5 ms·

This current iteration of "AI", by which I mean LLMs, seems only good at enabling generative spam and disinformation. I think regulations are exactly what's re

by nvy 2y ago

This current iteration of "AI", by which I mean LLMs, seems only good at enabling generative spam and disinformation. I think regulations are exactly what's required.

robertlagrant 2y ago
> seems only good at enabling generative spam and disinformation This doesn't seem to make sense. Although it uses the word "disinformation", so perhaps an LLM wrote it.
- nvy 2y ago
  What about it doesn't make sense to you? LLMs aren't intelligent and the hallucinations problem is fundamental to the technology, and cannot be solved. Therefore definitionally an LLM will never be fully trustworthy. But it's great at generating plausible-sounding stuff, which makes it great for propaganda and disinformation.
  - danielbln 2y ago
    I've implemented many use cases professionally where LLMs sit at the center, from customer support, to document classification, general knowledge management and many more. Personally I use them frequently for coding, to carry out various text based tasks (turning information from one format to another etc). Hallucination can be significantly reduced by injecting knowledge into the request, to the degree that LLM output is suitable for various business use cases. Saying LLMs are not useful for anything but generating spam is objectively wrong, as to disqualify your commenta from serious discussion about the topic. There are many issues with GenAI that are worth discussing, saying they are generally useless is not the way.
  - robertlagrant 2y ago
    > What about it doesn't make sense to you? That you said it was only good for those things. You changed it in this message, as I'm sure you're aware.
ch4s3 2y ago
> seems only good at enabling generative spam and disinformation. That seems overly pessimistic. They're pretty good at generating summaries of text, simple machine translation, generating alternative prose, turning unstructured text into table-like data, providing a text interface over a corpus of facts, and probably a bunch of things I'm not thinking of.
- nvy 2y ago
  >They're pretty good at generating summaries of text, simple machine translation, generating alternative prose They're not, though. For these tasks they are mediocre at best.
  - ch4s3 2y ago
    I disagree strongly. For the two two languages I can read the machine translations are quite good. I wouldn't count on it to faithfully render poetry or subtle elements in literature, but it's great for news articles or reading a form post. For generating alternative prose, what I mean is using a prompt like: given this sentence/paragraph give me 3 alternatives that capture a particular mood/theme/style. Doing that is a great way to try to rewrite flat or repetitive prose or to shake up a bit of writer's block. It's excellent for turning some strung together facts into an email that I can quickly edit into my own voice a bit then send. You may not find that valuable, but I assure you that other people do. Summaries also work pretty well in my experience and for my purposes. I don't need 100% perfection for every big document I come across sometimes I need a paragraph that hits the main idea(s) It's good enough for that and why wouldn't a text prediction tool be good at that?
oceanplexian 2y ago
The government has the legal authority to kill, imprison, and tax citizens. It can destroy entire industries with the stroke of a pen. The military can wipe other nations off the face of the Earth. And you want to wield that awesome power for some trivial problem like spam on the Internet? This is exactly the kind of overreach the founding fathers intended to prevent.
- ronsor 2y ago
  You've hit the nail on the head. The government could stop all theft if people couldn't leave their homes except with police escort. Obviously this is a ludicrous idea. Ergo we have collectively (and correctly) decided that tolerating some level of bad actors is better than melting away all freedom.
  - bumby 2y ago
    The hard part is finding the right balance between security and freedom, no?
- nvy 2y ago
  The "free market" mostly isn't. LLMs are actively harmful to society and in a way that individuals cannot combat, and spam/fraud is merely the obvious way.
smileson2 2y ago
[dead]

by nvy 2y ago

This current iteration of "AI", by which I mean LLMs, seems only good at enabling generative spam and disinformation. I think regulations are exactly what's required.

robertlagrant 2y ago

> seems only good at enabling generative spam and disinformation This doesn't seem to make sense. Although it uses the word "disinformation", so perhaps an LLM wrote it.

nvy 2y ago
What about it doesn't make sense to you? LLMs aren't intelligent and the hallucinations problem is fundamental to the technology, and cannot be solved. Therefore definitionally an LLM will never be fully trustworthy. But it's great at generating plausible-sounding stuff, which makes it great for propaganda and disinformation.
- danielbln 2y ago
  I've implemented many use cases professionally where LLMs sit at the center, from customer support, to document classification, general knowledge management and many more. Personally I use them frequently for coding, to carry out various text based tasks (turning information from one format to another etc). Hallucination can be significantly reduced by injecting knowledge into the request, to the degree that LLM output is suitable for various business use cases. Saying LLMs are not useful for anything but generating spam is objectively wrong, as to disqualify your commenta from serious discussion about the topic. There are many issues with GenAI that are worth discussing, saying they are generally useless is not the way.
- robertlagrant 2y ago
  > What about it doesn't make sense to you? That you said it was only good for those things. You changed it in this message, as I'm sure you're aware.

ch4s3 2y ago

> seems only good at enabling generative spam and disinformation. That seems overly pessimistic. They're pretty good at generating summaries of text, simple machine translation, generating alternative prose, turning unstructured text into table-like data, providing a text interface over a corpus of facts, and probably a bunch of things I'm not thinking of.

nvy 2y ago
>They're pretty good at generating summaries of text, simple machine translation, generating alternative prose They're not, though. For these tasks they are mediocre at best.
- ch4s3 2y ago
  I disagree strongly. For the two two languages I can read the machine translations are quite good. I wouldn't count on it to faithfully render poetry or subtle elements in literature, but it's great for news articles or reading a form post. For generating alternative prose, what I mean is using a prompt like: given this sentence/paragraph give me 3 alternatives that capture a particular mood/theme/style. Doing that is a great way to try to rewrite flat or repetitive prose or to shake up a bit of writer's block. It's excellent for turning some strung together facts into an email that I can quickly edit into my own voice a bit then send. You may not find that valuable, but I assure you that other people do. Summaries also work pretty well in my experience and for my purposes. I don't need 100% perfection for every big document I come across sometimes I need a paragraph that hits the main idea(s) It's good enough for that and why wouldn't a text prediction tool be good at that?

oceanplexian 2y ago

The government has the legal authority to kill, imprison, and tax citizens. It can destroy entire industries with the stroke of a pen. The military can wipe other nations off the face of the Earth. And you want to wield that awesome power for some trivial problem like spam on the Internet? This is exactly the kind of overreach the founding fathers intended to prevent.

ronsor 2y ago
You've hit the nail on the head. The government could stop all theft if people couldn't leave their homes except with police escort. Obviously this is a ludicrous idea. Ergo we have collectively (and correctly) decided that tolerating some level of bad actors is better than melting away all freedom.
- bumby 2y ago
  The hard part is finding the right balance between security and freedom, no?
nvy 2y ago
The "free market" mostly isn't. LLMs are actively harmful to society and in a way that individuals cannot combat, and spam/fraud is merely the obvious way.

smileson2 2y ago

[dead]