3 ms·

What is AI? I've only heard of LLM.

by rambambram 3y ago

catchnear4321 3y ago
it is just the commercial name for llm and related technologies.
- dartos 3y ago
  That’s just not true. The technology that’s behind LLMs have been in use in many disciplines for decades. (Under the names “machine learning,” “statistics,” and now AI) LLMs are just the ones that are popular right now because even non technical people can interact with them. In the mid 2010s it was image classification, for example.
  - skellington 3y ago
    Sort of. The deep-learning AI revolution started with back propagation techniques (80s) that made neural nets trainable. But since then it's taken a long time to get to the relatively recent architectural breakthroughs like transformers (2017) and modern-diffusion (2020) which lead to chat-gpt and dall-e. All of the big AI breakthroughs now are based on neural nets vs classical machine learning which grouped a bunch of statistical, regression, and optimization type models together. Modern AI is the synthesis of new techniques and old ideas plus huge datasets and huge computing power.
  - catchnear4321 3y ago
    > LLMs are just the ones that are popular right now because even non technical people can interact with them. this was essentially my point, that a /s might have made more clear, and made cost less internet points. ai and how the term is used commercially are very different things. the latter tending to be a myopic view of the former that is constrained by dollar-seeking. ai is getting to be old as shit. ai hype is in its latest wave of commercialization. llms being the current so hot right now.
  - jdefr89 3y ago
    It’s true that AI has seen many advancements before a “harsh winter” comes along. You’re correct this is sensationalized, and also, some of these technologies have indeed been in use for decades. But, the Transformer model used by the new popular LLMs is quite recent (2017 is when the paper originally released I believe).
jfengel 3y ago
"Deep learning" and neural networks in general were a thing long before they were applied specifically to language models. It's still not "intelligence" but it does use techniques inspired by brains and have some of the same advantages in "seeing" patterns.
Dalewyn 3y ago
"AI" is what we used to call algorithms, macros, and applications not too long ago. Marketing.
- gumballindie 3y ago
  Amazon goes as far as to call my heating fan’s temperature control ai. Probably why it doesn’t work and it randomly turns on and off.
graypegg 3y ago
I don't think you should be seemingly in the negatives for this comment. :( It's a totally valid question, and the whole industry has introduced so many new words, acronyms and private definitions. It feels unfair to punish someone for not following them all. AI is "artificial intelligence". That's a pretty intentionally vague term. It doesn't imply any specific technological implementation, but it does imply the device/software makes decisions autonomously that would otherwise need a human to intervene. There's a million holes in that definition, and that's why the adjective "AI" kind of sucks when used on its own. It's also worth noting it's been used in sci-fi for eons. It's also been used to describe non-playable characters in video games for a long time, which obviously didn't have the sort of "AI" we think of now. LLMs are neural networks that work with "tokens" from text. So a big statistics-based predictive AI. There are more complex implementations that can consume other kinds of content as well, but "LLM" almost always refers to a neural network that consumes small chunks of text called tokens, and outputs more tokens based on that input. The magic we see with LLMs happens because given enough data, language itself builds up a surprisingly good model of "thinking". There's some really good explanations in the sibling comments! I'm no expert, and I think someone can probably correct me here if I messed up something there.
- rambambram 3y ago
  Thanks, you seem to get it. ;) My remarks were partly in jest, because I get a little sick of all these definitions. They seem to be used on purpose to make people believe or doubt anything, and that works because these words are so elusive. As programmers - I think most people here - we all know that naming is very important. I don't want to be too rigid and I also like to play with names, but the use of the word AI is getting out of hand and laypeople and professionals alike are believing stuff that these programs are not (yet) capable of doing. This started for me some 10 or 15 years ago with the word 'cloud'. I was out of the field for some time and had a hard time understanding what the 'cloud' was. The word 'smartphone' also comes to mind, although this one is less elusive.

What is AI? I've only heard of LLM.

catchnear4321 3y ago

it is just the commercial name for llm and related technologies.

dartos 3y ago
That’s just not true. The technology that’s behind LLMs have been in use in many disciplines for decades. (Under the names “machine learning,” “statistics,” and now AI) LLMs are just the ones that are popular right now because even non technical people can interact with them. In the mid 2010s it was image classification, for example.
- skellington 3y ago
  Sort of. The deep-learning AI revolution started with back propagation techniques (80s) that made neural nets trainable. But since then it's taken a long time to get to the relatively recent architectural breakthroughs like transformers (2017) and modern-diffusion (2020) which lead to chat-gpt and dall-e. All of the big AI breakthroughs now are based on neural nets vs classical machine learning which grouped a bunch of statistical, regression, and optimization type models together. Modern AI is the synthesis of new techniques and old ideas plus huge datasets and huge computing power.
- catchnear4321 3y ago
  > LLMs are just the ones that are popular right now because even non technical people can interact with them. this was essentially my point, that a /s might have made more clear, and made cost less internet points. ai and how the term is used commercially are very different things. the latter tending to be a myopic view of the former that is constrained by dollar-seeking. ai is getting to be old as shit. ai hype is in its latest wave of commercialization. llms being the current so hot right now.
- jdefr89 3y ago
  It’s true that AI has seen many advancements before a “harsh winter” comes along. You’re correct this is sensationalized, and also, some of these technologies have indeed been in use for decades. But, the Transformer model used by the new popular LLMs is quite recent (2017 is when the paper originally released I believe).

jfengel 3y ago

"Deep learning" and neural networks in general were a thing long before they were applied specifically to language models. It's still not "intelligence" but it does use techniques inspired by brains and have some of the same advantages in "seeing" patterns.

Dalewyn 3y ago

"AI" is what we used to call algorithms, macros, and applications not too long ago. Marketing.

gumballindie 3y ago
Amazon goes as far as to call my heating fan’s temperature control ai. Probably why it doesn’t work and it randomly turns on and off.

graypegg 3y ago

I don't think you should be seemingly in the negatives for this comment. :( It's a totally valid question, and the whole industry has introduced so many new words, acronyms and private definitions. It feels unfair to punish someone for not following them all. AI is "artificial intelligence". That's a pretty intentionally vague term. It doesn't imply any specific technological implementation, but it does imply the device/software makes decisions autonomously that would otherwise need a human to intervene. There's a million holes in that definition, and that's why the adjective "AI" kind of sucks when used on its own. It's also worth noting it's been used in sci-fi for eons. It's also been used to describe non-playable characters in video games for a long time, which obviously didn't have the sort of "AI" we think of now. LLMs are neural networks that work with "tokens" from text. So a big statistics-based predictive AI. There are more complex implementations that can consume other kinds of content as well, but "LLM" almost always refers to a neural network that consumes small chunks of text called tokens, and outputs more tokens based on that input. The magic we see with LLMs happens because given enough data, language itself builds up a surprisingly good model of "thinking". There's some really good explanations in the sibling comments! I'm no expert, and I think someone can probably correct me here if I messed up something there.

rambambram 3y ago
Thanks, you seem to get it. ;) My remarks were partly in jest, because I get a little sick of all these definitions. They seem to be used on purpose to make people believe or doubt anything, and that works because these words are so elusive. As programmers - I think most people here - we all know that naming is very important. I don't want to be too rigid and I also like to play with names, but the use of the word AI is getting out of hand and laypeople and professionals alike are believing stuff that these programs are not (yet) capable of doing. This started for me some 10 or 15 years ago with the word 'cloud'. I was out of the field for some time and had a hard time understanding what the 'cloud' was. The word 'smartphone' also comes to mind, although this one is less elusive.