This is a highly accurate description of how AI models work. (Speaking as a programmer who has done significant user-space work (open source models, custom prompts, and harness development -- though not the lower-level model training).
The "video games depend on user input" and "system prompts as editorial decisions" are both solid analogies that should make sense to non-technical users using AI from web interfaces.
Nice work. I'm adopting those analogies for my own future discussions. I'm looking forward to the next post in this series!
We have laws against harming other people physically - including, say, mild cases of poisoning over a long period. To my knowledge we don't have laws about human speech or media injuring people in this way because - to date - speech and media haven't been the kinds of things that can cause physical harm (although "speech acts", in J.L. Austin’s sense, can lead directly to physical harm and thus may constitute illegal acts).
Let me offer a hypothetical. A social media platform has an algorithm that blindly maximizes user engagement for profit. It uses proven addictive behavioral schemes for this purpose. The platform's own scientists establish that this works very effectively - users experience addiction to the platform, and in some cases this is quite severe - measurable physically. Unlike in, say, gambling, the users have no way of knowing they are being subjected to these covertly manipulative addictive schemes. The algorithm generates huge profits and also has some undeniably beneficial uses, so the platform takes no action.
Should this algorithm be regulated as hazardous to health? Or should it be protected by the First Amendment? Should it be age-regulated or come with health warnings?
As I say - entirely hypothetical. Couldn't possibly happen in a world of responsible media platforms.
Put aside how the knowledge was "accumulated" (ahem). Hallucination is part of this tech and cannot be avoided, only obscured -- by design. Time and time again, AI makes things up and presents them with a plausible veneer. If anything is getting in the way of the world's knowledge, it is AI. This is a massive liability issue.
This is a highly accurate description of how AI models work. (Speaking as a programmer who has done significant user-space work (open source models, custom prompts, and harness development -- though not the lower-level model training).
The "video games depend on user input" and "system prompts as editorial decisions" are both solid analogies that should make sense to non-technical users using AI from web interfaces.
Nice work. I'm adopting those analogies for my own future discussions. I'm looking forward to the next post in this series!
All? No. It doesn’t apply to people labeled mentally ill, and organizations like FIRE doesn’t pretend to defend such people.
This is like declaring all men are created but carving out an exception for men.
We have laws against harming other people physically - including, say, mild cases of poisoning over a long period. To my knowledge we don't have laws about human speech or media injuring people in this way because - to date - speech and media haven't been the kinds of things that can cause physical harm (although "speech acts", in J.L. Austin’s sense, can lead directly to physical harm and thus may constitute illegal acts).
Let me offer a hypothetical. A social media platform has an algorithm that blindly maximizes user engagement for profit. It uses proven addictive behavioral schemes for this purpose. The platform's own scientists establish that this works very effectively - users experience addiction to the platform, and in some cases this is quite severe - measurable physically. Unlike in, say, gambling, the users have no way of knowing they are being subjected to these covertly manipulative addictive schemes. The algorithm generates huge profits and also has some undeniably beneficial uses, so the platform takes no action.
Should this algorithm be regulated as hazardous to health? Or should it be protected by the First Amendment? Should it be age-regulated or come with health warnings?
As I say - entirely hypothetical. Couldn't possibly happen in a world of responsible media platforms.
Put aside how the knowledge was "accumulated" (ahem). Hallucination is part of this tech and cannot be avoided, only obscured -- by design. Time and time again, AI makes things up and presents them with a plausible veneer. If anything is getting in the way of the world's knowledge, it is AI. This is a massive liability issue.