5 ms·

I'm a natural skeptic, and I believe we're still on the rising edge of the "AI" hype cycle. Five years ago, it was "blockchain", and everyone was trying to ram

I'm a natural skeptic, and I believe we're still on the rising edge of the "AI" hype cycle. Five years ago, it was "blockchain", and everyone was trying to ram blockchain into everything, attracting lots of VC and media attention, etc. It seems that blockchain is beyond the honeymoon phase: I haven't seen an NFT or even a Bitcoin headline in HN for a while. So I'm trying to wrap my head around what an "AI Engineer" is. As I see it, it's all about calling a function that takes some text as an input, and getting some text as output. That function, of course, is being run on some big hardware that in most cases, you don't own. So is the "engineering" part of this finessing the input and massaging the output? Do most "AI Engineers" actually understand what's going on in that function, beyond what they learned in the "LLM 101" videos and articles that have been flooding the web over the past year? While an application developer doesn't need to know all of the ins-and-outs of the underlying operating systems they run on, those developers that do have a deep understanding of the OS are the ones who write more performant code, and can really get to the bottom of issues that arise do to how their code uses the OS. Can the same be said of "AI Engineers"? All of a sudden, everyone's an AI Engineer. Where where these experts hiding five years ago? https://trends.google.com/trends/explore?date=today%205-y&geo=US&q=blockchain,GPT https://trends.google.com/trends/explore?date=today%205-y&ge...

voz_ 3y ago
Imagine comparing blockchain and AI.
- aerhardt 3y ago
  LLMs and GenAI are already useful at scale but the current hype that they will lead to infinitely generalizable models, myriad groundbreaking applications in all fields and industries, or even AGI could be overblown - let's at least admit that as a possibility.
  - lost_tourist 3y ago
    If AGI reached "human level" capabilities and took off from there how in the world could that possibly be overblown? That would change everything, comparing it to discovering fire would probably be apt, and maybe the first time in recorded history that making that statement isn't overly dramatic.
- sanderjd 3y ago
  I would submit that we don't yet have enough evidence to say whether the comparison is apt or inapt. I personally think what's going on with these "generative" models seems like a bigger deal than blockchain, but it's fiendishly difficult to know what is or isn't hype while embedded within a hype cycle.
  - janalsncm 3y ago
    Seriously? The only comparison you can make is that both were hyped. Digging even a millimeter under the surface reveals they’re completely different. Blockchain was and still is rife with scams and “you just don’t understand the technology bro” hype men. Just check out Dirty Bubble Media. Blockchain was rarely if ever a product that solved a problem, the whole point is The Line Goes Up. That’s why no one uses blockchain in industry, and crypto bros were constantly finding themselves proposing silly use cases like ticket sales and property deeds. These are people who have apparently never heard of a relational database. The hype around AI is due to increased attention on things that have already existed and have already been studied, used, and improved for decades now. There was never much R&D into blockchain tech because the tech isn’t the point. For ML, there are researchers who have worked on these problems for decades. It doesn’t need to justify its own existence, the justification is that it can solve real problems.
    - sanderjd 3y ago
      Again, I do tend to agree that there is a lot more "there" there with generative AI. But I think it's also true that it's too early to be sure. You're comparing the two technologies at totally different points in their hype cycles. The comparison point to where AI is right now is to Bitcoin / very early Ethereum in the late 2000s to early 2010s. Nobody knew where it was all going, some people saw endless potential, other people saw nonsense and scams. The explosion of bitcoin into mainstream consciousness in the early 2010s is akin to the explosion of ChatGPT over the past 6 to 9 months. But what's next? That's what matters. The early 2010s bitcoin boom now pretty clearly looks like a fad, in hindsight. Was ChatGPT also mostly a fad, or is it going to be a lasting fixture of productivity and/or entertainment moving forward? I think it's the latter - it has already changed my habits at work in ways that I think will be permanent - but I just think it's too early to say for sure. (And to be clear, I'm not talking about machine learning as an academic discipline; I totally agree with you that there is definitely enough evidence to say there is a lot more "there" there than research into chained hashing to solve double-spend-like problems.)
- minimaxir 3y ago
  Unfortunately, web3 bros pivoted to AI, which just adds even more noise to the space.
  - wpietri 3y ago
    For me that's absolutely the strongest ground for comparison. In recent decades, we've had two big waves of tech advance: the Web and mobile. A lot of people have lived through them both, giving them an expectation that another such wave should be along soon. You could see that in the decade of blockchain/ICO/DAO/NFT/web3 hype, where people, many with shaky credentials, touted the transformation soon to come, taking in a lot of cash. In retrospect, from Mt Gox to FTX we can see that it was all horseshit. The main real advance was decentralizing not finance or property or computing, but the Ponzi scheme. Despite this failure, and echoing The Great Disappointment [1], we see a lot of the same hype and even the same people. Is it possible that this is different, that there's more substance here? Or is this going to be another Groupon or Metaverse? That does not matter to the hypesters. They're going to run the same routine that worked before. They're going to take in a lot of money, which is the primary goal. It's possible that some of them will, by luck or accident, latch on to something that isn't a total fraud. Surely most of them won't. But we should never forget that is basically irrelevant to a lot of people in the early stages of a cycle. And not just for the fraudsters, but for anybody who makes their living on the upswing of the the hype cycle, including a notable fraction of investors, "experts", and journalists. how ever it turns out, they'll get paid just fine. [1] https://en.wikipedia.org/wiki/Great_Disappointment https://en.wikipedia.org/wiki/Great_Disappointment
    - minimaxir 3y ago
      Web3 / Metaverse was a solution in search of a problem. AI is solving problems today.
      - wpietri 3y ago
        You're responding as if I said they were the same, but I was pretty careful to say the opposite. My point is explicitly setting the utility of "AI" aside.
- _ea1k 3y ago
  It'll be fun to look back on these comments in a few years. It will be like looking back on the internet skeptics of the 90s. Most people have forgotten about those. Of course, there was a big boom and bust cycle back then too, but just like then, this cycle is nowhere near its peak.
  - ResearchCode 3y ago
    That's what they said about the coins and the tokens too.
phillipcarter 3y ago
> Where where these experts hiding five years ago? Prompt engineering has been around for a few years already, and there are a lot of existing ML engineers who have been able to quickly learn how to adapt their skills.
dr_dshiv 3y ago
ChatGPT is useful in a way that block chain never was. There may be inflated expectations, but we are already on “the plateau of productivity.” I don’t think LLMs are overhyped relative to the normal background levels of tech hype.
spmurrayzzz 3y ago
> So is the "engineering" part of this finessing the input and massaging the output? I don't know if I'll ever use the phrase "AI Engineer" myself, but there's plenty of meaningful engineering work in that space that strays pretty far from just calling some provider's APIs. A few that come to mind just for LLMs: - Custom fine-tuning of foundational models both in the classic sense and with more modern strategies like PEFT/QLoRA - Data preprocessing pipelines to help automate fine-tuning, vectorization, etc - Continuous integration suites to evaluate models on standard benchmarks as they change over time - Vector db / semantic search engineering to help decorate context windows effectively - Architecting ensemble models infrastructure to accommodate more complex task processing I think many of those probably going into what folks are calling the "MLOps" bucket, but I think its a more broadly a combination of research, application engineering, and operations engineering. Edit: for clarity, my position is that the line in the article between AI Engineer and ML Engineer need not be that bright. Just like software engineers today that write/operate their own devops tooling to deploy and manage the apps they build.
- majesticglue 3y ago
  of all the things mentioned in this whole thread, this resonates the most. from my take, everyone here mentions AI engineering as someone just plugging in LLM (ie OpenAI) into every software application. However I believe there is a nice burgeoning of the "AI Engineer" where it involves more of the data processing and specialized AI fine-tuning that engineers have a role in, but of course it requires more specialization than just learning LangChain and calling it a day.
minimaxir 3y ago
> Where where these experts hiding five years ago? You could use GPT-3 in 2020 but it was expensive and difficult to make it behave. Iterations of GPT-3 starting in 2021-2022 allowed it to obey commands (InstructGPT) and made it more feasible to "engineer" with it. The true inflection point was due to free, accessible and good-enough-quality AI generation in the form of Midjourney and ChatGPT.
PheonixPharts 3y ago
AI Engineer here. To start with, I share some of your skepticism about AI and hype (but I love these problems, so I'm happy to take the risk of overhyping to try to solve these challenges). But there is a lot of real work going on in this space. Though most of these are basically the same answers as you'd get in the article. > it's all about calling a function that takes some text as an input, and getting some text as output. Real world AI applications involve non-trivial prompts that are often composed of many different components and dynamically changed based on user interaction with the environment. So it's not quite as simple in practice as just calling an API. > So is the "engineering" part of this finessing the input and massaging the output? You could make this claim about all software engineering at the end of the day. If you want to understand whether or not any of the billion companies shipping "AI" products right now are really doing AI, the big term to ask about is "evaluations". It is not trivial to evaluate the performance of LLM output across a broad range of tasks. However if you're not doing this, then you can't possibly know how your efforts are doing. The companies that are slapping "AI" stickers on old products are largely ignoring this issue. The next challenge is "how do you improve bad outputs?" Prompt engineering is one solution, but there are potentially may other engineering solutions to recovering from a bad state. None of these are trivial. A rapidly growing part of this space is working with "agents", that is you have multiple LLMs that are capable of interacting with each other. This area is changing rapidly. Vector databases are also becoming very important of the work as not all LLM/AI work is just throwing around prompts, but often working with embeddings. > All of a sudden, everyone's an AI Engineer. Where where these experts hiding five years ago? It's not that mysterious. Everyone I know working in this space right now was either a very engineering focused data scientist in their last role, or an ML engineer working near this space. In either case they're people that have been interested in this space before that have all the skills necessary to change roles. > Can the same be said of "AI Engineers"? At least in my circle, everyone doing this work right now has a long history of working in machine learning and quantitative problem solving. Of course that used to be true of ML engineers as well (and I've not far to many MLEs that don't understand gradient descent).
chefandy 3y ago
> Where where these experts hiding five years ago? I'll bet that many of them were trying to ram blockchain into everything! Personally, I'm curiously watching emerging prompt pen testing scene. It brings a tear of cyberpunk joy to my eye to consider that we've built important, powerful systems advanced enough to be vulnerable to social engineering attacks... and pretty dumb ones, too.
- ryandrake 3y ago
  What we're seeing now is a frantic attempt by companies to ram "AI" into everything. Some force (Wall Street?) is expecting companies to say "We're using AI." What do you need it for? "We don't know, but damn we've got to use it for something!" A few years ago, few were talking about AI. Today, if it's not somehow crammed into your product roadmap, you might as well start looking for another job.
- gaogao 3y ago
  On the cyberpunk side, the LLM demoscene is also incredibly fascinating. We folks beating out the megacorps on model hacks like NKT and hecking crazy model names like TheBloke/WizardLM-Uncensored-SuperCOT-StoryTelling-30B-SuperHOT-8K-GGML
pcthrowaway 3y ago
It really wasn't blockchain 5 years ago. It was blockchain 6 years ago (though to a much lesser degree, and focused on different iterations of the distributed ledger idea). Then it was really blockchain 2 years ago (but focused on protocols built on smart-contract-enabled blockchains). Interestingly, those two targets are completely different. 6 years ago the application was the ledger, and the language was usually C, 2 years ago the application was "a financial product built to run on a resource-constrained virtual machine that executes a custom language (or Rust)". The demand follows the money, so if the whole space experiences another "bull run" there will be lots of demand for "blockchain" again, and I'm sure the actual associated skills expected will be different as well
buffalobuffalo 3y ago
I'm also a skeptic, but for a slightly different reason. There are currently two types of business use cases that seem to be the focal point of this generation of AI. 1) Tooling. This one I think will probably bear fruit. It will likely result in huge productivity gains (I mean, it already has for me). But i don't know if it will result in a paradigm shift. 2) Agents. This is where most of the hype is focused. The idea is that you can cut humans out of the loop. If doable, this would be revolutionary. But from what I've seen, the currently technology is not likely to do this. The reason is that agents that are required to perform a long series of independent actions without human intervention can fall victim to an accumulation of small errors that occur in each step. This kind of "snowball effect" will often result in complete garbage being produced. As a result of point two, I think a lot of the new AI products are likely to fall flat. And as this happens, the hype/funding will dry up pretty quickly. All the same, I think this will have more positive economic outcome than all the crypto stuff did.
- __loam 3y ago
  I disagree completely with your last point. Flooding the internet with automated bullshit and people locking down their data because they're worried they'll get scraped is going to be a hugely negative consequence, to say nothing of what this will do to people like artists and writers because of the perception of this technology that the people paying them are going to have.
  - buffalobuffalo 3y ago
    I actually agree with you. I think the societal impacts of this are going to be huge. I was speaking more to the monetization potential of AI. And I guess I should have specified that this refers legal monetization, since I'm guessing that for a time, illegal bot operators will do quite well. In terms of societal impact, I suspect that this will ultimately result in the death of all open platforms on the internet and a withdrawal from online spaces in general.
Xenoamorphous 3y ago
I think comparing AI to blockchain is not fair. Sure, it was a hyped technology a few years back, but only in tech environments. My mom has never heard of blockchain, I bet. AI is in the mainstream media all the time. But ultimately it's a matter of scope. AI has the potential of, at the very least, transforming lots of jobs. Blockchain never had that potential. The company I work for, a non-tech one, never ever mentioned blockchain. But they're trying to get AI everywhere.
badrequest 3y ago
> Do most "AI Engineers" actually understand what's going on in that function, beyond what they learned in the "LLM 101" videos and articles that have been flooding the web over the past year? Do they need to, to be effective at their jobs?
visarga 3y ago
When your task is too nuanced to be described in a prompt and a few demonstrations you need to use one of the fine-tuning scripts to bake into the model much more supervised data. That's still doable for AI Engineers, prepare data, call LoRA fine tuning script, deploy model. The problem is having a good supervised training set. But recently it can be generated with LLMs + plugins and language chains, basically amplified LLMs. In other cases you need to collect human preference data and apply a different kind of fine-tuning. Still mostly dataset curation and iterative model building, something and AI Engineer should be able to do. Working with LLMs is very different from building neural nets like in 2017. Tons of those old skills are not needed anymore, so many of the old tasks are basically solved at human level. And a whole new set of problems appeared.
- __loam 3y ago
  I've been encountering machine learning on and off for the last 8 years or so. The idea that people are generating data with LLMs to then turn around and train LLMs without a hint of irony is hilarious and astonishing to me.
fuddle 3y ago
Bitcoin is probably a better comparison - https://trends.google.com/trends/explore?date=today%205-y&geo=US&q=bitcoin,GPT https://trends.google.com/trends/explore?date=today%205-y&ge...
lispisok 3y ago
A lot of people trying to refute you claiming the latest AI hype has actual value. The new models people are raving about do have value way above what blockchain had but I'd argue the AI hype is also way higher. Overall I'd say the gap between AI hype and AI value is bigger than blockchain despite blockchain value basically being zero if you dont count transactions for illicit goods.
- __loam 3y ago
  ChatGPT was quite literally optimized to be convincing to humans thanks to HFRL. I think it's going to be a huge negative externality thanks to the amount of ai noise it's going to be pumping onto the internet for the next couple of years. It's going to get harder and harder to connect with real humans and get information that we know is true rather than just proliferated by an automated system that knows nothing about quality. That negative impact might be even worse than blockchain if we see fewer and fewer people participating in creative skills, or even being able to make a living thanks to the lies the AI industry has sold to MBAs.

voz_ 3y ago

Imagine comparing blockchain and AI.

aerhardt 3y ago
LLMs and GenAI are already useful at scale but the current hype that they will lead to infinitely generalizable models, myriad groundbreaking applications in all fields and industries, or even AGI could be overblown - let's at least admit that as a possibility.
- lost_tourist 3y ago
  If AGI reached "human level" capabilities and took off from there how in the world could that possibly be overblown? That would change everything, comparing it to discovering fire would probably be apt, and maybe the first time in recorded history that making that statement isn't overly dramatic.
sanderjd 3y ago
I would submit that we don't yet have enough evidence to say whether the comparison is apt or inapt. I personally think what's going on with these "generative" models seems like a bigger deal than blockchain, but it's fiendishly difficult to know what is or isn't hype while embedded within a hype cycle.
- janalsncm 3y ago
  Seriously? The only comparison you can make is that both were hyped. Digging even a millimeter under the surface reveals they’re completely different. Blockchain was and still is rife with scams and “you just don’t understand the technology bro” hype men. Just check out Dirty Bubble Media. Blockchain was rarely if ever a product that solved a problem, the whole point is The Line Goes Up. That’s why no one uses blockchain in industry, and crypto bros were constantly finding themselves proposing silly use cases like ticket sales and property deeds. These are people who have apparently never heard of a relational database. The hype around AI is due to increased attention on things that have already existed and have already been studied, used, and improved for decades now. There was never much R&D into blockchain tech because the tech isn’t the point. For ML, there are researchers who have worked on these problems for decades. It doesn’t need to justify its own existence, the justification is that it can solve real problems.
  - sanderjd 3y ago
    Again, I do tend to agree that there is a lot more "there" there with generative AI. But I think it's also true that it's too early to be sure. You're comparing the two technologies at totally different points in their hype cycles. The comparison point to where AI is right now is to Bitcoin / very early Ethereum in the late 2000s to early 2010s. Nobody knew where it was all going, some people saw endless potential, other people saw nonsense and scams. The explosion of bitcoin into mainstream consciousness in the early 2010s is akin to the explosion of ChatGPT over the past 6 to 9 months. But what's next? That's what matters. The early 2010s bitcoin boom now pretty clearly looks like a fad, in hindsight. Was ChatGPT also mostly a fad, or is it going to be a lasting fixture of productivity and/or entertainment moving forward? I think it's the latter - it has already changed my habits at work in ways that I think will be permanent - but I just think it's too early to say for sure. (And to be clear, I'm not talking about machine learning as an academic discipline; I totally agree with you that there is definitely enough evidence to say there is a lot more "there" there than research into chained hashing to solve double-spend-like problems.)
minimaxir 3y ago
Unfortunately, web3 bros pivoted to AI, which just adds even more noise to the space.
- wpietri 3y ago
  For me that's absolutely the strongest ground for comparison. In recent decades, we've had two big waves of tech advance: the Web and mobile. A lot of people have lived through them both, giving them an expectation that another such wave should be along soon. You could see that in the decade of blockchain/ICO/DAO/NFT/web3 hype, where people, many with shaky credentials, touted the transformation soon to come, taking in a lot of cash. In retrospect, from Mt Gox to FTX we can see that it was all horseshit. The main real advance was decentralizing not finance or property or computing, but the Ponzi scheme. Despite this failure, and echoing The Great Disappointment [1], we see a lot of the same hype and even the same people. Is it possible that this is different, that there's more substance here? Or is this going to be another Groupon or Metaverse? That does not matter to the hypesters. They're going to run the same routine that worked before. They're going to take in a lot of money, which is the primary goal. It's possible that some of them will, by luck or accident, latch on to something that isn't a total fraud. Surely most of them won't. But we should never forget that is basically irrelevant to a lot of people in the early stages of a cycle. And not just for the fraudsters, but for anybody who makes their living on the upswing of the the hype cycle, including a notable fraction of investors, "experts", and journalists. how ever it turns out, they'll get paid just fine. [1] https://en.wikipedia.org/wiki/Great_Disappointment https://en.wikipedia.org/wiki/Great_Disappointment
  - minimaxir 3y ago
    Web3 / Metaverse was a solution in search of a problem. AI is solving problems today.
    - wpietri 3y ago
      You're responding as if I said they were the same, but I was pretty careful to say the opposite. My point is explicitly setting the utility of "AI" aside.
_ea1k 3y ago
It'll be fun to look back on these comments in a few years. It will be like looking back on the internet skeptics of the 90s. Most people have forgotten about those. Of course, there was a big boom and bust cycle back then too, but just like then, this cycle is nowhere near its peak.
- ResearchCode 3y ago
  That's what they said about the coins and the tokens too.

phillipcarter 3y ago

> Where where these experts hiding five years ago? Prompt engineering has been around for a few years already, and there are a lot of existing ML engineers who have been able to quickly learn how to adapt their skills.

dr_dshiv 3y ago

ChatGPT is useful in a way that block chain never was. There may be inflated expectations, but we are already on “the plateau of productivity.” I don’t think LLMs are overhyped relative to the normal background levels of tech hype.

spmurrayzzz 3y ago

> So is the "engineering" part of this finessing the input and massaging the output? I don't know if I'll ever use the phrase "AI Engineer" myself, but there's plenty of meaningful engineering work in that space that strays pretty far from just calling some provider's APIs. A few that come to mind just for LLMs: - Custom fine-tuning of foundational models both in the classic sense and with more modern strategies like PEFT/QLoRA - Data preprocessing pipelines to help automate fine-tuning, vectorization, etc - Continuous integration suites to evaluate models on standard benchmarks as they change over time - Vector db / semantic search engineering to help decorate context windows effectively - Architecting ensemble models infrastructure to accommodate more complex task processing I think many of those probably going into what folks are calling the "MLOps" bucket, but I think its a more broadly a combination of research, application engineering, and operations engineering. Edit: for clarity, my position is that the line in the article between AI Engineer and ML Engineer need not be that bright. Just like software engineers today that write/operate their own devops tooling to deploy and manage the apps they build.

majesticglue 3y ago
of all the things mentioned in this whole thread, this resonates the most. from my take, everyone here mentions AI engineering as someone just plugging in LLM (ie OpenAI) into every software application. However I believe there is a nice burgeoning of the "AI Engineer" where it involves more of the data processing and specialized AI fine-tuning that engineers have a role in, but of course it requires more specialization than just learning LangChain and calling it a day.

minimaxir 3y ago

> Where where these experts hiding five years ago? You could use GPT-3 in 2020 but it was expensive and difficult to make it behave. Iterations of GPT-3 starting in 2021-2022 allowed it to obey commands (InstructGPT) and made it more feasible to "engineer" with it. The true inflection point was due to free, accessible and good-enough-quality AI generation in the form of Midjourney and ChatGPT.

PheonixPharts 3y ago

AI Engineer here. To start with, I share some of your skepticism about AI and hype (but I love these problems, so I'm happy to take the risk of overhyping to try to solve these challenges). But there is a lot of real work going on in this space. Though most of these are basically the same answers as you'd get in the article. > it's all about calling a function that takes some text as an input, and getting some text as output. Real world AI applications involve non-trivial prompts that are often composed of many different components and dynamically changed based on user interaction with the environment. So it's not quite as simple in practice as just calling an API. > So is the "engineering" part of this finessing the input and massaging the output? You could make this claim about all software engineering at the end of the day. If you want to understand whether or not any of the billion companies shipping "AI" products right now are really doing AI, the big term to ask about is "evaluations". It is not trivial to evaluate the performance of LLM output across a broad range of tasks. However if you're not doing this, then you can't possibly know how your efforts are doing. The companies that are slapping "AI" stickers on old products are largely ignoring this issue. The next challenge is "how do you improve bad outputs?" Prompt engineering is one solution, but there are potentially may other engineering solutions to recovering from a bad state. None of these are trivial. A rapidly growing part of this space is working with "agents", that is you have multiple LLMs that are capable of interacting with each other. This area is changing rapidly. Vector databases are also becoming very important of the work as not all LLM/AI work is just throwing around prompts, but often working with embeddings. > All of a sudden, everyone's an AI Engineer. Where where these experts hiding five years ago? It's not that mysterious. Everyone I know working in this space right now was either a very engineering focused data scientist in their last role, or an ML engineer working near this space. In either case they're people that have been interested in this space before that have all the skills necessary to change roles. > Can the same be said of "AI Engineers"? At least in my circle, everyone doing this work right now has a long history of working in machine learning and quantitative problem solving. Of course that used to be true of ML engineers as well (and I've not far to many MLEs that don't understand gradient descent).

chefandy 3y ago

> Where where these experts hiding five years ago? I'll bet that many of them were trying to ram blockchain into everything! Personally, I'm curiously watching emerging prompt pen testing scene. It brings a tear of cyberpunk joy to my eye to consider that we've built important, powerful systems advanced enough to be vulnerable to social engineering attacks... and pretty dumb ones, too.

ryandrake 3y ago
What we're seeing now is a frantic attempt by companies to ram "AI" into everything. Some force (Wall Street?) is expecting companies to say "We're using AI." What do you need it for? "We don't know, but damn we've got to use it for something!" A few years ago, few were talking about AI. Today, if it's not somehow crammed into your product roadmap, you might as well start looking for another job.
gaogao 3y ago
On the cyberpunk side, the LLM demoscene is also incredibly fascinating. We folks beating out the megacorps on model hacks like NKT and hecking crazy model names like TheBloke/WizardLM-Uncensored-SuperCOT-StoryTelling-30B-SuperHOT-8K-GGML

pcthrowaway 3y ago

It really wasn't blockchain 5 years ago. It was blockchain 6 years ago (though to a much lesser degree, and focused on different iterations of the distributed ledger idea). Then it was really blockchain 2 years ago (but focused on protocols built on smart-contract-enabled blockchains). Interestingly, those two targets are completely different. 6 years ago the application was the ledger, and the language was usually C, 2 years ago the application was "a financial product built to run on a resource-constrained virtual machine that executes a custom language (or Rust)". The demand follows the money, so if the whole space experiences another "bull run" there will be lots of demand for "blockchain" again, and I'm sure the actual associated skills expected will be different as well

buffalobuffalo 3y ago

I'm also a skeptic, but for a slightly different reason. There are currently two types of business use cases that seem to be the focal point of this generation of AI. 1) Tooling. This one I think will probably bear fruit. It will likely result in huge productivity gains (I mean, it already has for me). But i don't know if it will result in a paradigm shift. 2) Agents. This is where most of the hype is focused. The idea is that you can cut humans out of the loop. If doable, this would be revolutionary. But from what I've seen, the currently technology is not likely to do this. The reason is that agents that are required to perform a long series of independent actions without human intervention can fall victim to an accumulation of small errors that occur in each step. This kind of "snowball effect" will often result in complete garbage being produced. As a result of point two, I think a lot of the new AI products are likely to fall flat. And as this happens, the hype/funding will dry up pretty quickly. All the same, I think this will have more positive economic outcome than all the crypto stuff did.

__loam 3y ago
I disagree completely with your last point. Flooding the internet with automated bullshit and people locking down their data because they're worried they'll get scraped is going to be a hugely negative consequence, to say nothing of what this will do to people like artists and writers because of the perception of this technology that the people paying them are going to have.
- buffalobuffalo 3y ago
  I actually agree with you. I think the societal impacts of this are going to be huge. I was speaking more to the monetization potential of AI. And I guess I should have specified that this refers legal monetization, since I'm guessing that for a time, illegal bot operators will do quite well. In terms of societal impact, I suspect that this will ultimately result in the death of all open platforms on the internet and a withdrawal from online spaces in general.

Xenoamorphous 3y ago

I think comparing AI to blockchain is not fair. Sure, it was a hyped technology a few years back, but only in tech environments. My mom has never heard of blockchain, I bet. AI is in the mainstream media all the time. But ultimately it's a matter of scope. AI has the potential of, at the very least, transforming lots of jobs. Blockchain never had that potential. The company I work for, a non-tech one, never ever mentioned blockchain. But they're trying to get AI everywhere.

badrequest 3y ago

> Do most "AI Engineers" actually understand what's going on in that function, beyond what they learned in the "LLM 101" videos and articles that have been flooding the web over the past year? Do they need to, to be effective at their jobs?

visarga 3y ago

When your task is too nuanced to be described in a prompt and a few demonstrations you need to use one of the fine-tuning scripts to bake into the model much more supervised data. That's still doable for AI Engineers, prepare data, call LoRA fine tuning script, deploy model. The problem is having a good supervised training set. But recently it can be generated with LLMs + plugins and language chains, basically amplified LLMs. In other cases you need to collect human preference data and apply a different kind of fine-tuning. Still mostly dataset curation and iterative model building, something and AI Engineer should be able to do. Working with LLMs is very different from building neural nets like in 2017. Tons of those old skills are not needed anymore, so many of the old tasks are basically solved at human level. And a whole new set of problems appeared.

__loam 3y ago
I've been encountering machine learning on and off for the last 8 years or so. The idea that people are generating data with LLMs to then turn around and train LLMs without a hint of irony is hilarious and astonishing to me.

fuddle 3y ago

Bitcoin is probably a better comparison - https://trends.google.com/trends/explore?date=today%205-y&geo=US&q=bitcoin,GPT https://trends.google.com/trends/explore?date=today%205-y&ge...

lispisok 3y ago

A lot of people trying to refute you claiming the latest AI hype has actual value. The new models people are raving about do have value way above what blockchain had but I'd argue the AI hype is also way higher. Overall I'd say the gap between AI hype and AI value is bigger than blockchain despite blockchain value basically being zero if you dont count transactions for illicit goods.

__loam 3y ago
ChatGPT was quite literally optimized to be convincing to humans thanks to HFRL. I think it's going to be a huge negative externality thanks to the amount of ai noise it's going to be pumping onto the internet for the next couple of years. It's going to get harder and harder to connect with real humans and get information that we know is true rather than just proliferated by an automated system that knows nothing about quality. That negative impact might be even worse than blockchain if we see fewer and fewer people participating in creative skills, or even being able to make a living thanks to the lies the AI industry has sold to MBAs.