The biggest misconception about AI is how close it is to being really useful. We are currently at a peak hype cycle for AI. Remember 10 years ago when everyone was saying self driving cars were just around the corner and would take over the world? Remember 5 years ago when the world was about to put everything on the block chain? If your company didn't mention crypto somehow you were not trying hard enough. That is the context of the current AI hype. It is very likely that in another 5-10 years we haven't made that much progress. Models can even easily regress if they are fed data produced by other models.
The current generation of LLMs are really good at auto complete for problems that have been solved a million times. Even then they include the same human errors that can be found when you search for a solution with google. They are only as good as the data they are fed which is often only partially correct. Also, they are just not very good at solving new and unique problems.
If you start using AI, at first you think it is amazing, then the longer you use it you realize how many problems it has.
What they are really good at is generating solutions where they just need to be close enough. Such as generating images that look mostly like what you describe.
Maybe Korean ones don't, but American AI engineers get paid and treated very well. I've seen many companies offering $200k+ and crazy benefits (free IVF, free pet daycare, free pet insurance, 6 weeks PTO) for 5 years of experience.
The people putting together the training data don't always get paid well though - I've heard that's often people getting paid very little in poorer countries to do a pretty boring job.
Also, most of the AI engineers I know are working ~35 hour weeks in nice offices, all between 9 and 5. If you're the person training the AI, it's a fairly glamourous and very well-paying job.
It's not a super sexy high paying field. It's a grind in a dark room that leads most of the engineers into depression, eating disorders, and addiction.
I was a network engineer for 20 years, now retired. Oh, as rothbard said, it’s not exactly a glamorous job. It seems like making good money from home remoting into the little devices that run the internet is a cool job. You work at night, you work alone, and you deal with the most hideously boring data you can imagine. And when you report to the testing center to take your engineering exams, you look around the waiting room and see the most stressed out people in the world. I mean, people actually kill themselves for failing these exams.
The biggest misconception is that AI, AS IT IS NOW CONSTITUTED, will become sentient. What I mean is the large language models, the generative AI’s like ChatGPT, are actually beginning to reach the limits of the data they can learn (like literally, all the data in the entire world). They are now beginning to use simulated data and problems are developing with feedback loops of contaminated data.
OK, enough with the gobbledygook talk. Plain English time: It’s looking like generative AI is collapsing long before it reaches the sentience everyone was talking about.
But, hundreds of thousands of my IT peers have already lost their jobs due to this narrow AI being REALLY GOOD at very specific tasks. That’s only going to increase.
What no one is talking about is the growth of self-learning devices. Weird, strangely shaped robots that learn how to move and do things. The misconception here is these will just replace factory workers. Maybe initially, but, I think this is the vein of AI where true sentience may emerges. Little robots learning about the world just like we did, only, much, much faster.
AI Engineers/ ML Engineers/ Data Scientists aren't Network Engineers (not saying Network Engineers don't use AI/ML), and I haven't met a single one who works nights.
The current generation of LLMs are really good at auto complete for problems that have been solved a million times. Even then they include the same human errors that can be found when you search for a solution with google. They are only as good as the data they are fed which is often only partially correct. Also, they are just not very good at solving new and unique problems.
I've noticed people who were never good at finding information, like people who didn't know to search for "site:stackexchange.com" and then a few keywords, are championing AI a lot. It's good-ish at regurgitating information for well-solved problems, but it's not always right.
If I didn't know the first thing about RegEx AND I was bad at Googling stuff and teaching myself stuff, then ChatGPT would be mindblowing because it could teach me the basics of writing regular expressions in a few minutes and even help me write a very basic one, maybe.
AI Engineers/ ML Engineers/ Data Scientists aren't Network Engineers (not saying Network Engineers don't use AI/ML), and I haven't met a single one who works nights.
Totally agree. For folks that don’t know, “network engineer” included, for me, maintaining a bunch servers. We worked crappy hours because you just can’t take any of that down in the day to work on it unless it was an emergency. I/We were the “grunts” of the IT world. I did get to use some AI when working on security policies, but not nearly as much as I wanted.
Anyway, TMI, the IT universe is vast and confusing to most folks.
Momma’s don’t let your children grow up to be network engineers. Go into AI/ML/Data!
I've found the best LLMs (ChatGPT isn't the best for coding anymore) are usually faster than searching stackexchange if it's something that has been asked multiple times on stackexchange or explained many times in textbooks. I think it's now extremely good at giving info for well-solved problems, and even pretty good at algorithms. A good LLM can summarize all the good info on sites like stackexchange, and then explain how it applies to your specific code. It's also very good at some corporate creative tasks - writing ad copy for example.
But it's not always great when your issue is different from those already on the internet- for example it's still pretty bad at embedded code because only 10 people have ever posted about the microcontroller you're using.
And over-relying on it can make you lazy. For example, if you only need to do work in c++ once every ~2 months, you might find you can usually get working c++ code just by prompting an LLM. But this is a problem because it means you don't have an amazing understanding of your own code. What will happen if the LLM introduces a bug that you can't catch because you still don't know c++?
I work in venture capital and I can tell you 99% of “AI” companies are not AI at all. Simple pdf scrapers are called AI. Simple VBA programs are called AI. It’s the most overused term ever.
This.
But also the general publics' lack of knowledge and loosely throwing around these terms really muddies the water. In 2024, the news presenters, papers, blogs, youtube channels are all throwing AI on everything, and it's about as accurate as all of us sat around in 2004 asking how someone made a powerpoint and being told "well, technology did it".... yes but what technology, what machine, a computer?, what program, what did it do, how
Alternatively imagine being introduced to a friend of a friend who says in their free time they 'do sport'.. as if that gets us any closer at all to understanding wtf they're talking about. AI is a horribly vague umbrella term and once people start understanding what it covers, we'll start to get a better understanding of its applications and use cases, and stop imagining Sky-Net or flying cars or killer robots, and also be able to realise what actually isnt AI at all.
I teach English to Korean AI engineers so I am tangently in that industry.
It's not a super sexy high paying field. It's a grind in a dark room that leads most of the engineers into depression, eating disorders, and addiction.
It's not a super sexy high paying field. It's a grind in a dark room that leads most of the engineers into depression, eating disorders, and addiction.
I was a network engineer for 20 years, now retired. Oh, as rothbard said, it’s not exactly a glamorous job. It seems like making good money from home remoting into the little devices that run the internet is a cool job. You work at night, you work alone, and you deal with the most hideously boring data you can imagine. And when you report to the testing center to take your engineering exams, you look around the waiting room and see the most stressed out people in the world. I mean, people actually kill themselves for failing these exams.
The biggest misconception is that AI, AS IT IS NOW CONSTITUTED, will become sentient. What I mean is the large language models, the generative AI’s like ChatGPT, are actually beginning to reach the limits of the data they can learn (like literally, all the data in the entire world). They are now beginning to use simulated data and problems are developing with feedback loops of contaminated data.
OK, enough with the gobbledygook talk. Plain English time: It’s looking like generative AI is collapsing long before it reaches the sentience everyone was talking about.
But, hundreds of thousands of my IT peers have already lost their jobs due to this narrow AI being REALLY GOOD at very specific tasks. That’s only going to increase.
What no one is talking about is the growth of self-learning devices. Weird, strangely shaped robots that learn how to move and do things. The misconception here is these will just replace factory workers. Maybe initially, but, I think this is the vein of AI where true sentience may emerges. Little robots learning about the world just like we did, only, much, much faster.
I've been in IT for 30 years and do a lot of network management and server stuff. I think AI is going to destroy jobs like mine. I have a lot of arcane knowledge for setting up switches and devices that rely on simply understanding the vocabulary and being able to enter text-based commands on switching gear. Some of it (Meraki) is also GUI based. Point being, if I am setting up a network and I need to create a DHCP scope, I know exactly what that is and how to do it. Once AI gets its hands on this, I see no reason an end user (or regular office worker) couldn't say "I need this network to distribute unique internal IP addresses to the ten computers on this network, but to use this one gateway address for internet traffic" and the switch will know how to translate that into the appropriate commands and do the setup itself. Fortunately, I'm two years away from retirement, so there's little impact on me if this becomes reality.
I don’t work directly with AI but for one of the companies at the forefront of development. They hosted a “AI safety” learning session for those interested about the work.
Basically there is zero safety plan. AI is unpredictable and will be wickedly smart. Not sure why they hosted the session because it was not encouraging.
I was a network engineer for 20 years, now retired. Oh, as rothbard said, it’s not exactly a glamorous job. It seems like making good money from home remoting into the little devices that run the internet is a cool job. You work at night, you work alone, and you deal with the most hideously boring data you can imagine. And when you report to the testing center to take your engineering exams, you look around the waiting room and see the most stressed out people in the world. I mean, people actually kill themselves for failing these exams.
The biggest misconception is that AI, AS IT IS NOW CONSTITUTED, will become sentient. What I mean is the large language models, the generative AI’s like ChatGPT, are actually beginning to reach the limits of the data they can learn (like literally, all the data in the entire world). They are now beginning to use simulated data and problems are developing with feedback loops of contaminated data.
OK, enough with the gobbledygook talk. Plain English time: It’s looking like generative AI is collapsing long before it reaches the sentience everyone was talking about.
But, hundreds of thousands of my IT peers have already lost their jobs due to this narrow AI being REALLY GOOD at very specific tasks. That’s only going to increase.
What no one is talking about is the growth of self-learning devices. Weird, strangely shaped robots that learn how to move and do things. The misconception here is these will just replace factory workers. Maybe initially, but, I think this is the vein of AI where true sentience may emerges. Little robots learning about the world just like we did, only, much, much faster.
I've been in IT for 30 years and do a lot of network management and server stuff. I think AI is going to destroy jobs like mine. I have a lot of arcane knowledge for setting up switches and devices that rely on simply understanding the vocabulary and being able to enter text-based commands on switching gear. Some of it (Meraki) is also GUI based. Point being, if I am setting up a network and I need to create a DHCP scope, I know exactly what that is and how to do it. Once AI gets its hands on this, I see no reason an end user (or regular office worker) couldn't say "I need this network to distribute unique internal IP addresses to the ten computers on this network, but to use this one gateway address for internet traffic" and the switch will know how to translate that into the appropriate commands and do the setup itself. Fortunately, I'm two years away from retirement, so there's little impact on me if this becomes reality.
Anyone who has ever used an automated customer service chatbot can see why AI is no significant danger to IT or other similar professions. AI can be "taught" to follow a basic logic tree for troubleshooting, analysis and learning, but it fundamentally lacks many true human intuitive qualities that knows when to cut past needless logic trees and processes to more quickly address advanced, complex issues.
AI at best will just waste needless time to get to a solution a skilled human can identify and execute right away. More often it will just frustrate engineers and end users alike as it leads you into the dead end of a preconstructed rat maze.
Like self driving cars, tech leadership's egos are substantially overrating AI's future impact on society. It will help us in untold ways sure, maybe hurt us in others, but there's too many confounding and unworkable holes in its fundamental limitations for it to ever workably replace all human skill.
Help us build the best running shoe review site for a chance to win a LetsRun t-shirt.Help us build the best running shoe review site for a chance to win one of 10 LetsRun t-shirts.