Found your blog through the popular twitter post and have been steadily reading through all the posts. It has been quite gratifying, thank you for writing them!
So my internet provider and bank, they both have this verr verr int-ellergin-t bot when calling on phone, it always feels like ground hog day, never get to talk to a person because the bot insists that it is an employee with as much value as an actual human and therefore I _need_ to take to _it_ instead of a human. Even that it doesn't understand anything I say. But I found out - through lots of pain - that if I cough and then yell "LALALALALALA" in a very high pitched voice, it will actually send me to a human operator. I LOVE chatbots!
To be fair, google search has been degrading in quality for a decade at this point because of censorship, LLMs just accelerated the degree and speed at which it was falling apart.
It seems that the experience of "spending 10 minutes trying to verbosely explain the code you need to ChatGPT and then realizing it would be faster to just write the code yourself" is universal
For the past weeks I have gotten a lot of Indian and Arabic vids recommended here on YT. I'm Middle-European through and through tho and I have shot down countless of these vids, yet YT persists. Shit's cursed. Edit: i haven't read any replies but I don't just get "casual" videos of those groups. It is weird shit like hoax health stuff or cult advertisments. Indiscribable thumbnails too.
@@comedyman4896why is it always so hard for leftoids to hide their political leaning. No one cares that you don't like right wing grifters but suck off left wing grifters like Hasan. I will admit, it must be nice to be so unispiringly typical
This video encapsulates so much anger I've been feeling! I'm working on a cancer detection app. Guess what this requires??? A CNN and swish activated neural network. Does it mean I have to plaster everything with AI this AI that? No it means the front page says "This isn't official advice". It means that I have to ensure that actual links to nearby professionals are nearby. It means prompts for how to take a photo that doesn't screw with the model is taken. Deep learning is a fine technology to use IF the user is aware, IF the data is stored safely (preferably stored locally and encrypted), and most importantly IF IT WASN'T BECAUSE OF ANOTHER (former) CRYPTO BRO, ECON MAJOR, OR BUISSNESS MAJOR.
This is the most based thing I've read in a UA-cam comment section. It's also a reminder that it's not all doom and gloom, and that there are good people out in the world doing good work, and using technology for the enrichment of human lives over personal commercial gain, or data farming.
In Peru we had a congress man present projects that literally had "I’m sorry, but as an AI language model, I do not have the ability to access...." in them
I interviewed some guy who literally said "as an AI language model..." After I asked him a question about unit tests. I tried so hard not to laugh during the interview, I barely succeeded
@@mronewheeler no it wasn't, it was a video call, throughout the interview the guy was very obviously typing in my questions into Chatgpt and then reading out the answers. I knew he was stalling every time I asked a question and there's always a gap between the time I asked a question to him reading out the answers. And then when I asked him "so when you write unit tests for your code, do you write the tests before the code, or the code before the tests?" He answered "as an AI language model I don't write unit tests per se... But (yada yada)" I couldn't believe hearing him say that out loud lmao
Just had an interview Monday. Guy asked me about polymorphism. I know it, but in my nervousness I couldnt really verbalize it. He then asked me if i knew about overriding and overloading and how to do both in C# which was easy. He then explained that he changed the question because he finds that sometimes developers are unable to link what theyre doing to official terms and lingo. But also could provide proper use cases and explanations for the what and why for what theyre doing
@Tom-jy3in conceptually no, but leveraging overrides and overloading are ways in which you can execute polymorphic paradigms in C#. But I'm curious to hear what you have to say about it?
It is crazy how much money people are spending on adding a chatbot to their app that people will maybe use once and then forget about. But as someone who has added features the client has requested that no user wanted or uses… Yeah I guess it’s par for the course.
nahh bruh AI is so good men u dont know what you're saying, it can search stuff on google for you!!! I mean it's all hype, even tho it makes content crazily fast, the content created needs a human input and its often full of errors. it will never surpass human intelligence. I use AI but it wont code for you, you will need to read the docs at the end of the day.
@51:40 I had a discussion recently with someone learning programming and who uses AI for it very often. He said that phrase exactly to me, you will be left behind if you don't use AI like those who refused to use computers. I then asked him the following question: "Which skill is easier to learn, good understanding of programming or using chat gpt?". You will not be left behind if you don't use AI tools. Adapting to using it is easy. Getting the needed knowledge when you hit the limitations of programming with AI is hard.
I don't need to program compilers to program code. I don't need to code LLM to stuff the useful bits its learned about customer behavior into a marketing email.
@@KevinJDildonik That's such a shitty example, because there's plenty of times when using ChatGPT where has no clue how to solve a problem, or it throws out an incorrect answer. What are you gonna do then?
@@TheArrowedKnee Yeah I'm still waiting to see this supposed AI work flow for actual programming jobs. Great it can write out a basic 40 line script or function. But really so what? That may cut it for programming class, but for real world projects you need something that can understand your whole code base and tie to together. And at that point you're probably going to be spending way too much time explaining the problem to the AI to make it worth while. Then you can't even trust the AI so you have to carefully go over and test its code. And if it gets it wrong... well then you're just screwed, because often times it can't fix it.
I so feel his pain... We're having these talks right now, and given i am the only one that actually already did ML, i just asked "OK, but what problem is the model supposed to solve and where is the data it will learn from?". Cue totally blank stares... Segue into "and a single model isn't viable if we want accuracy, we'll need one for each particular 'group' of decision types, as they have no correlation". More blank stares... Then it finally comes, form the back, the dreaded whisper... "Can't we just use ChatGPT?". In my mind, a mix of Clockwork Orange and Aliens starts playing, with Sweating Bullets as the backing track...
The dangerous ones are the ones who know enough to know that "overfitting" is a problem, but not enough to know that they are lightyears away from overfitting being _their_ problem.
@@chris-hu7tm Pray tell, what is there so outlandish that could bring it down to 20%? p) Not knowing what they want to solve? q) Not having (enough) data to do it? r) Not understanding that a single model can't solve every problem? s) (Idiots) Believing ChatGPT can do everything? Given q/r/s are probably in the 90-100% range, you'd have to lean REAL HARD on people knowing PRECISELY what they want to get a 20%. You do know how to calculate the joint probability of independent events, right?
1:30 jokes aside, there are actuall graphic designers losing their jobs to AIs that have been trained with the designers' previous work... so, it is no joke for those professions
Yeah, I've been paying more attention to specific stuff in the space since writing this, and have gone from "I am annoyed at middle managers going on about this" to "These people have no respect for artists, other people's work, and the law".
@@LudicityHackernews the issue is that, at the end of all things, what we all require is the end result. if I need 5 icons for my web app; and the choice is use a free AI tool to get results I'm happy with vs love, respect and bow to the artist and pay them for the icons because love; I'm using the AI. It's a no-brainer. They're going to lose their jobs. and there's still going to be that 1-5% top artists who are going to be creating high art and receiving millions for it, and no one's going to lose actual RESPECT or LOVE for the art and the artists. But your neighbour Millie who used to draw up icons and logos and moonlight as a comics designer here and there, she's gonna go out of work.
This article is written SO well. Compare this one the article read that is saying AI is great--the one that uses horror movies for chapters--and the difference is wild.
Most "AI" projects are literally just Actual Indians. Google search algorithm? Turns out to at some degree be controlled by a crew of Indians who manually rate links to determine quality. Amazon automated stores? Indians. Most coding demos, especially the ones that for some reason take a few hours? Work is just batched to India.
When the tech layoffs fired 20% or the workforce? AI took up the slack. Every Actual Indian I work with has been taking a lot of unplanned vacation lately, mostly mental health days, from all the stress.
@@KevinJDildonik People keep elling me this, but none of them have actually been able to SHOW ME this supposed work it is generating. Right now, if "AI" can prduce it then it is mot likely of no value. If you wish to argue the point please take it up with Air Canada. I do not give a single f**k.
@@davea136 the Devin AI chatbot that writes code is intentionally made to replace programmers. That said, it's not very good, yet. Last time i looked into it, it only solved problems around 15% of the time. But that's the thing with ai. It improves at a specific task drastically faster than humans do. Opensource tested out their AI against proffesional dota 2 players around 6 or 7 years ago. The first show matches, the pros destroyed the bot easily. Two weeks later none of the pros could take a game off it. Also canadians are incompetent at the best of times. Their entire existence is "were insignificant so no one cares about us but everyone knows we live next to the quarterback."
It's really tiring to see "ponzi scheme" used everywhere in every single conversation like it's bread for breakfast. We *can* say something is unhealthy, bullshit or genuinely bad without having to try to make it fit in the only mold we somewhat heard of that with some adjustments might have a chance of looking a bit similar in a low light environment when squinting hard enough. Not everything bad is a ponzi scheme.
@@ExilumI feel this way about how people use the word "grifter" these days. People cant just say "this guy is an idiot" or "llms seem like theyre overhyped" it needs to be some buzzword because the lexicon of your average twitter, reddit, youtube commentor contains only like 10 words. God forbid people try and include some kind of nuance in their language.
It really resonates with me when you were talking about using your hands to learn. I am currently trying to make sense of nix, nixpkgs and NixOs. For the past, 6 months, I've been trying to just ask GPT "I would like to do this in a nix flake", or some kind of code. It would, NEVER, work. I always had to change stuff around. 1 week ago, I told myself, sick of just typing stuff "to make it work" (or couldn't make it work in many cases). And I decided to open multiple wikis, and docs for nix. In one week, I am way above what chatgpt tried to understand, I'm no professionnal at it, but I was able to make stuff work that I thought would've been to hard not too long ago. Really enjoyed this article, a breath of fresh air, in a polluted climate of AI.
As a rule the more rarely used or verbose a language, the worse AI will be at writing it, because you need a longer context window for verbose languages and you need more training data for rarely used languages. Anything that's not python or js AI is pretty bad at.
Do you have tips for nixos ? My home server is on nixos and it's so bothersome to do simple stuff I could do with a simple docker compose file ... I'm starting to get burn out of trying installing apps the nix way
I had the same experience when messing around with arch, it can help with really basic stuff, but anything that involves even a few moving pieces gets really dangerous, I even bricked the system listening to the advice from GPT.
On the Elixir point, it's also absolutely dreadful at Laravel. Me and my team gave up asking it Laravel questions because it straight up hallucinates answers that contain things that have never existed in any version of Laravel, ever.
Use a context file to add to your prompt. Show it examples of similar code and/or what not to do. It’s not going to always get everything right from a 2 sentence prompt.
It worked great for me for Laravel, and also a bit of React now (same project). It is beginner stuff, but my god is it good. Literally all the code I got worked without changing except one database migration where it hallucinated the command for adding db checks (and when I told it that it doesn't work, it gave me the right one). Programing is probably one of the best applications of LLMs, and I'm so happy for that. We are the lucky few that can reap at least some of the benefits of it.
it cant even read the official doc and give you a wise answer. for simple stuff is great but for most complex stuff its useless. and everything he gives is in the docs already! I just go straight to the docs nowadays.
Once you understand the underlying statisical models even on a basic level, you'll quickly understand how absolutely bullshit this current LLM gravy train is.
tech priests would laugh at AI, but if they saw the pyramids they'd want to run into the tombs and find relics and fight necrons that doesn't exist in our tombs, thank god for that..
The role of a software developer isn't to implement features and fix bugs. The role of a software developer is to create a system (of features) that is maintainable even if said developer has left 5 years ago. The primary tools of a software developer aren't the coding language, the framework, the repository system, or anything so specifically. The primary tools of a software developer are selection and projection: "Will I understand this solution in 3 months, 6 months, 3 years?" "Will anyone other than me understand this solution in that time?" "Will this work without me or anyone else babysitting every single step of the way to production and beyond?" "Does this language/framework/process create more solutions than problems?" "Does this only solve a very specific problem, or can this solve a general category of problems that really occurs?" Ask yourself these question, and try to answer them without hubris and ego.
its human bias, I mean the bias it has is the algo which some human or humans coded. its has no personal imagination or preference. it gives you an answer based on optimal statistics with greater chances of success. in this case to get the job done. but this stats and probability it calculates in miliseconds all comes from some human algo that tells it what to do.
@@ricardomartinhodacruz I dont think this case reflects human bias. This is due to it being an LLM. Based off the text, it tries to predict based on all the data its ever seen. Thats all, it has no understanding.
it's human stupidity to assume that names in this task don't matter or mean anything. It was given extremely little context, and the names could very well mean something.
@@lystic9392 No this just shows that you cant ever rely on LLMs to reason. They cannot because they do not have understanding. They provide probabilties based off how each letter/word/token compares to every other it has ever seen (trained on). This is one of the simplest examples I have seen to convey this.
4:20 AI winter was actually a thing back in the 20th century. Also, there have been several growth periods and even more stagnation time inbetween. Btw, before LLMs in 2020s, there was also boom in deep learning about a decade earlier (hardware got fast enough, we ditched complex activation functions instead going with fast stuff like ReLU and most importantly ideas like ResNet solved the major issues of backpropagation).
I had to lookup the meaning of "piledrive", expecting some tech stuff to do with drivers and stacks... My shock when the first result is from webMD talking about adult fun time positions... my relief when the second result talks about wrestling (with a very cute 31 frame stickman animation), and my shock again when the third result is that it's the model name of the (selfaclaimed) most powerful air rifle... And finally, a dictionary result talking about heavy duty machinery. So yeah, TIL a new word... Anyway, now onto actually watching the video, not anymore to hear something about AI, but to find out which kind of piledriving the OP of the article was threathening (a good time?) with
@justiny2552 ooh, yeah i'm Dutch so I cognate a lot of English terms, I wouldn't have even thought to split the term. Now I get results exclusively about the heavy duty machines to hammer pylons into the earth. Tempting rabbit hole that, on the history of constructions. Nothing about wrestling tho, so that one's ethologically closer to the naughty one. Wonder which one came first (not like that) @bradweir3085 not that I particularly care if you believe me or not, am just procrastinating here; I've never come across the term, wrestling isn't that big outside the Americas. Or do you not believe that particular order of results, because I did lie a bit: in between the fourth and third results I mentioned was another adulttime related one. Cus I use Google engine through Startpage on a mobile network, I don't think there's personalisation outside country localisation going on, before people start claiming that about me lol
As a guy who works in a giant indian enterprise, i manuvally do thing that i can automate in an hour (infact, as my collage project, i automated entire work in my role just by making a chrome extention and a excel vba, and today i presented it)
A lot of the criticism on LLM's boils down to one thing : some things are so obviously true that nobody in the history of mankind ever bothered to write it down. Yann LeCun gave this example : when there is a cup on the table and you move the table then the cup will move with it. How do you even begin to figure out what these baseline truths even are? Let alone write them all down and then train your AI on them first. And then also make sure it won't forget about it when you keep training it? There is only so much that language can do for us.
And he was wrong, try and ask any AI model this question and you will get the correct answer. His examples were wrong. Rational Animations did a good video on this topic, you should check them out.
It is especially true when it comes to culture and history. People don’t write down what is so obvious to them that it doesn’t even occur to them to write down about it.
@@TheManinBlack9054 Your comprehension skills and imagination are inline with an AI The point is what questions and information you do not realise simply do not get written down despite how intuitively we understand them. There is no single question that defeats this point of concern.
FYI: The owner of that blog and author of the article is Nikhil Suresh, not Ludic Mataroa. Ludicity is the name of his blog and mataroa is a blogging platform he uses.
AI became quite lucrative well before LLMs. Expert systems have been controlling things like speech recognition and flight scheduling at airports for decades. What LLMs did is make AI mainstream in a popular way. When people say "AI Winter" what they really mean is that we don't have Data from Star Trek yet. Until we get Data, we'll always bein on the cusp of or going through an "AI Winter." But the truth is that people have been using and making money off of AI for a long time.
When I became aware of the increasing role of AI in generating code, especially for learners, I knew it was a good time to learn how to program. Somebody is going to have to know how this stuff actually works, the OG chads won't live forever and if you are "learning" with this tech, you won't really know much besides how to proompt. This is great for people that can actually do the work as the skill divide becomes a yawning chasm. 😎
20:50 - Here's the problem: Details matter. It said "been assisting with strategic decisions". Using LLMs for ideas and not necessarily correct explanations on how things work, is exactly what they are good for. This can of course be used as an input for decisions, maybe often just as an inspiration.
I guess in the case of customer support LLMs could be used to protect the employee against the "impatient" customers. So here would be my suggestion : customer support gets to "rate" the customer at the end of the call / chat. Ratings are shared across companies. If a phone number / customer ID has bad enough ratings, the system defaults to giving them AI support, if the customer has good enough rating, they get a real person every time, considering that the pool of available agents is not much bigger. There would be a way to appeal the rating. This way, overall, assholes are filtered out and I'm rewarded for talking to customer support like they're actual human beings by being given actual human beings
I still don't know why remote live support is still valued. If, in 2024, someone is using a computer but doesn't know how to ctrl+f a faq and has to get a call centre employee to do it for them, they are too stupid to do business with.
@telnobynoyator_6183 Sorry to disappoint you but I years ago I watched a docu about call centers and their shady tactics.. Allegedly, each caller receives a rating attached to their number: friendly callers get a positive mark, while frustrated callers receive a negative one. Unfortunately, most large call centers operate under a single umbrella or utilize the same software, allowing unrelated service numbers to identify "troublemakers" in advance. They can choose to ignore these calls and move on to the next one.
And then, the AI (LLMs) will mix up some or all parts of those answers with all the crap non-answers it’s also been fed with. Hopefully you’ll be able to tell the difference.
I love laughing at the "just fine tune the model"... on what? the 2 files and 12 database records I have in my dev server? or the 100 files and 10,000 records in my production server littered with PII. Either way that's not enough data to fine tune anything.
Todoist is phenomenal - you write "meeting with Primeagen tomorrow at 15:30 !15mb" and it sets and event/task "meetings with primeagen" with a deadline tomorrow at 15:30 and a notification 15 minutes before the deadline. Works with other natural language phrases like "every other wednesday", "every last thursday of every month", etc.
@@Jabberwockybird Todoist doesn't use LLMs for this. The LLM feature lets you write more complex queries over your tasks, but 99% of the value comes from what are probably a bunch of regexes.
About a month ago, the CEO of the company I work for basically went "Hey, you know that alert and suggestion system we asked you to implement? The one with zero user inputs, and can be done by using templates and swapping out a couple of words? Can you add an AI there?"
I'll not press 1, I'll press 2 for deep doubt for current AI. I have no idea what will be in the next 5 to 10 years, but the thing we have now is partly usabe, deeply unreliable, and hyped to high heavens.
The way I like to use the gippity for dev is by using it to explore unfamiliar problems. Not that it can solve them for me, but it sometimes pushes out solutions using techniques or libraries that i may not have even thought to google for because i dont know the domain well enough. Often theres some code worth copying, I usually end up optimising, but it gives me a atarting point. Often a lot of the boilerplate I'll throw away.
38:53 Why do we do this to him? Why can't we just read our own articles? I don't have dyslexia. I can read pretty well. I'm not going to stop, though. He's too entertaining.
"... add chatbot support to every application under the sun when half of the industry hasn't worked out how to test database backups regularly" made me felt so heard. I had a system stand for 3 weeks losing millions because database backups weren't checked. Had to manually recreate 3 months of orders. One week before that incident we had a meeting on how to utilize AI.
So I use LLM “whatever” to rewrite my cv to make it match a job description. Prospect employer uses the same LLM to calculate the match value of my cv in relation to the job description. So, who won?
In fairness to AI in cybersecurity, it's a pretty serious threat in the context of your programmers using it to create massive security vulnerabilities in your own software and the utility in improving or mass producing phishing attempts.
39:20 same man, haha, can end up reading the same word so many times that it starts to not even look like a real word anymore and then nothing is real. good ol' adhd
35:14 I’ve seen one of the pitches for “LLM on PR” company. Their opening statement was- “currently we have GPT writing 10% of the code, soon it will be 90%”.
36:20 if legislation becomes the spiritual equivalent of folk tales told by LLMs, will anyone ask "Wait, didn't we start writing laws down, so people don't just make sht up?"
42:05 My big thing is phind. It tells me what to do *IF* I ask the right question, gives me useless code, and tells me how to make the code as if it was good. I come away with sources, steps to start researching, and new concepts to figure out. Phind has a lot of issues, but that's what helps. It's like talking to a toddler that has access to the sparknotes of all things. One day, I hope to make my own that feels like a middleschooler or better, and phind gave me some idea how to do that, once I figure our how. 😅
34:50 I got a hot take: Asking an AI to write unit tests for the code that the same AI wrote is like asking a corporate entity to investigate itself for any wrongdoing. You can only trust it to do whatever it takes to make itself look good, in spite of any and all flaws that indicate otherwise.
I started reading LOTR: fellowship a few weeks ago (got to Moria but haven't picked it up in a while). Interestingly, there are a lot of songs and poems in the book that I did not expect. Some of them are quite touching or fun, but some of them are pretty throwaway lol.
Those two magical letters are gonna get piledriven on faster than that elevator guy's face in "Drive" and there's nothing we do about it except make it super uncool.
I think one of the overlooked things working in Copilot and other coding assistant AIs favor, is what happens when ya step outside the coding for money setting, strangely. It allows specializations created by other industries carry people so much further solo, letting people focus on the things they are good at during the early stages of a project while letting the AI worry about the code they don't really understand. Once that website is a proof of concept, or that app is an alpha, or the game is functionally playable long enough to show off a bit, and that proof of concept is figured out, then they can hope to find the funds they never had before and hire somebody to patch the glaring security holes and fix the idiot mistakes, or lets be honest in a lot of cases gut the entire mess and do it right. But still, a lot of good ideas likely die because a person lacks the time to learn to code or just aint got the head for it. And yes, some never will and will attempt to keep going with AI generated code and they will suffer as a result... so like, be extra careful who ya give that credit card number to for a bit I guess?
43:00 This is so fucking real. Starting out in 2022 I started learning and taking courses in school that taught and utilized Python, learned a fuck ton. Discovered GPT and Co-Pilot later that summer and over the past year and a half my brain just started leaking all of the useful problem solving skills and syntax out. Recently turned co-pilot off and I just spent 10 minutes staring at the fucking screen. The shit is poison for your brain.
Human strongly disagree: Like the inventor of the insulin production that donated the recipe for it to be available to everyone. But seems the news did not reach the US of fing A and you ends up the only country in the world to have people die because insulin is not affordable xD
4:35 Good argument, but I'd like to inform you that an 80+ year old ex-judge (fairly high in the court order as well) I got to interact with was pretty certain that the development will accelerate and we will all be soon (within the next 10 years) be crushed under the feet of skynet. To be fair to him though, who knows maybe he's right 🤷♂
I was hired to produce AI content for a company. I said it won't work as they want it to. Hired me anyway. I was fired 9 months later because it wouldn't work the way they wanted it to. Last week they wanted to re-hire me because they can't find anyone to make it work like they want it to.
Here's something else that's problematic about AI usage - even in the places where there's a good use case, it can still go out of bounds from your original intentions in ways you couldn't have imagined.
On the topic of not using copilot to learn and learning by doing, there is a quote that I absolutely love. I am a mathematics student and whenever someone asks me how I managed to become good at math I tell them this quote: "Mathematics is learnt by banging your head against a wall". Math can be so difficult that the ONLY way you can learn it is by closing the textbook and actually trying. You may fail and eventually need to look up the solution, but then you move on to the next question and repeat. Then you come back a week later and redo everything without looking up the solutions. You learn by doing and your research (or as you call it, review) is simply a tool to can aid in that learning; however, you cannot learn without actually doing -- with or without that review.
about CoPilot, this is VERY TRUE! .. for anyone can't make the connection and is older than 30 can conform, driving with and without street Navigation, has the same effect, if the navi keeps telling you turn by turn where to go, once the device is off, you will ba really lost, because you never rmemorized the streets but were just the bot that dives after being told to.
I worked for a company that couldn’t manage to roll out office 365 until 2022 and they use excel sheets to schedule manufacturing capacity. There is no way they are using AI effectively within the next decade.
What's wrong with Excel? It's pretty good at modeling most data, and it's much more intuitive than programming languages for most people. (I say this as a programmer)
I setup automated backups for our DB at this startup i was at, and left it running. At some point it stopped working, and there was apparently no backups for 3 months lmao. Don't even remember what the issue was. While i was still there, would just checked the bucket every few days to make sure it was still running.
44:15 Zero-trust security architecture? very interesting. I was thinking since knew about zero-knewledge proofs, if that can be used to build secure systems with no potential of leaking data to corporations or 3rd parties
As a guy who brainwashed himself with ai content and building python stuff around llms for 1,5 years i agree. It is sickening and useless😅 also it seems to make your hair go grey
I think GPT really overshadows everything else AI can do, because it's the most visible to your average person. Stuff like weather predictions, replacing repetitive manual labor tasks (why do they always assume it's tech jobs that get replaced?), self driving cars, and so on. Perhaps the thing I'm most interested in, is if AI could intelligently decompile software and make it human readable with insightful variable names. That would be an absolute nightmare. I assume the NSA or some TLA is working on that tool as we speak.
![The moment we stopped understanding AI [AlexNet]](http://i.ytimg.com/vi/UZDiGooFs54/mqdefault.jpg)
I am the author and have been so excited to watch this, but also live in constant fear of Senpai Primagen's judgement. Wish me luck, Godspeed.
That's a lot of +1s in chat, god bless you maniacs.
Senpai loves me, I'm selling bathwater.
39:10 Primagen confirmed can only handle Low T content
man i loved you article so much it was fun to read, all respect Ludic senpai
Found your blog through the popular twitter post and have been steadily reading through all the posts. It has been quite gratifying, thank you for writing them!
Dave: "Open the pod-bay doors, Hal"
Hal 9000: "I'm Sorry Dave, I'm Afraid I Can't Do That"
Dave: "you can!"
Hal 9000: "Opening pod-bay doors"
Made me laugh so hard, after so long! Thanks man! ❤
“sudo open the pod bay doors, Hal”
LLMs will be instrumental in solving the degradation in search engine results caused by LLMs.
So my internet provider and bank, they both have this verr verr int-ellergin-t bot when calling on phone, it always feels like ground hog day, never get to talk to a person because the bot insists that it is an employee with as much value as an actual human and therefore I _need_ to take to _it_ instead of a human. Even that it doesn't understand anything I say. But I found out - through lots of pain - that if I cough and then yell "LALALALALALA" in a very high pitched voice, it will actually send me to a human operator. I LOVE chatbots!
I can't wait until we start seeing jobs for "Guy who knows how to word questions to get ChatGPT to respond truthfully."
It started with SCOs
Classic. Create a problem that solves the problem and breed more problems.
To be fair, google search has been degrading in quality for a decade at this point because of censorship, LLMs just accelerated the degree and speed at which it was falling apart.
It seems that the experience of "spending 10 minutes trying to verbosely explain the code you need to ChatGPT and then realizing it would be faster to just write the code yourself" is universal
Chat gpt is the best rubber duck, just never actually send the message
@@gwentarinokripperinolkjdsf683 I tried to develop an algorithm using ChatGPT, and this was 100% my full experience.
I think you never used ChatGPT
Makes me think my idea of starting a "how to phrase your AI request" school is a great idea ;) Or possibly a class at a tech school. hahaha
@@pRoFlT doesnt matter how you phrase it, chatgpt is bad at logic
13:20 Twitter knows that all I look at is porn and yet it regularly promotes politics and political figures that would not align with my porn choice.
For the past weeks I have gotten a lot of Indian and Arabic vids recommended here on YT.
I'm Middle-European through and through tho and I have shot down countless of these vids, yet YT persists. Shit's cursed.
Edit: i haven't read any replies but I don't just get "casual" videos of those groups. It is weird shit like hoax health stuff or cult advertisments. Indiscribable thumbnails too.
_"...I could never vote for an ass-man"_
I swear the Twitter AI simply will not stop shoving right wing grifters down my for-you feed no matter how many porn stars I follow
@@comedyman4896why is it always so hard for leftoids to hide their political leaning. No one cares that you don't like right wing grifters but suck off left wing grifters like Hasan. I will admit, it must be nice to be so unispiringly typical
@@comedyman4896 surely you dont follow left-wing grifters on twitter, that would just be ridiculous
"Do you test your backups to make sure you can recover your application?" "Nah fuckit, we ball" The response heard around the world.
This video encapsulates so much anger I've been feeling! I'm working on a cancer detection app. Guess what this requires??? A CNN and swish activated neural network. Does it mean I have to plaster everything with AI this AI that? No it means the front page says "This isn't official advice". It means that I have to ensure that actual links to nearby professionals are nearby. It means prompts for how to take a photo that doesn't screw with the model is taken. Deep learning is a fine technology to use IF the user is aware, IF the data is stored safely (preferably stored locally and encrypted), and most importantly IF IT WASN'T BECAUSE OF ANOTHER (former) CRYPTO BRO, ECON MAJOR, OR BUISSNESS MAJOR.
If you want it to sell you will plaster AI all over it
Same man working on malaria parasite detection at edge on low powered device.
Underrated comment
a single finance bro is thousands of times worse than a libertarian socialist gender studies grad
This is the most based thing I've read in a UA-cam comment section. It's also a reminder that it's not all doom and gloom, and that there are good people out in the world doing good work, and using technology for the enrichment of human lives over personal commercial gain, or data farming.
In Peru we had a congress man present projects that literally had "I’m sorry, but as an AI language model, I do not have the ability to access...." in them
They are dumb
I interviewed some guy who literally said "as an AI language model..." After I asked him a question about unit tests. I tried so hard not to laugh during the interview, I barely succeeded
@@conancatWth 💀
@@conancat I presume this was an interview in text? Or was it just a really good joke?
@@mronewheeler no it wasn't, it was a video call, throughout the interview the guy was very obviously typing in my questions into Chatgpt and then reading out the answers. I knew he was stalling every time I asked a question and there's always a gap between the time I asked a question to him reading out the answers. And then when I asked him "so when you write unit tests for your code, do you write the tests before the code, or the code before the tests?" He answered "as an AI language model I don't write unit tests per se... But (yada yada)" I couldn't believe hearing him say that out loud lmao
Just had an interview Monday. Guy asked me about polymorphism. I know it, but in my nervousness I couldnt really verbalize it. He then asked me if i knew about overriding and overloading and how to do both in C# which was easy. He then explained that he changed the question because he finds that sometimes developers are unable to link what theyre doing to official terms and lingo. But also could provide proper use cases and explanations for the what and why for what theyre doing
but... polymorphism has nothing to do with overrides and overloads
@@Tom-jy3in what is polymorfism?
had an interview like that where i panicked once and dropped my spaghetti. that interviewer sounds cool, mine wasnt as forgiving
@Tom-jy3in conceptually no, but leveraging overrides and overloading are ways in which you can execute polymorphic paradigms in C#. But I'm curious to hear what you have to say about it?
@@Tom-jy3in Polymorphism is a broader concept but overrides and overloads are absolutely and example of parametric polymorphism in type theory.
It is crazy how much money people are spending on adding a chatbot to their app that people will maybe use once and then forget about.
But as someone who has added features the client has requested that no user wanted or uses… Yeah I guess it’s par for the course.
The latter is part of the job, as much as it sometimes drives me up the wall still.
I come from a user-centred design philosophy and the disconnect from users in this ass-backwards approach baffles me.
Especially since all LLMs make it very obvious yo're talking to AI by how they phrase responses lol
nahh bruh AI is so good men u dont know what you're saying, it can search stuff on google for you!!! I mean it's all hype, even tho it makes content crazily fast, the content created needs a human input and its often full of errors. it will never surpass human intelligence. I use AI but it wont code for you, you will need to read the docs at the end of the day.
Reminds me of the social media sharing buttons back in the day
@51:40 I had a discussion recently with someone learning programming and who uses AI for it very often. He said that phrase exactly to me, you will be left behind if you don't use AI like those who refused to use computers. I then asked him the following question: "Which skill is easier to learn, good understanding of programming or using chat gpt?". You will not be left behind if you don't use AI tools. Adapting to using it is easy. Getting the needed knowledge when you hit the limitations of programming with AI is hard.
I don't need to program compilers to program code. I don't need to code LLM to stuff the useful bits its learned about customer behavior into a marketing email.
@@KevinJDildonik you need to if you want to make good code.
@@KevinJDildonik That's such a shitty example, because there's plenty of times when using ChatGPT where has no clue how to solve a problem, or it throws out an incorrect answer. What are you gonna do then?
@@TheArrowedKnee Yeah I'm still waiting to see this supposed AI work flow for actual programming jobs. Great it can write out a basic 40 line script or function. But really so what? That may cut it for programming class, but for real world projects you need something that can understand your whole code base and tie to together. And at that point you're probably going to be spending way too much time explaining the problem to the AI to make it worth while. Then you can't even trust the AI so you have to carefully go over and test its code. And if it gets it wrong... well then you're just screwed, because often times it can't fix it.
@@TheArrowedKnee Don't worry, your boss probably won't notice and in our society rewards have very little to do with actual performance.
I so feel his pain... We're having these talks right now, and given i am the only one that actually already did ML, i just asked "OK, but what problem is the model supposed to solve and where is the data it will learn from?". Cue totally blank stares... Segue into "and a single model isn't viable if we want accuracy, we'll need one for each particular 'group' of decision types, as they have no correlation". More blank stares... Then it finally comes, form the back, the dreaded whisper... "Can't we just use ChatGPT?". In my mind, a mix of Clockwork Orange and Aliens starts playing, with Sweating Bullets as the backing track...
lmao this is so true
The dangerous ones are the ones who know enough to know that "overfitting" is a problem, but not enough to know that they are lightyears away from overfitting being _their_ problem.
Something similar happened but I dont believe ur story, ill say 20% is true
@@chris-hu7tm Pray tell, what is there so outlandish that could bring it down to 20%?
p) Not knowing what they want to solve?
q) Not having (enough) data to do it?
r) Not understanding that a single model can't solve every problem?
s) (Idiots) Believing ChatGPT can do everything?
Given q/r/s are probably in the 90-100% range, you'd have to lean REAL HARD on people knowing PRECISELY what they want to get a 20%.
You do know how to calculate the joint probability of independent events, right?
There's a war inside my head!
one of these days a company will get the idea to use a LLM for its CEO and they will be astounded at the money they save.
I remember reading about a chinese company doing that a couple years ago.
LLM for government!
NetDragon Websoft!
There's also Mika, a Polish company that has an AI powered Robot as CEO.
who would you rather work with
Just wait for exclusive and totally not monopoly inducing contracts between OpenAI and the government
1:30 jokes aside, there are actuall graphic designers losing their jobs to AIs that have been trained with the designers' previous work... so, it is no joke for those professions
Yeah, I've been paying more attention to specific stuff in the space since writing this, and have gone from "I am annoyed at middle managers going on about this" to "These people have no respect for artists, other people's work, and the law".
@@LudicityHackernews they only care about the bottom line, which is understandable I guess...
@@LudicityHackernews the issue is that, at the end of all things, what we all require is the end result.
if I need 5 icons for my web app; and the choice is use a free AI tool to get results I'm happy with vs love, respect and bow to the artist and pay them for the icons because love; I'm using the AI.
It's a no-brainer. They're going to lose their jobs.
and there's still going to be that 1-5% top artists who are going to be creating high art and receiving millions for it, and no one's going to lose actual RESPECT or LOVE for the art and the artists. But your neighbour Millie who used to draw up icons and logos and moonlight as a comics designer here and there, she's gonna go out of work.
This article is written SO well. Compare this one the article read that is saying AI is great--the one that uses horror movies for chapters--and the difference is wild.
Just released that one today
AI is just an acronym for Another Indian
Most "AI" projects are literally just Actual Indians. Google search algorithm? Turns out to at some degree be controlled by a crew of Indians who manually rate links to determine quality. Amazon automated stores? Indians. Most coding demos, especially the ones that for some reason take a few hours? Work is just batched to India.
When the tech layoffs fired 20% or the workforce? AI took up the slack. Every Actual Indian I work with has been taking a lot of unplanned vacation lately, mostly mental health days, from all the stress.
@@KevinJDildonik People keep elling me this, but none of them have actually been able to SHOW ME this supposed work it is generating.
Right now, if "AI" can prduce it then it is mot likely of no value.
If you wish to argue the point please take it up with Air Canada. I do not give a single f**k.
I thought it was meaning "A*s Insertions"
Now I'm confused...
@@davea136 the Devin AI chatbot that writes code is intentionally made to replace programmers. That said, it's not very good, yet. Last time i looked into it, it only solved problems around 15% of the time. But that's the thing with ai. It improves at a specific task drastically faster than humans do. Opensource tested out their AI against proffesional dota 2 players around 6 or 7 years ago. The first show matches, the pros destroyed the bot easily. Two weeks later none of the pros could take a game off it.
Also canadians are incompetent at the best of times. Their entire existence is "were insignificant so no one cares about us but everyone knows we live next to the quarterback."
This just ripped the feelings right out of my soul and layed them bare with more grace than I could ever muster.
The implied unending rise of AI in culture has really big Ponzi scheme vibes.
It's really tiring to see "ponzi scheme" used everywhere in every single conversation like it's bread for breakfast.
We *can* say something is unhealthy, bullshit or genuinely bad without having to try to make it fit in the only mold we somewhat heard of that with some adjustments might have a chance of looking a bit similar in a low light environment when squinting hard enough.
Not everything bad is a ponzi scheme.
@@Exilum youre right. but all these AI startups (or most of them) are ponzi schemes. a lot of them have already been exposed
So we need a new term for Ponzi scheme that is more tech-related?
@@Salantor Blockchain scheme? IoT scheme?
@@ExilumI feel this way about how people use the word "grifter" these days.
People cant just say "this guy is an idiot" or "llms seem like theyre overhyped" it needs to be some buzzword because the lexicon of your average twitter, reddit, youtube commentor contains only like 10 words. God forbid people try and include some kind of nuance in their language.
It really resonates with me when you were talking about using your hands to learn.
I am currently trying to make sense of nix, nixpkgs and NixOs. For the past, 6 months, I've been trying to just ask GPT "I would like to do this in a nix flake", or some kind of code. It would, NEVER, work. I always had to change stuff around.
1 week ago, I told myself, sick of just typing stuff "to make it work" (or couldn't make it work in many cases). And I decided to open multiple wikis, and docs for nix.
In one week, I am way above what chatgpt tried to understand, I'm no professionnal at it, but I was able to make stuff work that I thought would've been to hard not too long ago.
Really enjoyed this article, a breath of fresh air, in a polluted climate of AI.
As a rule the more rarely used or verbose a language, the worse AI will be at writing it, because you need a longer context window for verbose languages and you need more training data for rarely used languages. Anything that's not python or js AI is pretty bad at.
Do you have tips for nixos ? My home server is on nixos and it's so bothersome to do simple stuff I could do with a simple docker compose file ... I'm starting to get burn out of trying installing apps the nix way
I had the same experience when messing around with arch, it can help with really basic stuff, but anything that involves even a few moving pieces gets really dangerous, I even bricked the system listening to the advice from GPT.
Reading the docs remains the greatest super-power out there
@@xenio8736 I put everything "that just works" tm in docker-compose, but for example, my nextcloud/samba share is on my nix config.
On the Elixir point, it's also absolutely dreadful at Laravel. Me and my team gave up asking it Laravel questions because it straight up hallucinates answers that contain things that have never existed in any version of Laravel, ever.
apparently also asking about php it loves to answer about php6
Use a context file to add to your prompt. Show it examples of similar code and/or what not to do. It’s not going to always get everything right from a 2 sentence prompt.
It worked great for me for Laravel, and also a bit of React now (same project). It is beginner stuff, but my god is it good. Literally all the code I got worked without changing except one database migration where it hallucinated the command for adding db checks (and when I told it that it doesn't work, it gave me the right one).
Programing is probably one of the best applications of LLMs, and I'm so happy for that. We are the lucky few that can reap at least some of the benefits of it.
it cant even read the official doc and give you a wise answer. for simple stuff is great but for most complex stuff its useless. and everything he gives is in the docs already! I just go straight to the docs nowadays.
Once you understand the underlying statisical models even on a basic level, you'll quickly understand how absolutely bullshit this current LLM gravy train is.
❌ Mentioning LLMs
✅ Mentioning the Omnissiah
Ai is in fact heresy .
@@Nickname863 Assuming they were actually intelligent, yes.
That's the reason it's gpt4O
@@salgadev GPT40k
tech priests would laugh at AI, but if they saw the pyramids they'd want to run into the tombs and find relics and fight necrons that doesn't exist in our tombs, thank god for that..
The role of a software developer isn't to implement features and fix bugs. The role of a software developer is to create a system (of features) that is maintainable even if said developer has left 5 years ago. The primary tools of a software developer aren't the coding language, the framework, the repository system, or anything so specifically. The primary tools of a software developer are selection and projection:
"Will I understand this solution in 3 months, 6 months, 3 years?"
"Will anyone other than me understand this solution in that time?"
"Will this work without me or anyone else babysitting every single step of the way to production and beyond?"
"Does this language/framework/process create more solutions than problems?"
"Does this only solve a very specific problem, or can this solve a general category of problems that really occurs?"
Ask yourself these question, and try to answer them without hubris and ego.
Dear lord 8:48 "No bias in AI" is the funniest thing I have seen from AI yet.
its human bias, I mean the bias it has is the algo which some human or humans coded.
its has no personal imagination or preference. it gives you an answer based on optimal statistics with greater chances of success.
in this case to get the job done. but this stats and probability it calculates in miliseconds all comes from some human algo that tells it what to do.
@@ricardomartinhodacruz I dont think this case reflects human bias. This is due to it being an LLM. Based off the text, it tries to predict based on all the data its ever seen. Thats all, it has no understanding.
it's human stupidity to assume that names in this task don't matter or mean anything. It was given extremely little context, and the names could very well mean something.
@@lystic9392 No this just shows that you cant ever rely on LLMs to reason. They cannot because they do not have understanding. They provide probabilties based off how each letter/word/token compares to every other it has ever seen (trained on). This is one of the simplest examples I have seen to convey this.
@@ricardomartinhodacruzautism
I am highly convinced that Autistic Intelligence will take my job first
Trust me you dont want to be autistic in a world that is rapidly changing
4:20 AI winter was actually a thing back in the 20th century. Also, there have been several growth periods and even more stagnation time inbetween.
Btw, before LLMs in 2020s, there was also boom in deep learning about a decade earlier (hardware got fast enough, we ditched complex activation functions instead going with fast stuff like ReLU and most importantly ideas like ResNet solved the major issues of backpropagation).
I got an ad for an AI CV enhancer site while prime talked about adding prompts to a CV
LMAO, the remove dog from the pic had me rolling.
I had to lookup the meaning of "piledrive", expecting some tech stuff to do with drivers and stacks... My shock when the first result is from webMD talking about adult fun time positions... my relief when the second result talks about wrestling (with a very cute 31 frame stickman animation), and my shock again when the third result is that it's the model name of the (selfaclaimed) most powerful air rifle...
And finally, a dictionary result talking about heavy duty machinery.
So yeah, TIL a new word... Anyway, now onto actually watching the video, not anymore to hear something about AI, but to find out which kind of piledriving the OP of the article was threathening (a good time?) with
Top tier comment
The spacing I wonder... Nonsexual usage i've always seen is "pile drive".
I don't believe you.
Just finished the video, i'm still not sure which form of piledriving OP meant
@justiny2552 ooh, yeah i'm Dutch so I cognate a lot of English terms, I wouldn't have even thought to split the term. Now I get results exclusively about the heavy duty machines to hammer pylons into the earth. Tempting rabbit hole that, on the history of constructions. Nothing about wrestling tho, so that one's ethologically closer to the naughty one. Wonder which one came first (not like that)
@bradweir3085 not that I particularly care if you believe me or not, am just procrastinating here; I've never come across the term, wrestling isn't that big outside the Americas. Or do you not believe that particular order of results, because I did lie a bit: in between the fourth and third results I mentioned was another adulttime related one. Cus I use Google engine through Startpage on a mobile network, I don't think there's personalisation outside country localisation going on, before people start claiming that about me lol
AI=Actually India!
As a guy who works in a giant indian enterprise, i manuvally do thing that i can automate in an hour
(infact, as my collage project, i automated entire work in my role just by making a chrome extention and a excel vba, and today i presented it)
Context: Google’s Amazon Go stores didn’t have AI auto-checkout, just powered by 1000 people in india.
"But AI is going to be everywhere soon"
Yeah, like Java newer than 8?
A lot of the criticism on LLM's boils down to one thing : some things are so obviously true that nobody in the history of mankind ever bothered to write it down. Yann LeCun gave this example : when there is a cup on the table and you move the table then the cup will move with it. How do you even begin to figure out what these baseline truths even are? Let alone write them all down and then train your AI on them first. And then also make sure it won't forget about it when you keep training it? There is only so much that language can do for us.
And he was wrong, try and ask any AI model this question and you will get the correct answer. His examples were wrong. Rational Animations did a good video on this topic, you should check them out.
It is especially true when it comes to culture and history.
People don’t write down what is so obvious to them that it doesn’t even occur to them to write down about it.
@@TheManinBlack9054 Your comprehension skills and imagination are inline with an AI
The point is what questions and information you do not realise simply do not get written down despite how intuitively we understand them. There is no single question that defeats this point of concern.
FYI: The owner of that blog and author of the article is Nikhil Suresh, not Ludic Mataroa.
Ludicity is the name of his blog and mataroa is a blogging platform he uses.
AI became quite lucrative well before LLMs. Expert systems have been controlling things like speech recognition and flight scheduling at airports for decades. What LLMs did is make AI mainstream in a popular way. When people say "AI Winter" what they really mean is that we don't have Data from Star Trek yet. Until we get Data, we'll always bein on the cusp of or going through an "AI Winter." But the truth is that people have been using and making money off of AI for a long time.
Facts
Lmao imagine investing yourself so much into a fictional TV show
When I became aware of the increasing role of AI in generating code, especially for learners, I knew it was a good time to learn how to program. Somebody is going to have to know how this stuff actually works, the OG chads won't live forever and if you are "learning" with this tech, you won't really know much besides how to proompt. This is great for people that can actually do the work as the skill divide becomes a yawning chasm. 😎
Bruh AI will be sentient and AGI creating new code
20:50 - Here's the problem: Details matter. It said "been assisting with strategic decisions". Using LLMs for ideas and not necessarily correct explanations on how things work, is exactly what they are good for. This can of course be used as an input for decisions, maybe often just as an inspiration.
I guess in the case of customer support LLMs could be used to protect the employee against the "impatient" customers. So here would be my suggestion : customer support gets to "rate" the customer at the end of the call / chat. Ratings are shared across companies. If a phone number / customer ID has bad enough ratings, the system defaults to giving them AI support, if the customer has good enough rating, they get a real person every time, considering that the pool of available agents is not much bigger. There would be a way to appeal the rating.
This way, overall, assholes are filtered out and I'm rewarded for talking to customer support like they're actual human beings by being given actual human beings
Brilliant
You know, I think this could work really well.
I still don't know why remote live support is still valued. If, in 2024, someone is using a computer but doesn't know how to ctrl+f a faq and has to get a call centre employee to do it for them, they are too stupid to do business with.
@telnobynoyator_6183 Sorry to disappoint you but I years ago I watched a docu about call centers and their shady tactics..
Allegedly, each caller receives a rating attached to their number: friendly callers get a positive mark, while frustrated callers receive a negative one.
Unfortunately, most large call centers operate under a single umbrella or utilize the same software, allowing unrelated service numbers to identify "troublemakers" in advance. They can choose to ignore these calls and move on to the next one.
The writer is the same writer who wrote "I accidentally saved half million dollars" the one you reacted to a few months ago. His entire blog is a gem.
Thank you for the kind words!
11:44 AI *can't* tell you the magical answers to the universe. Not until someone first feeds those answers to it.
And then, the AI (LLMs) will mix up some or all parts of those answers with all the crap non-answers it’s also been fed with. Hopefully you’ll be able to tell the difference.
I love laughing at the "just fine tune the model"... on what? the 2 files and 12 database records I have in my dev server? or the 100 files and 10,000 records in my production server littered with PII. Either way that's not enough data to fine tune anything.
Todoist is phenomenal - you write "meeting with Primeagen tomorrow at 15:30 !15mb" and it sets and event/task "meetings with primeagen" with a deadline tomorrow at 15:30 and a notification 15 minutes before the deadline. Works with other natural language phrases like "every other wednesday", "every last thursday of every month", etc.
Ya, but that kind of thing is really easy to program even without an LLM. Google calendar can do this
@@Jabberwockybird Todoist doesn't use LLMs for this. The LLM feature lets you write more complex queries over your tasks, but 99% of the value comes from what are probably a bunch of regexes.
About a month ago, the CEO of the company I work for basically went "Hey, you know that alert and suggestion system we asked you to implement? The one with zero user inputs, and can be done by using templates and swapping out a couple of words? Can you add an AI there?"
I'll not press 1, I'll press 2 for deep doubt for current AI. I have no idea what will be in the next 5 to 10 years, but the thing we have now is partly usabe, deeply unreliable, and hyped to high heavens.
Skynut.
I'm bothered by the sense of inevitability the tech world is putting on us.
"yeah, it hallucinates, but eventually you'll get over that."
The way I like to use the gippity for dev is by using it to explore unfamiliar problems. Not that it can solve them for me, but it sometimes pushes out solutions using techniques or libraries that i may not have even thought to google for because i dont know the domain well enough.
Often theres some code worth copying, I usually end up optimising, but it gives me a atarting point. Often a lot of the boilerplate I'll throw away.
38:53 Why do we do this to him? Why can't we just read our own articles? I don't have dyslexia. I can read pretty well.
I'm not going to stop, though. He's too entertaining.
"... add chatbot support to every application under the sun when half of the industry hasn't worked out how to test database backups regularly" made me felt so heard. I had a system stand for 3 weeks losing millions because database backups weren't checked. Had to manually recreate 3 months of orders. One week before that incident we had a meeting on how to utilize AI.
So I use LLM “whatever” to rewrite my cv to make it match a job description. Prospect employer uses the same LLM to calculate the match value of my cv in relation to the job description.
So, who won?
In fairness to AI in cybersecurity, it's a pretty serious threat in the context of your programmers using it to create massive security vulnerabilities in your own software and the utility in improving or mass producing phishing attempts.
39:20 same man, haha, can end up reading the same word so many times that it starts to not even look like a real word anymore and then nothing is real. good ol' adhd
35:14 I’ve seen one of the pitches for “LLM on PR” company. Their opening statement was- “currently we have GPT writing 10% of the code, soon it will be 90%”.
Every ad I got watching this was about one of those doomed AI use cases. Really put it into perspective.
36:20 if legislation becomes the spiritual equivalent of folk tales told by LLMs, will anyone ask "Wait, didn't we start writing laws down, so people don't just make sht up?"
prime being swooned by Australian use of curse words is adorable
42:05 My big thing is phind. It tells me what to do *IF* I ask the right question, gives me useless code, and tells me how to make the code as if it was good. I come away with sources, steps to start researching, and new concepts to figure out.
Phind has a lot of issues, but that's what helps. It's like talking to a toddler that has access to the sparknotes of all things. One day, I hope to make my own that feels like a middleschooler or better, and phind gave me some idea how to do that, once I figure our how. 😅
If you have a good disaster recovery plan, testing backups usually happens per quarter or around a major update or implementation.
Synergy greg is a genius, he hired Tom
"Its not a competition... but he's winning?!" HAHAHA I fucked died laughing.
I love this, my inner troll gets excited to talk more about AI, seems some are really angry
34:50 I got a hot take: Asking an AI to write unit tests for the code that the same AI wrote is like asking a corporate entity to investigate itself for any wrongdoing. You can only trust it to do whatever it takes to make itself look good, in spite of any and all flaws that indicate otherwise.
The engineer with their passwords in a TXT file is the reason your security guy drinks😢
He just need a reason
I started reading LOTR: fellowship a few weeks ago (got to Moria but haven't picked it up in a while). Interestingly, there are a lot of songs and poems in the book that I did not expect. Some of them are quite touching or fun, but some of them are pretty throwaway lol.
Flip deserves more love, piledriving the edits 💕
I have laughed so hard in this video, thank you so much for reading the article.
The besttt youtube dev channel for continuous learning without much effort!
“Just let a man have pancakes” - sums up the actual crisis we face, given a properly realistic context. If you have to ask, you’ll never know.
"It's predictive text on GPUs" was a great comment.
Literally dying of laughter about how companies solve problems. "We talk to each other to solve problems." bahahahahahahaha🤣
LOL you're break down from the tripple are's was hilarious
Those two magical letters are gonna get piledriven on faster than that elevator guy's face in "Drive" and there's nothing we do about it except make it super uncool.
I think one of the overlooked things working in Copilot and other coding assistant AIs favor, is what happens when ya step outside the coding for money setting, strangely. It allows specializations created by other industries carry people so much further solo, letting people focus on the things they are good at during the early stages of a project while letting the AI worry about the code they don't really understand. Once that website is a proof of concept, or that app is an alpha, or the game is functionally playable long enough to show off a bit, and that proof of concept is figured out, then they can hope to find the funds they never had before and hire somebody to patch the glaring security holes and fix the idiot mistakes, or lets be honest in a lot of cases gut the entire mess and do it right. But still, a lot of good ideas likely die because a person lacks the time to learn to code or just aint got the head for it. And yes, some never will and will attempt to keep going with AI generated code and they will suffer as a result... so like, be extra careful who ya give that credit card number to for a bit I guess?
43:00
This is so fucking real. Starting out in 2022 I started learning and taking courses in school that taught and utilized Python, learned a fuck ton.
Discovered GPT and Co-Pilot later that summer and over the past year and a half my brain just started leaking all of the useful problem solving skills and syntax out.
Recently turned co-pilot off and I just spent 10 minutes staring at the fucking screen. The shit is poison for your brain.
Human strongly disagree: Like the inventor of the insulin production that donated the recipe for it to be available to everyone. But seems the news did not reach the US of fing A and you ends up the only country in the world to have people die because insulin is not affordable xD
43:02 - "I do not learn with my eye. He who learns with his eye has forgotten the face of his father. When I learn, I learn with my hand."
Flip, that zoom in the intro, chef’s kiss, bruh, chef’s kiss.
Thanks!
4:35 Good argument, but I'd like to inform you that an 80+ year old ex-judge (fairly high in the court order as well) I got to interact with was pretty certain that the development will accelerate and we will all be soon (within the next 10 years) be crushed under the feet of skynet. To be fair to him though, who knows maybe he's right 🤷♂
In the company I work at we test our backups every 10min and we backup every client database hourly 💪😎
Based and ops-pilled
So far the best use I got out of AI is making RGB lights around my house shine in color and rhythm of my music
Every time you hear Prime saying "let's see", you know he misread something :D
I was hired to produce AI content for a company.
I said it won't work as they want it to. Hired me anyway.
I was fired 9 months later because it wouldn't work the way they wanted it to.
Last week they wanted to re-hire me because they can't find anyone to make it work like they want it to.
This guy isn't a data scientist, he's a poet.
watch 10:35 - 11:05
if you dont understand by that statement alone...idk what to tell you.
Comfort zone? Nah, Prime lives in the trauma zone, coronating the forbidden dream of reading articles to an audience and having fun with it!
FLIP! Great editing, the squeaky mic movements were masterful. A++
PiledriveMeDaddyGen
Here's something else that's problematic about AI usage - even in the places where there's a good use case, it can still go out of bounds from your original intentions in ways you couldn't have imagined.
Prime makes my world better, thank you bro!
Less than 2 minutes and already 9 thumbs? Maybe it was watched by AI 🤣
Nah the title slaps
Maybe people watched it on stream..
pre-watched, pre-read
Everything by me is liked immediately unless they prove unqualified (everyone is free unless proven guilty)
Those thumbs are clearly human
AI can't generate fingers
On the topic of not using copilot to learn and learning by doing, there is a quote that I absolutely love. I am a mathematics student and whenever someone asks me how I managed to become good at math I tell them this quote: "Mathematics is learnt by banging your head against a wall". Math can be so difficult that the ONLY way you can learn it is by closing the textbook and actually trying. You may fail and eventually need to look up the solution, but then you move on to the next question and repeat. Then you come back a week later and redo everything without looking up the solutions. You learn by doing and your research (or as you call it, review) is simply a tool to can aid in that learning; however, you cannot learn without actually doing -- with or without that review.
about CoPilot, this is VERY TRUE! .. for anyone can't make the connection and is older than 30 can conform, driving with and without street Navigation, has the same effect, if the navi keeps telling you turn by turn where to go, once the device is off, you will ba really lost, because you never rmemorized the streets but were just the bot that dives after being told to.
One man woke up and chose violence.
10:42
"I was right until I was wrong"
Woah.
I worked for a company that couldn’t manage to roll out office 365 until 2022 and they use excel sheets to schedule manufacturing capacity. There is no way they are using AI effectively within the next decade.
What's wrong with Excel? It's pretty good at modeling most data, and it's much more intuitive than programming languages for most people.
(I say this as a programmer)
I setup automated backups for our DB at this startup i was at, and left it running. At some point it stopped working, and there was apparently no backups for 3 months lmao. Don't even remember what the issue was. While i was still there, would just checked the bucket every few days to make sure it was still running.
0:39 zoom in, enhance
Flip screwed up.
44:15 Zero-trust security architecture? very interesting. I was thinking since knew about zero-knewledge proofs, if that can be used to build secure systems with no potential of leaking data to corporations or 3rd parties
I am sick and tired hearing about AI.
As a guy who brainwashed himself with ai content and building python stuff around llms for 1,5 years i agree. It is sickening and useless😅 also it seems to make your hair go grey
@@denisblack9897it is useful for what it is. The issue is that some tech bros have decided ir has to be the ultimate solution to every problem
I think GPT really overshadows everything else AI can do, because it's the most visible to your average person. Stuff like weather predictions, replacing repetitive manual labor tasks (why do they always assume it's tech jobs that get replaced?), self driving cars, and so on. Perhaps the thing I'm most interested in, is if AI could intelligently decompile software and make it human readable with insightful variable names. That would be an absolute nightmare. I assume the NSA or some TLA is working on that tool as we speak.
My coworker uploads specific IC manuals (mcus, socs, FPGAs, etc) and then asks it specific questions
7:25 👈Prime spittin' mad truth right there! That should be on a t-shirt
To live for something you need something to live for.