There is this seeming need to discredit AI from some people that goes overboard. Some friends and family who have never really used LLMs outside of Google search feel compelled to tell me how bad it is.
But generative AIs are really good at tasks I wouldn’t have imagined a computer doing just a few year ago. Even if they plateaued in place where they are right now it would lead to major shakeups in humanity’s current workflow. It’s not just hype.
The part that is over hyped is companies trying to jump the gun and wholesale replace workers with unproven AI substitutes. And of course the companies who try to shove AI where it doesn’t really fit, like AI enabled fridges and toasters.
“AI technology is exceptionally expensive, and to justify those costs, the technology must be able to solve complex problems, which it isn’t designed to do.”
Generative AI can indeed do impressive things from a technical standpoint, but not enough revenue has been generated so far to offset the enormous costs. Like for other technologies, It might just take time (remember how many billions Amazon burned before turning into a cash-generating machine? And Uber has also just started turning some profit) + a great deal of enshittification once more people and companies are dependent.
Or it might just be a bubble.
As humans we’re not great at predicting these things including of course me. My personal prediction? A few companies will make money, especially the ones that start selling AI as a service at increasingly high costs, many others will fail and both AI enthusiasts and detractors will claim they were right all along.
Even if they plateaued in place where they are right now it would lead to major shakeups in humanity’s current workflow
Like which one? Because it’s now 2 years we have chatGPT and already quite a lot of (good?) models.
Which shakeup do you think is happening or going to happen?
No, not that either. Unless you consider “use LLM to summarize the changes/errors/inaccuracies, then have a human read the whole thing again” an improvement over “just have a human read the whole thing”.
Because LLM will do all these things:
point you toward issues
point you toward non-issues
not point you toward issues
change stuff even when “instructed” not to
If there is one thing you don’t want to throw an LLM at without full, unbiased review, it’s documents where the wording is legally binding. And if you have to do a full, unbiased review to begin with, where you can’t even trust your tool to have highlighted all the important parts, you may as well not bother with the tool.
I quit my previous job in part because I couldn’t deal with the influx of terrible, unreliable, dangerous, bloated, nonsensical, not even working code that was suddenly pushed into one of the projects I was working on. That project is now completely dead, they froze it on some arbitrary version.
When junior dev makes a mistake, you can explain it to them and they will not make it again. When they use llm to make a mistake, there is nothing to explain to anyone.
I compare this shake more to an earthquake than to anything positive you can associate with shaking.
And so, the problem wasn’t the ai/llm, it was the person who said “looks good” without even looking at the generated code, and then the person who read that pull request and said, again without reading the code, “lgtm”.
If you have good policies then it doesn’t matter how many bad practice’s are used, it still won’t be merged.
The only overhead is that you have to read all the requests but if it’s an internal project then telling everyone to read and understand their code shouldn’t be the issue.
I hardly see it changed to be honest. I work in the field too and I can imagine LLMs being good at producing decent boilerplate straight out of documentation, but nothing more complex than that.
I often use LLMs to work on my personal projects and - for example - often Claude or ChatGPT 4o spit out programs that don’t compile, use inexistent functions, are bloated etc.
Possibly for languages with more training (like Python) they do better, but I can’t see it as a “radical change” and more like a well configured snippet plugin and auto complete feature.
LLMs can’t count, can’t analyze novel problems (by definition) and provide innovative solutions…why would they radically change programming?
That is my experience, it’s generally quite decent for small and simple stuff (as I said, distillation of documentation). I use it for rust, where I am sure the training material was much smaller than other languages. It’s not a matter a prompting though, it’s not my prompt that makes it hallucinate functions that don’t exist in libraries or make it write code that doesn’t compile, it’s a feature of the technology itself.
GPTs are statistical text generators after all, they don’t “understand” the problem.
It’s also pretty young, human toddlers hallucinate and make things up. Adults too. Even experts are known to fall prey to bias and misconception.
I don’t think we know nearly enough about the actual architecture of human intelligence to start asserting an understanding of “understanding”. I think it’s a bit foolish to claim with certainty that LLMs in a MoE framework with self-review fundamentally can’t get there. Unless you can show me, materially, how human “understanding” functions, we’re just speculating on an immature technology.
As much as I agree with you, humans can learn a bunch of stuff without first learning the content of the whole internet and without the computing power of a datacenter or consuming the energy of Belgium. Humans learn to count at an early age too, for example.
I would say that the burden of proof is therefore reversed. Unless you demonstrate that this technology doesn’t have the natural and inherent limits that statistical text generators (or pixel) have, we can assume that our mind works differently.
Also you say immature technology but this technology is not fundamentally (I.e. in terms of principle) different from what Weizenabum’s ELIZA in the '60s. We might have refined model and thrown a ton of data and computing power at it, but we are still talking of programs that use similar principles.
So yeah, we don’t understand human intelligence but we can appreciate certain features that absolutely lack on GPTs, like a concept of truth that for humans is natural.
humans can learn a bunch of stuff without first learning the content of the whole internet and without the computing power of a datacenter or consuming the energy of Belgium. Humans learn to count at an early age too, for example.
I suspect that if you took into consideration the millions of generations of evolution that “trained” the basic architecture of our brains, that advantage would shrink considerably.
I would say that the burden of proof is therefore reversed. Unless you demonstrate that this technology doesn’t have the natural and inherent limits that statistical text generators (or pixel) have, we can assume that our mind works differently.
I disagree. I’d argue evidence suggests we’re just a more sophisticated version of a similar principle, refined over billions of years. We learn facts by rote, and learn similarities by rote until we develop enough statistical text (or audio) correlations to “understand” the world.
Conversations are a slightly meandering chain of statistically derived cliches. English adjective order is universally “understood” by native speakers based purely on what sounds right, without actually being able to explain why (unless you’re a big grammar nerd). More complex conversations might seem novel, but they’re just a regurgitation of rote memorized facts and phrases strung together in a way that seems appropriate to the conversation based on statistical experience with past conversations.
Also you say immature technology but this technology is not fundamentally (I.e. in terms of principle) different from what Weizenabum’s ELIZA in the '60s. We might have refined model and thrown a ton of data and computing power at it, but we are still talking of programs that use similar principles.
As with the evolution of our brains, which have operated on basically the same principles for hundreds of millions of years. The special sauce between human intelligence and a flatworm’s is a refined model.
So yeah, we don’t understand human intelligence but we can appreciate certain features that absolutely lack on GPTs, like a concept of truth that for humans is natural.
I’m not sure you can claim that absolutely. That kind of feature is an internal experience, you can’t really confirm or deny if a GPT has something similar. Besides, humans have a pretty tenuous relationship with the concept of truth. There are certainly humans that consider objective falsehoods to be Truth.
No actually it has changed pretty fundamentally. These aren’t simply a bunch of FCNs put together. Look up what a transformer is, that was one of the major breakthroughs that made modern LLMs possible.
Exactly this. Things have already changed and are changing as more and more people learn how and where to use these technologies. I have seen even teachers use this stuff who have limited grasp of technology in general.
My kid’s teachers had what I thought was a fantastic approach - have the kids write an outline. Use an LLM to generate an essay from that outline, then critique the essay
Computers have always been good at pattern recognition. This isn’t new. LLM are not a type of actual AI. They are programs capable of recognizing patterns and Loosely reproducing them in semi randomized ways. The reason these so-called generative AI Solutions have trouble generating the right number of fingers. Is not only because they have no idea how many fingers a person is supposed to have. They have no idea what a finger is.
The same goes for code completion. They will just generate something that fills the pattern they’re told to look for. It doesn’t matter if it’s right or wrong. Because they have no concept of what is right or wrong Beyond fitting the pattern. Not to mention that we’ve had code completion software for over a decade at this point. Llms do it less efficiently and less reliably. The only upside of them is that sometimes they can recognize and suggest a pattern that those programming the other coding helpers might have missed. Outside of that. Such as generating act like whole blocks of code or even entire programs. You can’t even get an llm to reliably spit out a hello world program.
I never know what to think when I come across a comment like this one—which does describe, even if only at a surface level, how an LLM works—with 50% downvotes. Like, are people angry at reality, is that it?
The part that is over hyped is companies trying to jump the gun and wholesale replace workers with unproven AI substitutes. And of course the companies who try to shove AI where it doesn’t really fit, like AI enabled fridges and toasters.
This is literally the hype. This is the hype that is dying and needs to die. Because generative AI is a tool with fairly specific uses. But it is being marketed by literally everyone who has it as General AI that can “DO ALL THE THINGS!” which it’s not and never will be.
The obsession with replacing workers with AI isn’t going to die. It’s too late. The large financial company that I work for has been obsessively tracking hours saved in developer time with GitHub Copilot. I’m an older developer and I was warned this week that my job will be eliminated soon.
So the company that is obsessed with money that you work for has discovered a way to (they think) make more money by getting rid of you and you’re surprised by this?
At least you’ve been forewarned. Take the opportunity to abandon ship. Don’t be the last one standing when the music stops.
I never said that I was surprised. I just wanted to point out that many companies like my own are already making significant changes to how they hire and fire. They need to justify their large investment in AI even though we know the tech isn’t there yet.
See now, I would prefer AI in my toaster. It should be able to learn to adjust the cook time to what I want no matter what type of bread I put in it. Though is that realky AI? It could be. Same with my fridge. Learn what gets used and what doesn’t. Then give my wife the numbers on that damn clear box of salad she buys at costco everytime, which take up a ton of space and always goes bad before she eats even 5% of it. These would be practical benefits to the crap that is day to day life. And far more impactful then search results I can’t trust.
There’s a good point here that like about 80% of what we’re calling AI right now… isn’t even AI or even LLM. It’s just… algorithm, code, plain old math. I’m pretty sure someone is going to refer to a calculator as AI soon. “Wow, it knows math! Just like a person! Amazing technology!”
(That’s putting aside the very question of whether LLMs should even qualify as AIs at all.)
In my professional experience, AI seems to be just a faster way to generate an algorithm that is really hard to debug. Though I am dev-ops/sre so I am not as deep in it as the devs.
I remined of the time researchers used an evolutionary algorithm to devise a circuit that would emit a tone on certain audio inputs and not on others. They examined the resulting circuit and found an extra vestigial bit, but when they cut it off, the chip stopped working. So they re-enabled it. Then they wanted to show off their research at a panel, and at the panel it completely failed. Dismayed they brought it back to their lab to figure out why it stopped working, and it suddenly started working fine.
After a LOT of troubleshooting they eventually discovered that the circuit was generating the tone by using the extra vestigial bit as an antenna that picked up emissions from a CRT in the lab and downconverted it to the desired tone frequency. Turn of the antenna, no signal. Take the chip away from that CRT, no signal.
That’s what I expect LLMs will make. Complex, arcane spaghetti stuff that works but if you look at it funny it won’t work anymore, and nobody knows how it works at all.
There is this seeming need to discredit AI from some people that goes overboard. Some friends and family who have never really used LLMs outside of Google search feel compelled to tell me how bad it is.
But generative AIs are really good at tasks I wouldn’t have imagined a computer doing just a few year ago. Even if they plateaued in place where they are right now it would lead to major shakeups in humanity’s current workflow. It’s not just hype.
The part that is over hyped is companies trying to jump the gun and wholesale replace workers with unproven AI substitutes. And of course the companies who try to shove AI where it doesn’t really fit, like AI enabled fridges and toasters.
Goldman Sachs, quote from the article:
Generative AI can indeed do impressive things from a technical standpoint, but not enough revenue has been generated so far to offset the enormous costs. Like for other technologies, It might just take time (remember how many billions Amazon burned before turning into a cash-generating machine? And Uber has also just started turning some profit) + a great deal of enshittification once more people and companies are dependent. Or it might just be a bubble.
As humans we’re not great at predicting these things including of course me. My personal prediction? A few companies will make money, especially the ones that start selling AI as a service at increasingly high costs, many others will fail and both AI enthusiasts and detractors will claim they were right all along.
Like which one? Because it’s now 2 years we have chatGPT and already quite a lot of (good?) models. Which shakeup do you think is happening or going to happen?
Review of legal documents.
Oh boy…what can possibly go wrong for documents where small minutiae like wording can make a huge difference.
Creating legal documents, no. Reviewing legal documents for errors and inaccuracies totally.
No, not that either. Unless you consider “use LLM to summarize the changes/errors/inaccuracies, then have a human read the whole thing again” an improvement over “just have a human read the whole thing”.
Because LLM will do all these things:
If there is one thing you don’t want to throw an LLM at without full, unbiased review, it’s documents where the wording is legally binding. And if you have to do a full, unbiased review to begin with, where you can’t even trust your tool to have highlighted all the important parts, you may as well not bother with the tool.
Computer programming has radically changed. Huge help having llm auto complete and chat built in. IDEs like Cursor and Windsurf.
I’ve been a developer for 35 years. This is shaking it up as much as the internet did.
@remindme@mstdn.social 1 year. Let me know about the seachange of new 10x transform based programmers that have automated me out of a job.
I quit my previous job in part because I couldn’t deal with the influx of terrible, unreliable, dangerous, bloated, nonsensical, not even working code that was suddenly pushed into one of the projects I was working on. That project is now completely dead, they froze it on some arbitrary version.
When junior dev makes a mistake, you can explain it to them and they will not make it again. When they use llm to make a mistake, there is nothing to explain to anyone.
I compare this shake more to an earthquake than to anything positive you can associate with shaking.
And so, the problem wasn’t the ai/llm, it was the person who said “looks good” without even looking at the generated code, and then the person who read that pull request and said, again without reading the code, “lgtm”.
If you have good policies then it doesn’t matter how many bad practice’s are used, it still won’t be merged.
The only overhead is that you have to read all the requests but if it’s an internal project then telling everyone to read and understand their code shouldn’t be the issue.
I hardly see it changed to be honest. I work in the field too and I can imagine LLMs being good at producing decent boilerplate straight out of documentation, but nothing more complex than that.
I often use LLMs to work on my personal projects and - for example - often Claude or ChatGPT 4o spit out programs that don’t compile, use inexistent functions, are bloated etc. Possibly for languages with more training (like Python) they do better, but I can’t see it as a “radical change” and more like a well configured snippet plugin and auto complete feature.
LLMs can’t count, can’t analyze novel problems (by definition) and provide innovative solutions…why would they radically change programming?
ChatGPT 4o isn’t even the most advanced model, yet I have seen it do things you say it can’t. Maybe work on your prompting.
That is my experience, it’s generally quite decent for small and simple stuff (as I said, distillation of documentation). I use it for rust, where I am sure the training material was much smaller than other languages. It’s not a matter a prompting though, it’s not my prompt that makes it hallucinate functions that don’t exist in libraries or make it write code that doesn’t compile, it’s a feature of the technology itself.
GPTs are statistical text generators after all, they don’t “understand” the problem.
It’s also pretty young, human toddlers hallucinate and make things up. Adults too. Even experts are known to fall prey to bias and misconception.
I don’t think we know nearly enough about the actual architecture of human intelligence to start asserting an understanding of “understanding”. I think it’s a bit foolish to claim with certainty that LLMs in a MoE framework with self-review fundamentally can’t get there. Unless you can show me, materially, how human “understanding” functions, we’re just speculating on an immature technology.
As much as I agree with you, humans can learn a bunch of stuff without first learning the content of the whole internet and without the computing power of a datacenter or consuming the energy of Belgium. Humans learn to count at an early age too, for example.
I would say that the burden of proof is therefore reversed. Unless you demonstrate that this technology doesn’t have the natural and inherent limits that statistical text generators (or pixel) have, we can assume that our mind works differently.
Also you say immature technology but this technology is not fundamentally (I.e. in terms of principle) different from what Weizenabum’s ELIZA in the '60s. We might have refined model and thrown a ton of data and computing power at it, but we are still talking of programs that use similar principles.
So yeah, we don’t understand human intelligence but we can appreciate certain features that absolutely lack on GPTs, like a concept of truth that for humans is natural.
I suspect that if you took into consideration the millions of generations of evolution that “trained” the basic architecture of our brains, that advantage would shrink considerably.
I disagree. I’d argue evidence suggests we’re just a more sophisticated version of a similar principle, refined over billions of years. We learn facts by rote, and learn similarities by rote until we develop enough statistical text (or audio) correlations to “understand” the world.
Conversations are a slightly meandering chain of statistically derived cliches. English adjective order is universally “understood” by native speakers based purely on what sounds right, without actually being able to explain why (unless you’re a big grammar nerd). More complex conversations might seem novel, but they’re just a regurgitation of rote memorized facts and phrases strung together in a way that seems appropriate to the conversation based on statistical experience with past conversations.
As with the evolution of our brains, which have operated on basically the same principles for hundreds of millions of years. The special sauce between human intelligence and a flatworm’s is a refined model.
I’m not sure you can claim that absolutely. That kind of feature is an internal experience, you can’t really confirm or deny if a GPT has something similar. Besides, humans have a pretty tenuous relationship with the concept of truth. There are certainly humans that consider objective falsehoods to be Truth.
No actually it has changed pretty fundamentally. These aren’t simply a bunch of FCNs put together. Look up what a transformer is, that was one of the major breakthroughs that made modern LLMs possible.
Exactly this. Things have already changed and are changing as more and more people learn how and where to use these technologies. I have seen even teachers use this stuff who have limited grasp of technology in general.
My kid’s teachers had what I thought was a fantastic approach - have the kids write an outline. Use an LLM to generate an essay from that outline, then critique the essay
Computers have always been good at pattern recognition. This isn’t new. LLM are not a type of actual AI. They are programs capable of recognizing patterns and Loosely reproducing them in semi randomized ways. The reason these so-called generative AI Solutions have trouble generating the right number of fingers. Is not only because they have no idea how many fingers a person is supposed to have. They have no idea what a finger is.
The same goes for code completion. They will just generate something that fills the pattern they’re told to look for. It doesn’t matter if it’s right or wrong. Because they have no concept of what is right or wrong Beyond fitting the pattern. Not to mention that we’ve had code completion software for over a decade at this point. Llms do it less efficiently and less reliably. The only upside of them is that sometimes they can recognize and suggest a pattern that those programming the other coding helpers might have missed. Outside of that. Such as generating act like whole blocks of code or even entire programs. You can’t even get an llm to reliably spit out a hello world program.
I never know what to think when I come across a comment like this one—which does describe, even if only at a surface level, how an LLM works—with 50% downvotes. Like, are people angry at reality, is that it?
Downvoting someone on the Internet is easier than tangentially modifying reality in a measurable way
Downvoting sounds like a task that’s ripe for automation with AI!
This is literally the hype. This is the hype that is dying and needs to die. Because generative AI is a tool with fairly specific uses. But it is being marketed by literally everyone who has it as General AI that can “DO ALL THE THINGS!” which it’s not and never will be.
The obsession with replacing workers with AI isn’t going to die. It’s too late. The large financial company that I work for has been obsessively tracking hours saved in developer time with GitHub Copilot. I’m an older developer and I was warned this week that my job will be eliminated soon.
So the company that is obsessed with money that you work for has discovered a way to (they think) make more money by getting rid of you and you’re surprised by this?
At least you’ve been forewarned. Take the opportunity to abandon ship. Don’t be the last one standing when the music stops.
I never said that I was surprised. I just wanted to point out that many companies like my own are already making significant changes to how they hire and fire. They need to justify their large investment in AI even though we know the tech isn’t there yet.
See now, I would prefer AI in my toaster. It should be able to learn to adjust the cook time to what I want no matter what type of bread I put in it. Though is that realky AI? It could be. Same with my fridge. Learn what gets used and what doesn’t. Then give my wife the numbers on that damn clear box of salad she buys at costco everytime, which take up a ton of space and always goes bad before she eats even 5% of it. These would be practical benefits to the crap that is day to day life. And far more impactful then search results I can’t trust.
There’s a good point here that like about 80% of what we’re calling AI right now… isn’t even AI or even LLM. It’s just… algorithm, code, plain old math. I’m pretty sure someone is going to refer to a calculator as AI soon. “Wow, it knows math! Just like a person! Amazing technology!”
(That’s putting aside the very question of whether LLMs should even qualify as AIs at all.)
In my professional experience, AI seems to be just a faster way to generate an algorithm that is really hard to debug. Though I am dev-ops/sre so I am not as deep in it as the devs.
I remined of the time researchers used an evolutionary algorithm to devise a circuit that would emit a tone on certain audio inputs and not on others. They examined the resulting circuit and found an extra vestigial bit, but when they cut it off, the chip stopped working. So they re-enabled it. Then they wanted to show off their research at a panel, and at the panel it completely failed. Dismayed they brought it back to their lab to figure out why it stopped working, and it suddenly started working fine.
After a LOT of troubleshooting they eventually discovered that the circuit was generating the tone by using the extra vestigial bit as an antenna that picked up emissions from a CRT in the lab and downconverted it to the desired tone frequency. Turn of the antenna, no signal. Take the chip away from that CRT, no signal.
That’s what I expect LLMs will make. Complex, arcane spaghetti stuff that works but if you look at it funny it won’t work anymore, and nobody knows how it works at all.