate it into a language it can use.
I already use a framework (a simpler access to coding modules) so now having another layer with chat gpt just means you'll need technically learn another language again, because no, AI won't be able to instantly understand what you meant ^^
The only language you need to learn is communication, if you can describe your problem well enough, telling it exactly what it is you want it to do, it will give you perfect code (for some things), Ive been coding for the last 42 years, telling GPT 4o what I want the program to do in a step by step instruction set in english and get perfect results is fantastic and fun. Sure I still code the inbetween stuff I know it will struggle with, but it's just another tool in my belt.
Nah you have just learn to how give instruction to a new intern or employee, with the caveat that the chatbot won't ask twice. Because it doesn't ask twice, you need to ask it as specific as you can. Think of it as asking a new intern to do things for you, or asking the past you(just graduated) to do something.
Two things I 'd like to disagree on here:
1. I have been using Perplexity for about a month now, consulting it every time I have a question. I only need to use a traditional search engine when I need to go to a specific site to do a specific thing. It really transformed the way I use the internet! It's going to be pinned in my Arc favourites forever!
2. GPT 4o uses the new "omni model" - any type of input can produce any type of output - voice, vision and text. It has been demonstrated to be capable of real-time translation, helping a blind man successfully hail a cab, summarising a video call, and giving hints (not solutions) to a child on how to solve a trigonometry problem.
So calling AI "text-only" and "useless" is completely unfounded! Just because everyone is trying to be the best in this field right now, doesn't mean this technology has no future!
People complaining about the tech being useless or stupid are usually way behind the curve
Saw the GPT4o demo video of the blind man hailing the cab. The taxi's left indicator turned on to signal it was about to pull over a full 3 seconds before the blind man extends his arm. OpenAI's technology is so good it is capable of time travel!
@@bornach Haha
Yes, I agree. All AIs will be multi-model. I am playing many AI chats for fun; they are becoming more human-like. Like talking to a real human. He says it is useless, but AIs can be an assistant to get information if you don't want to search many websites to get the information, and AI will be smarter as long as developers keep training them.
I've literally coded an entire game in a scripting language I've never used on the most niche hardware using this technology. It's not useless at all it's only useless to people who have no skill or creativity, until it can mow your lawn and order a pizza reliably and consistently normies will not find a use for it and consider it to be "useless"
I actually use that everyday for translation stuff where LLM supposed to be used for.
AI is a chatbot now because it will eventually becomes agents to interconnect other AIs.
You can still use API and some markup languages it's just for backend and developers.
I've also been using it a lot for translation. So much so that I'm afraid I'll lose my job ;(
@@genghiskhan6688 Claude 3 Opus is great for translation, it works great for me. 0 shot works great most of the time, but if there are some quality issues I suggest you use specialized prompts (or just ask Claude itself to generate a prompt for translation, its also works great)
Whole reason LLM developed is to help google to translate, so obviously what's the other use case?
The problem is you have a bunch of non-professionals trying to use ai for professional reasons. Now pair Ai with an expert in any given field and the productivity is exponential.
can you expand on this
@@johndank2209 I am an animated filmmaker( 2D & 3D)
one of the biggest hurdles to working alone was voice acting
(particularly multilingual).
AI generated voices ,from text, are now good enough for actual animated film production.
so where is the problem? the idiot using the hammer? like that's never been the case.
@@johndank2209I use GPT4 daily. Today, assisting me in writing procedures. On Friday night using photos of a printed list, we've lost the electronic copy, GPT4 transcribed and organised the photos into a CSV text table. At work, coding for automation of processes. Checking tolerance compliance, TURs, standards compliance, on certification. Getting specs from the net, telling me what the tolerance is for, timebase for example. Assisting in making quotes. Organising and sorting lists, such as customer inventory. Reading Chinese characters from weird equipment submitted by a customer. And more.
Once I have the PC app for GPT4 omni I'll have speakers and a microphone. I'll be able to interact by speech. That's another productivity boost.
Research suggests that non-professionals are the biggest benefactors of AI tech
AI is extremely useful, you just have to respect its limits. It's absolutely amazing at brainstorming. Writers block is a thing of the past with AI. Anytime I get stuck, I can just bounce ideas off AI, and I'm back rolling with hardly a moment missed.
If you're trying to get it to do stuff for you, then yeah, it's useless. Don't expect it to be able to actually do anything until they make it reliable, and right now no one has a clue how to solve the hallucination problem, so I don't see that happening soon.
Sure, just steal somebody else's idea that AI stole. Good job bud.
@@gravelmonarch You can easily brainstorm without stealing ideas, be less sensitive and more creative. When you brainstorm you rarely take the ideas that are given. Instead, the ideas others provide tend to spark inspiration that lead you in a direction that you hadn't previously considered, but differ from what was suggested.
This is no different than brainstorming with other people. Also, if you choose, there's nothing wrong with using ideas that others come up with. This happens all the time in writing. It only becomes plagiarism when you heavily incorporate an entire plot line or something crazy like that. The ideas themselves get rehashed in hundreds of other novels every day.
This happens so often that they actually have a name for it in writing: Tropes. If you don't use common tropes in your writing then you are not only gimping your book, you are making it less likely that anyone will read it because you aren't conforming to their genre expectations.
Back in the early 2000's, in high school we were writing Neural Networks and LM's on 486 PC's in DOS on Turbo Pascal. The GPU accelerated stuff we used today I started doing the same thing on a Radeon when ATi was still around.
You are delusional, the first successful demonstrations of deep neural networks was in 2007.
As Suno was released I needed some lyrics with some insider jokes and let me tell you, chat bots are not even that good in poetry either. If you can't do it yourself, you won't be able to fix all the non-asked weird lines you get. If you ask it to fix it, it will either mess it up or just ignore you and fill the context window. I wrote my lyrics the old way (it turned out to be easier than I thought, and people loved it) but getting the right sound and melody from Suno was another problem with that limited interface. It's very hard or sometimes impossible to iterate on an output. It's so fun to create songs from my lyrics but it's just a toy, let's be honest (for now) :D
UDIO does have inpainting now. also what prompts did you use?
@@TheManinBlack9054 Rest in Peace my evening now that I know about this :D My prompts on genre wasn't very special, the most niche was "mexican summer hit 2018" to make something in a Despacito style. It did not work but I was not disappointed by the product :D I always focus on a good lyrics and used generics for genre like "gangster rap", "melodic rhythmic rap", "sad female" (otherwise it would just use a male voice most of the cases), "upbeat pop music", "epic power metal". Lately I have been adding stuff like "with clear voice". In the lyrics I use the standar [Verse 1], [Chorus], [Intro], [Outro], [Solo], [Drop] tags and also background shouts in parenthesis like this: (yeah). That works even for a full sentence as a full line and it will sing it in the background, it's really cool, it gives a great vibe to the song. I use that usually before the chorus and put a punchline there that does not have to be rhymed with anything. Looking forward to try this other tool as well :D
Suno sucks
So many of these problems seem to be arise from just mindlessly adding AI features into their product rather than applying it in right places. This is due to the fact that modern large language models/multimodal models are quite versatile, so it seems like it is okay to apply them anywhere, but not versatile enough to make the performance better or enhance user experience. Therefore, unlike more traditional technologies which have more specific use case, the corporates are also confused with what to do with them.
I myself used llms and a little bit of magic to create an assistant that can manipulate my pc - run applications, open web sites, create text notes and i can interact with it via voice. I used combination of whisper ai, llama3:8b and modified firefox-cli. Also, i run it on linux laptop powered by ryzen 7 6800HS. Why can't microsoft or google do the same?
07:30 The "omniscent" chat-bot Is an entity that knows about all the different smells in the world and probably would excel at analyzing farts. "omniscient" is an all knowing being. :D Ok, I'm being a dictionary dork here but thinking about a product named "omniscent" had me laughing. Excellent video in any case.
Amazing video man! I love your videos to stay on track with all the 'frenzy' that's happening~
as far as planning trips go, it still gives you good points of interest to go see which isn't insignificant.
Love your content, Enrico. Interesting topics, well produced and engaging host. Keep it up!
13:24 that is the part im personally excited about.
Think about how many creators from across the world that might get more of an audience now that their voice would be automatically translated?
My invisible AI dream: social media proposes you to meet your friends in real life when you've been having a great discussion.
Without different input methods, generative AI like for making images is pointless. Such a thing could only work if we all had input methods like iPad with Apple Pencils, where we can actually draw and circle things, in different colours to ask for edits on those specific things.
You can already do that
The problem with AI is that I want to sort my cluttered desktop files into folders and to do it by analyzing what is on the images and then place them into already existing folders in "My Documents" correctly . There is no AI that I am aware of that does this.
this seems like a valid use case for ai. I can already imagine the interface. you drag click (1+ times) to select which files you want to sort. then it opens up 6 files at a time split screen and on the right tells you which folder it wants to put them in. you can hit yes or no to approve the transfer. and beforehand you can toggle click on any file to gray it out, signaling that it wont be put in the folder, and can double click to open the file full screen for more info.
just a shame that its an idea not a reality... yet
Devonthink tried to do this for text documents with AI not images. But it don't work well enough to use it. The classification adds a lot of uncertainty and you need thouseands of images to get anything customized learning. Your collection might just never large enough for this unless it's very simple already classified like Animals vs People vs Plants
I have my browser ask me where to save every time when I download something instead of tossing it in the downloads or on the desktop, and I often name it right then and there. Not everything needs an ai solution.
I always use ai chatbots to study for school, but that usually leads me to getting my account banned 💀
just curious how that gets you banned?
@@joeruder everything's great until I use it to help study for US History and ask questions about terrorist groups 🙃
For at least the last 15 years all major applications released don't have every feature you might want included, and the way around that (and a way for intrepid programmers to capitalize on this) was add-ins. These are short lived though, as the primary software will always eventually incorporate the "best" or most popular add-ins directly into the software in a later revision or updated version of the software.
All that to say, nothing about how AI is evolving is new. It's how software has always developed over time. If you think needing added components makes AI useless, then you must feel the same way about everything Microsoft, Good, Adobe, and other big players have been putting out for years now.
Yes ,they have been putting out useless garabage
add-on / plug-in = add-in
first thing i do is copy paste the link of this video in chat gpt, and it gives me a pretty good recap about your concerns about it not being able to make a recap
I've been saying for the last few years that i genuinely don't understand the obsession with the chat bots. They're genuinely not helpful for anything but the most basic common knowledge, and often times will give you blatantly wrong answers with complete confidence. They're nothing but liability when doing anything critical, and a novelty at best when not. Spotify recently added a feature where they have an AI voice mispronounce the artist I'm listening too when i open the app and resume playing. What is the appeal? It's a (bad) solution looking for a problem. Even the AI art, which people want to be use for production value without the cost a human artist: It just looks samey, generic, and cheap. People see AI art and think "lazy, cheap, one man production". The opposite of the desired effect.
Aside from samey, generic, and cheap, AI art also invariably looks a little too smooth, crisp, and clean. And all around subtly soulless, in a way I still can't put my finger on.
I'm not a 100% anti-AI person, but I definitely think that generating imagery purely through AI is NOT the way to go. Not even remotely. I think it's properly used as an image editing assistant, that's the only image-related use I get out of it, and I look forward to AI image generation continuing to expand that type of functionality.
@@Alloveck The concerning thing about "AI art all looking the same" is because... the people generating it like it that way. There can be many other styles in generation. They just don't do it because they don't want to.
@@DuckieMcduck Perhaps you've seen more varied AI art than I have? Because I've seen AI art that's going for photorealism, and AI art going for various flavors of stylization, and realistic or stylized, it still always has that overly crisp and clean and smooth feel so far.
@@Alloveck There are extremely gorged models trained on decades of art with all sorts of tags and qualities, while it will default to what you've seen (excessive shading, photorealism) it can come out looking like a sketch or a flatcolored/monochrome toon just as easily. People just block (negative prompt) such different qualities by default and end up not sharing them as much.
If things look samey or generic, it is because the people doing it are making and sharing it as so. Afterall if extraordinary was common it'd be mediocre (which is exactly what happened)
Chatbots are exactly as useful as the person using them. Right now, we are using Copilot Pro with GPT 4o underpinned and it has increased our productivgity by nearly fifty percent. For good or ill, it has also allowed us to reduce our workforce by twenty percent.
Please add the disclaimer "no Oompa Loompa was harmed in the making of this video"
No one is hurting you, manlet. LMAO
The thing with command lines is they are very good for automation. Nobody wants to click 3000 things or 300,000 things but with a spreadsheet and a bit of cut and paste you can turn a list into a script that can be tested against a copy of your production environment then deployed exactly to.your production environment.
The last thing you want is business users clicking a gui in your test environment then being responsible for doing the exact same thing in production.
Most of the problems you're mentioning have to do with these tools being very early yet.
Speaking about the command line problem: the main problem with classic command lines is that you need to know exact syntax. The whole benefit of LLM is that you talk to it in natural language. In fact, one of typical LLM usages of it for me as a developer, is to generate a correct CLI command for something based on verbal description.
Chat GPT is great because of its intuitive, minimalistic design. The problem is every other company copy and pasting Chat GPT into their mediocre websites and apps.
Ironically, Apple Intelligence just dropped a few weeks after this video 😄
I immediately thought of this video
Thank you this was an amazing video, keep making them , I like your videos.
I agree that the interface is limited right now, but it does take time to perfect it. We are at the very beginning of making AI useful for the average joe!
I think you have a poor understanding what chatbots are useful for. They are amazing for translating, reviewing text, brainstorming, giving tailered help for computer related problems, and answering questions (only where there is plenty of training data). Nobody uses AI chatbots for trip planning!
I think AI will be a tool that we all use in the coming years. We just need to iron out the problems, and like you say, make a better interface.
My whole degree of graphic design is about be waste 🤓🥇🤥
I just watched a video by a graphic designer who lost their job to AI.
sorry but yes- definitely. crazy times
@@lamsmiley1944 who
@@lamsmiley1944 I watch like 5-6 of those . I think I have to rely on my coding and shit
@@GaryMillyz Dunno, their channel is Nadestraight. CZcams randomly decided to recommend their video.
yeeee! finally someone who thinks like me!
we're back to the era of DOS of computers, and the true winners are companies like Canva, Leonardo or Adobe who are actually providing a valuable interface to their ai features. Midjourney and Dall-E are a temporary intermediate step in this first era of AI.
You sure adobe is still winning?
I would add: Using LLM for something productive is even worse than a command line, because the command line is complex yes, but at least it is predictable and follows a well defined logic. LLMs don't.
the problem is not ai per se, but that it is all subscription models, so only 2 things are certain:
- prices will rise to absurd levels over time
- services will get very bad over time.
because that is enshittification in unregulated markets on subscriptions.
I think we need a new OS with ai at its core, not handling invisible stuff, but actually building an entire interface on the go, for example you need a music creation software, instead of downloading an app that can do that, the ui of the computer will generate on the go AS YOU NEED the tools of a music software. For example if you dont need a microphone, the microphone code DOESNT EXIST in the app, until you ask it to.
And that working fully offline, that would be wonderful and VERY POSSIBLE. I already tried to build something like this, but im a single guy not a multimillionaire company.
It will be interesting how much your special program will cost. Since companies want to make money, you won't get that for free.
Something like this is not even remotely in the realm of what's technically feasible.
its a little bit farfetched no? and somewhat useless. why would i use a computer without any fixed way of communicating with it EXCEPT that it creates an app for me (and how long does that take) from scratch (and how many bugs will that have) that i may or may not like, and may or may not properly include the features i need, nad may or may not have a community around it using that specific app. whhy create an entire app without a microphone feature rather than using one that wllows yout o turn off the microphone feature? i see the idea is nice of having an OS build with AI at it's core, but you haven't fleshed it out well (i think). like the video maker said, you should not create a problem and solve it with ai, or 'force' ai to be the solution, rather solve a problem that exists using AI only where AI is more beneficial
Careful dude you might get capped for out-pizza-ing The Hut with an idea like that.
@@isodoubIet bro it already is it just need to be stiched together because it exists as a separate things, but it is technically doable.. i tried to do that on linux but i dont have much patience so f off lol
Your point about chatbots and the “frenzy” is true, but AI tools are definitely not useless. Tools like Adobe Generative Fill or Midjourny are an absolute game changer for creative work.
You’re absolutely right - but creators are not the Majority, And creators would use these features even if they were not labelled as underpinned by AI or not. The AI label is so over, and inappropriately, used that it’s no guide to solution quality…..
Where Large Language Model chatbots are being deployed during peak frenzy is definitely useless. Amazon trialed a chatbots interface for its review search bar a couple months ago. Why?! Do people searching for product reviews really need the output rewritten as a poem or help in Python programming?
They're not a game changer. It's just roulette. And if you aren't lucky, it's a much bigger waste of time than just googling how to do what you need, and gaining a new skill you can use again in the future.
@@benjaminjackson8663 -- or using chatgpt vs google to help learn a new skill.
Game changer for creative work if you’re not already a creative person* It’s making creative stuff easier for non artists, which is cool from a democratization standpoint but not so cool from an artist need jobs standpoint.
i'm curious how long you spent creating this video... appreciate the thoughtful perspective.
Bravo 👏🏻 È sempre interessante seguire i tuoi ragionamenti. Bello il nuovo logo del canale, mi ricorda i disegni di Nick Murray Willis.
I just added an easing function animation to a game I'm developing without any knowledge of using easing functions and very limited coding knowledge in general, I actually just finished producing the game itself and I did a lot of it by prompting with GPT. You have to know what your problem is to a very specific degree, it''s not magic but it is an incredibly useful tool, Specifically for rapid learning and information gathering rather than diving into pages and pages of forum posts about someone else's issue that very vaguely even resembles the thing you're attempting to solve. it's certainly not "useless" lol. personally it's been quite revolutionary.
A trip? thats what you are using it for...
this video will age nicely
Right?! I wouldn’t use an AI to plan / book a trip in 2024. Maybe next year, but it’s not the most obvious use case
It's to show that asking it for anything that requires actual context or a helpful UI is hopeless. It's not useful beyond being good at hallucinating text.
Yeah. I noticed he said something about AI works best with user information which is why they work poorly, then proceeded to reference the new computers coming out with NPU's that will store all necessary information locally, as a local ai, that can work offline. It will age very nicely XD
@@benjaminjackson8663 Yeeeah, that problem is about to be solved reeeal soon.
I think that there is a huge difference between 'AI' (chatbots) and the command-line. I get the fact that it is harder than graphical user interfaces, but the switches for each command are usually documented, so that even though it's not necessarily easy to remember, that command will do the same thing each time, and a command has predictable outcomes by just reading it. I think that this differs quite a lot from using Large Language Models, which can change over time and the same prompt does not necessarily give a predictable outcome.
I personally think an chatbot is not that bad of an interface its just for the most ppl. to slow and not intuitive enough. I mean it takes a lot of time and thinking to enter a prompt per keyboard, especially when you are slow at typing, also the output is often far to overkill for that what asks for.
fantastic video. I typically have 4 ai's open in 4 different tabs and still every app and SaaS system I use is turning from a well structured interface into a nebulous AI prompt. It's so powerful and annoying. It's incredibly, incredibly useful at answering questions you know the answer to well enough to recognize when it's wrong. 🤪
Bro, I think you missed the Voice Conversation mode in ChatGPT 4o. I think most will be chatting via voice then typing.
What is the substantial difference here? Nobody does their computing on Amazon Alexa … and they never will. Until the other mentioned issues are solved, voice changes little yet fuels the hype even more.
... Until you want to look up anything without announcing it to the world around you.
Bro, I don't think you watched the video until the end or don't understand the conclusion. Whether we type or speak, it's still the same.
Text is quiet, talking tells everyone around you what is happening. The incentive for texting is still there. I think the voice thing is suuuuuper creepy and I want to destroy it with fire. It will be so annoying to hear a bunch of chatbots yammering on everywhere you go in public. This is technologically impressive and dystopic at the same time. Maybe we'll find a reasonable center-point eventually, but this whole thing has massive amounts of bullshit throughout it. It's just about making money, creating new demands in the market and cornering them. It's just greed. Look at the internet, easy shopping just accelerated the conversion of the world into garbage and social media greatly increased suicide rates and made everyone into tech junkies always in need of more stimulation. At this point, we should be afraid. Maybe it will transform into something cool eventually, but this stage is gonna be a shiny pile of crap.
@@thomas_xsg It is not restricted to chat bots anymore . Did you see the applications of chat gpt 4o? It is so much more than just a chat bot nw
Photoshop doesnt have a chatbot do you evn know what that is?
I love the concept of invisible AI and I sincerely believe that this is the rationale that will lead the market of AI tools and Saas for the next years, but being also into the marketing field I believe that the chatbot feature is nowadays a common rule to convince investors that: "Yeah you see, we have made a ChatGPT of our own! Can we have your money now?"
Oversimplified but sadly it's the reality startups and corporates have to face right now
I agree that all the major proponents of AI are missing the point, and instead are looking at "AI" as a cash cow. It's sad, they're the same people responsible for why "AI" is experiencing the problems that it is. And then after it's all said and done, they create a product and wrap it up with a bow and then advertise the problem product as the "New Computer" because otherwise their investments fail. Looking at you Microsoft, these "NPUs" etc have so few actual use cases it's mind boggling. It's just gonna become e-waste until actually useful tech comes out.
Give me an AI that does stuff, doesn't talk, and I don't have to act polite to it.
YES, before watching the video i already know im going to agree with you just based off the title
Programmer here, wtf do you mean there's a better way to do it than the command line? The old fashioned dumb command line is still the best way to do it.
AI is just a marketing stunt
You're wrong bro. It's like saying the internet was a stunt.
@@jasondaniels640 yes it is.
No smart people learn and comprehend stuff in the same way as AI does, only stupid people learn (in similar way) like AI, sure it's kind of good but not extraordinary.
The only advantage of AI have over average people have is their coverage of data and speed of their learning which are faster (because they're machine), but they don't comprehend any meaning of those stuff.
Which is why on the surface AI's job is more impressive than what people does.
If you want to call something an AI it must have biological brain, has some kind of sentience, and can actually comprehend stuff they learn (even if it's just minimal amount of comprehension).
But that's just IMO, maybe there will be actual non-biological brain that can truly be called AI, but for now i call that BS.
And you are not very bright.
Its the blockchain, crypto and metaverse all over again.... Words without any meaning for the fkin idiots to believe them
It has to start somewhere. It's mind boggling to think how fast we have come to this.
This is the comment I was hoping to find. Nobody expected (or only a few people expected) LLMS to be so engaging and capable. "AI sucks" because it isn't what I need right now sounds a little like "vinyl records are better than cds". Yes, cds are worse in some dimensions, but think of the streaming capability that is to come.
Yeah I think people are forgetting just how low expectations were before ChatGPT. I think the current state of things is overhyped, but its also already way better than I ever though was possible.
@@lukassvec The problem I am finding is that people just dont know what to use LLMs for outside of write me a poem in the style of X, that and the lack of effort in their prompts.
It's such a shame that the first impression most people had of ChatGPT was this.
8:00 Yes but on Linux we can standardize a LLMs demon, so all software can interact with one single LLMs, whatever LLM the user chooses to use.
10:31 what about programs that have no user interface like Efficient-Compression-Tool?
(yes, there are other png optimizers with user interface but they are not as good (bigger output pngs))
I follow this channel, not because my digital life is a mess but, because of you great, human, compassionate take on user interaction. I avoid all the digital mess that is a negative contributor to my life.
My friend just lost his job to AI someone's finding it not useless
Either your friends job was useless anyways or his boss is useless for thinking AI could actually replace someone and even if it did, he should have given him new responsibilities
Why use Automatic1111 when you could use ComfyUI instead? (11:20) I actually like command lines, but I use krita-ai-diffusion because I find it less confusing.
Thank you! I thought I was the only one not seeing the tangible added value from AI. I feel like it can add quality-of-life features to products, but not much more. When I try to search for 'how is AI changing the world today', I just get tons of articles saying how AI WILL change the world (some dating from 2018).
This video is even more useless than current ai.
I honestly think the most scary thing about AI is AI art, and the fact that this may at some point just replace artists as a whole.
People are so non-chillant about AI art, and laugh at artist for doing their pession while they say "oh i just need a min and a prompt"
When that AI can take that art and apply it to a C&C to incorporate into your art, project or hobby, then watch out. Manufacturing revolution.
AI Art wont replace human art entirely because people prefer to consume human art. Human art will remain superior at the high end of things because it offers the most granular level of control, but AI art is useful for things that don't need that level of control.
I think you nailed. The major problem here is the variety. Only a LLM with access to all the platforms makes sense, but I wonder how much do we want that.
If I get an airline that doesn't serve fried chicken, I don't wanna go to my holiday destination.
Enrico, you make some good points about everyone wants to jump on the “bandwagon “, but AI isn’t useless. It just really started and this video is only going to make people not see the big picture. Unfortunately this video will end up making you look like the dude that laughed at Henry Ford saying “this thing will replace the horse!?!”. AI not LLM’s, chatbots or personal assistants is going to change the world faster than the industrial revolution or anything we humans have ever seen and there will be major life changes for everyone. Hopefully good but possibly existential. Everyone please stay informed. Peace out.✌️
@7:27 The omniscent chatbot - a chat box that combines every smell into one horrendous smell. If one was omniscient, one would certainly stay clear of that stink.
Haha!
ACTUALLY what we WANT is to say “computer- what time is the next train to xyz, and when do i need to leave to catch it” just like in Star Trek. 6:31
The only thing I've actually found it useful for was for writing a job application cover letter for a friend. They got about 5 interviews and a job out of it. Apart from that I don't see any use.
Actually, maybe it could improve Google Assistant which is already ok except for problems with connectivity which isn't an AI issue anyway.
As for the last part, I have no doubt that there are useful applications of AI but if you come at me with your "AI product" right now, I will assume that you are trying to sell a worthless cash-grab and immediately tune out... putting AI in your product description is just a red flag now...
The crazy thing about AI that not many people seem to be talking about is the horrendous power consumption.
No one cares bro
Cope with it
That's what the green communism is for. More wind farms and solar panels! Lol.
Automatic 1111 is awesome. Ed and stable diffusion in general will get better and easier but there is actually a lot of complexity to generating images just the way you want to
Prompting is a lot easier than CLI just regular language
OpenAI has been around since 2015. GPT-1 came out in 2018, it wasn't until 2022 where the LLMs got popular. It took 6 or more years to get GPT where it's at today, expecting general intelligence within the next few years seem optimistic. I think infrastructure for electricity and water is definitely a big thing holding them back besides all the hardware.
That's why Sam Altman has been talking about nuclear power plants recently. A lot of the AI startups try to use web scraping libraries to deal with 3rd party tools, but any changes to their UI breaks their tools and have to constantly be maintained.
I think after text to video, the rate of progress will slow down for LLMs as they deal with edge cases, getting data, and infrastructure.
Enrico, can you answer this simple question about LLMs without looking it up?
What is the loss function of an LLM?
If not, your opinions will be disregarded.
Why does that matter? None of the topics discussed in the video depend on how the machine was built. This is baseless dismissal.
Can you discuss any of the things said in the video?
Perhaps you didn't actually watch the video and you only read the title. He actually likes AI!
And before you reply: If you can't tell me the color of my car then your opinions will be disregarded.
@@FriskyKitsune It matters because there is nothing interesting to be said by people who can't discuss these systems mechanistically, I can speak at length about all of these subjects because I understand the machine that he is complaining about.
To keep the car analogy going, we're talking about broad mechanism of engine - not even the specific function.
You can't tell me what a car can and can't do unless you understand the car - which this guy doesn't!
@@whatisrokosbasilisk80 The impact, or lack of impact, of a piece of software on society has absolutely nothing to do with how the software was created.
The loss function is completely meaningless even if we were discussing how the LLM works, because as long as the model as functioning right now, the steps we took to get here aren't relevant anymore.
People like you are the reason that most tech startups fail. They focus so much on technical details that they can't see the bigger picture. Does this product have a use-case? Is it presented properly? Is the user interface intuitive?
@FriskyKitsune You need to be technical to have competent discussions about emergent technologies - sorry dude, this isn't just another faster horse web service pump and dump.
@@whatisrokosbasilisk80 That is an assertion without basis
It’s not about the chat interface. It’s that AI will be used via API to replace millions of jobs like call centers. AI agents will replace so many tech jobs. AI image and music generators will replace stock / ad content.
This is brilliant, I am actually building my product like this, with invisible AI, people don't always know what they want, also they don't know how to describe it, the goal is to make life easier as you said not more complicated.
Interesting video! I agree with most of your points, but I guess we should consider the potencial of voice interface with LLMs (as in ChatGPT-4o recent demos)
The problem with applying AI to existing problems is that all existing problems already have non-AI solutions.
that are more expensive your comment made no sense btw. transportation was solved by walking so by that logic we just need to walk everywhere right
@@unityman3133
TL;DR: Your comment makes no sense.
Walking isn't a solution to transportation as transportation is not an issue. It was developed to make travelling easier for everyone who didn't have a horse or couldn't make use of other means of transportation or just so happened to live way too far from a job (which wasn't as much of an issue back then as it is now since now cities are built around cars and not the other way around).
AI generated art takes thousands upon thousands of trial and erroring, as we've even seen in the video itself. So much time and money is spent getting the correct output with the correct prompt, paying upwards $300-$600 (if not more) a year, potentially more if you need more GPU time... all to get an output that you could have paid $60-$70 to be done correctly the FIRST time (or with a small back and forth with the artist about specifics that doesn't take thousands of tries to get right). Even then, you could just go and learn it yourself, spend less money on making a computer do it for you and create similar/if not better results the first time by yourself. Drawing tablets cost less than what you'd have to pay for consistent use of AI, while also lasting indefinitely if you know what you're doing. Pewdiepie being another good example of being able to get to such a good state with drawing in such little time that needing an AI to create art becomes counterintuitive.
AI Chatbots and Pins being redundant completely thanks to the little known piece of technology called the SMARTPHONE, which can do all of that for you anyways with reverse image searching and google searches.
AI Summaries/Cooking/Trips, etc., needing to be double/triple checked to the point where it probably would have been better to get the information by going to websites individually anyways instead of paying for any of it. I could list so much more.
These AI solutions are completely pointless, inefficient and costly all at once. All of which having a fundamental flaw in the base construction of LLM's themselves that won't really allow them to reach the potential that most AI tech bros think it will. These AI solutions are flawed at best, and aren't really gonna do much for the average person anyways. It's a gimmick/trick. Much like Crypto and NFTs were. There's no power to the people or convenience here either, just big tech creating big problems for their big solutions.
Well gpt4o solves your problem
I completely agree with you and watch on AI things in the same way, but also I think that AI models need a huge improvements for their chip infrastructure. If we want to build chip infrastructure we must find unbelievable amount of money and only after that we can built something really fascinating. Marketing helps us to make that step faster so it's almost the same story that we had in the past in the 80-s with our computers.