4 ms·

‘AI-powered’ is tech’s meaningless equivalent of ‘all natural’ (2017)

37 points by nvr219 4y ago

dexwiz 4y ago
This article makes sense in 2017, which most AI was just a statistical model. But 6 years later we are living in a much different world.
- nwah1 4y ago
  Not that much different. 9/10 uses of the term are still complete bullshit. The other 1/10 are still mostly bullshit as well.
  - BoorishBears 4y ago
    If you don't understand the potential of a technology it's certainly easier to dismiss it as bullshit than to learn. Crypto/"Web3" was stupid because of the oracle problem: Your magical NFT can't actually enforce its license without leveraging the slightly centralized concept of the federal court system. AI as in LLMs and Latent Diffusion do not have that kind of obvious limitation to generating any actual value. There will clearly be specific limitations, but even within the confines of what has been built so far there is an insane amount of value and potential. Crypto detractors who are lumping AI into the "bullshit" bucket really come across as a broken clock that was right once, and I say that as someone who has been a crypto detractor for a long time.
    - nwah1 4y ago
      I didn't say machine learning is useless. I said most products claiming to have AI secret sauce have no secret sauce. Big difference.
      - BoorishBears 4y ago
        Saying: > I said most products claiming to have AI secret sauce have no secret sauce. Means you're implying > Most products claiming to AI have AI Which is exactly the opposite of the article and agreeing with the comment to you replied to while calling things bullshit... When the article was written they straight up were not using AI, now they are and mostly in novel ways by the simple virtue they're building on a novel tech.
    - marginalia_nu 4y ago
      Eh, I think a lot of the reasons to be skeptical of crypto still apply to "AI". AI may have actual real world uses (unlike crypto), but it's also the subject of a huge hype campaign (like crypto) largely consisting of the same people hyping crypto, with similar strategies of emphasizing FOMO (like crypto) and making wild and fantastical claims about unheard of growth and about the end of the existing world order through technology in a way that borders on eachatology (like crypto).
      - idopmstuff 4y ago
        Except that you can't make money on hype with AI the same way that you could with crypto. If you could generate FOMO about a coin, people would buy that coin and you'd make money. Generating FOMO about AI just leads people to go try ChatGPT. Unless you've built some domain-specific product on top that's meaningfully better than ChatGPT or Bing at whatever it's supposed to do, you can't actually profit (and if you have built that, then good for you, you've done something useful). That is all to say that there is definitely a lot of crypto hype, but it seems to be generated much more by legitimate excitement than by financial incentives. And given how much progress we've seen with AI serving legitimately useful functions, the hype seems much more warranted than the crypto hype (some of it, at least).
        add-sub-mul-div 4y ago
        > Except that you can't make money on hype with AI the same way that you could with crypto. People seem to think so. Have you seen all the shovelware getting posted here lately?
        marginalia_nu 4y ago
        I think these people have been useful idiots for crypto, and crypto grifters have fanned their flames and encouraged this cult of technology, but at this point they're an independent entity that will gladly latch onto any narrative that resonates them in the vacuum left behind crypto. They were never making money off this. If anything, they were the ones going broke in bait-and-switches.
      - 4y ago
    - esjeon 4y ago
      "Potential" is just a belief, and wielding it proves nor does nothing.
- version_five 4y ago
  It's been funny watching the "it's just a statistical model" crowd who were always quick to chime in (when that's not what 2016 deep learning was) now doing a 180 degree turn to "how do we know the brain isn't just a llm" or whatever silly confused by a mirror opinion they have now, when despite obvious advances in scale, theres no fundamental difference in the models. The only takeaway for me is that people love to passionately misunderstand anything that gets called AI.
  - cmarschner 4y ago
    For my part (phd in computational linguistics, first classical „AI“, then machine learning, then deep learning since 2014, not working on NLP for the past 5 years) - I thought I would know a thing or two about what‘s possible, and I am still in absolute awe what these models spit out. People might not be aware that just a few years ago formulating a grammatically correct sentence was a hard problem, let alone all the other things that once had names (anaphora resolution, semantics, pragmatics etc. etc.) and conference tracks and workshops dedicated to them. All that swept away by data and scalable transformer models. It is such a f*cking miracle. I would be curious what the state of understanding these models is. Haven‘t done any literature review on this but just reverse engineering whatever structure the model has come about could be enlightning (perhaps).
- nvr219 4y ago
  It's still that just with faster computers.
- pornel 4y ago
  We may seamlessly transition from "this isn't a real A.I, only a bit of simple fuzzy logic" to "this is needlessly using a billion-weights model to simulate a bit of simple fuzzy logic".
- esjeon 4y ago
  Welp, it's still a statistical model, and nothing is wrong with it. The only thing that's changed is that we now have a large-scale statistical method for extracting patterns out of a large corpus of highly complex data in short-enough time span. This progress was mainly backed by engineering efforts like cloud infrastructure, big data methodologies, GPGPU. The theory itself only got a small bump since the last AI winter (in terms of depth), and, combined with the engineering advancement, it opened up a whole lot of exploration opportunities (in terms of breadth).
  - scrubs 4y ago
    You stole my thunder.
  - famouswaffles 4y ago
    it's not statistical model. Unless the implication of the word has lost all meaning. The post shifting rampant in the discussions of this field is something to behold.
360macky 4y ago
Interesting. I see it. This will be the counterpart of "Human-designed" or "Made by humans".
joseph8th 4y ago
It's a marketing term rendered meaningless by the marketers. Even those who know better keep calling LLMs "AI"
- flangola7 4y ago
  What are LLMs if not AI? A machine that can look at a comic image and explain why it is funny, then play and win a chess game, then explain in Arabic how to solve a trigonometry problem, and then finally write a persuasive speech arguing for a comedy movie about the above three skills to be greenlit, is an intelligent entity.
- esjeon 4y ago
  Yeah, "AI" has been a marketing term for almost a decade. People who know better these days tend to avoid the term "AI" in professional contexts and use "model" instead. It's like "autopilot" from Tesla in the sense that the name itself makes people cast unfounded over-expectation (= overhype). You can see tons of people consider GPT as an AGI.
ralusek 4y ago
We run all of our bits through "Hot Dog/Not Hot Dog" AI. It turned out to be an identity function, because it thinks the 1s are hot dogs and the 0s aren't. All of our integration tests are still passing...things are just a lot slower around here, and multiple people swear that the servers now smell like a baseball game.

37 points by nvr219 4y ago

This article makes sense in 2017, which most AI was just a statistical model. But 6 years later we are living in a much different world.

nwah1 4y ago
Not that much different. 9/10 uses of the term are still complete bullshit. The other 1/10 are still mostly bullshit as well.
- BoorishBears 4y ago
  If you don't understand the potential of a technology it's certainly easier to dismiss it as bullshit than to learn. Crypto/"Web3" was stupid because of the oracle problem: Your magical NFT can't actually enforce its license without leveraging the slightly centralized concept of the federal court system. AI as in LLMs and Latent Diffusion do not have that kind of obvious limitation to generating any actual value. There will clearly be specific limitations, but even within the confines of what has been built so far there is an insane amount of value and potential. Crypto detractors who are lumping AI into the "bullshit" bucket really come across as a broken clock that was right once, and I say that as someone who has been a crypto detractor for a long time.
  - nwah1 4y ago
    I didn't say machine learning is useless. I said most products claiming to have AI secret sauce have no secret sauce. Big difference.
    - BoorishBears 4y ago
      Saying: > I said most products claiming to have AI secret sauce have no secret sauce. Means you're implying > Most products claiming to AI have AI Which is exactly the opposite of the article and agreeing with the comment to you replied to while calling things bullshit... When the article was written they straight up were not using AI, now they are and mostly in novel ways by the simple virtue they're building on a novel tech.
  - marginalia_nu 4y ago
    Eh, I think a lot of the reasons to be skeptical of crypto still apply to "AI". AI may have actual real world uses (unlike crypto), but it's also the subject of a huge hype campaign (like crypto) largely consisting of the same people hyping crypto, with similar strategies of emphasizing FOMO (like crypto) and making wild and fantastical claims about unheard of growth and about the end of the existing world order through technology in a way that borders on eachatology (like crypto).
    - idopmstuff 4y ago
      Except that you can't make money on hype with AI the same way that you could with crypto. If you could generate FOMO about a coin, people would buy that coin and you'd make money. Generating FOMO about AI just leads people to go try ChatGPT. Unless you've built some domain-specific product on top that's meaningfully better than ChatGPT or Bing at whatever it's supposed to do, you can't actually profit (and if you have built that, then good for you, you've done something useful). That is all to say that there is definitely a lot of crypto hype, but it seems to be generated much more by legitimate excitement than by financial incentives. And given how much progress we've seen with AI serving legitimately useful functions, the hype seems much more warranted than the crypto hype (some of it, at least).
      - add-sub-mul-div 4y ago
        > Except that you can't make money on hype with AI the same way that you could with crypto. People seem to think so. Have you seen all the shovelware getting posted here lately?
      - marginalia_nu 4y ago
        I think these people have been useful idiots for crypto, and crypto grifters have fanned their flames and encouraged this cult of technology, but at this point they're an independent entity that will gladly latch onto any narrative that resonates them in the vacuum left behind crypto. They were never making money off this. If anything, they were the ones going broke in bait-and-switches.
    - 4y ago
  - esjeon 4y ago
    "Potential" is just a belief, and wielding it proves nor does nothing.
version_five 4y ago
It's been funny watching the "it's just a statistical model" crowd who were always quick to chime in (when that's not what 2016 deep learning was) now doing a 180 degree turn to "how do we know the brain isn't just a llm" or whatever silly confused by a mirror opinion they have now, when despite obvious advances in scale, theres no fundamental difference in the models. The only takeaway for me is that people love to passionately misunderstand anything that gets called AI.
- cmarschner 4y ago
  For my part (phd in computational linguistics, first classical „AI“, then machine learning, then deep learning since 2014, not working on NLP for the past 5 years) - I thought I would know a thing or two about what‘s possible, and I am still in absolute awe what these models spit out. People might not be aware that just a few years ago formulating a grammatically correct sentence was a hard problem, let alone all the other things that once had names (anaphora resolution, semantics, pragmatics etc. etc.) and conference tracks and workshops dedicated to them. All that swept away by data and scalable transformer models. It is such a f*cking miracle. I would be curious what the state of understanding these models is. Haven‘t done any literature review on this but just reverse engineering whatever structure the model has come about could be enlightning (perhaps).
nvr219 4y ago
It's still that just with faster computers.
pornel 4y ago
We may seamlessly transition from "this isn't a real A.I, only a bit of simple fuzzy logic" to "this is needlessly using a billion-weights model to simulate a bit of simple fuzzy logic".
esjeon 4y ago
Welp, it's still a statistical model, and nothing is wrong with it. The only thing that's changed is that we now have a large-scale statistical method for extracting patterns out of a large corpus of highly complex data in short-enough time span. This progress was mainly backed by engineering efforts like cloud infrastructure, big data methodologies, GPGPU. The theory itself only got a small bump since the last AI winter (in terms of depth), and, combined with the engineering advancement, it opened up a whole lot of exploration opportunities (in terms of breadth).
- scrubs 4y ago
  You stole my thunder.
- famouswaffles 4y ago
  it's not statistical model. Unless the implication of the word has lost all meaning. The post shifting rampant in the discussions of this field is something to behold.

360macky 4y ago

Interesting. I see it. This will be the counterpart of "Human-designed" or "Made by humans".

joseph8th 4y ago

It's a marketing term rendered meaningless by the marketers. Even those who know better keep calling LLMs "AI"

flangola7 4y ago
What are LLMs if not AI? A machine that can look at a comic image and explain why it is funny, then play and win a chess game, then explain in Arabic how to solve a trigonometry problem, and then finally write a persuasive speech arguing for a comedy movie about the above three skills to be greenlit, is an intelligent entity.
esjeon 4y ago
Yeah, "AI" has been a marketing term for almost a decade. People who know better these days tend to avoid the term "AI" in professional contexts and use "model" instead. It's like "autopilot" from Tesla in the sense that the name itself makes people cast unfounded over-expectation (= overhype). You can see tons of people consider GPT as an AGI.

ralusek 4y ago

We run all of our bits through "Hot Dog/Not Hot Dog" AI. It turned out to be an identity function, because it thinks the 1s are hot dogs and the 0s aren't. All of our integration tests are still passing...things are just a lot slower around here, and multiple people swear that the servers now smell like a baseball game.