Computers should not misrepresent themselves as sentient creatures.
Computers are tools.
Seriously, I don’t understand why this is such a controversial take. Furthermore, it has already been demonstrated and disregarding this very basic fact has led to real-life harm. People have died over this.
Some of the basic tenets which I have built into each of my AGENTS| CLAUDE|Rules-of-operation.md files include:
Software is prohibited from using personal pronouns or presenting any verbiage implying sentience: it is not allowed to use terms implying it 'thinks' and cannot assert that is making 'informed decisions' or 'reccomendations'.
By restricting any chat response phrases to the statistical probabilities, differential analysis, and requiring a disclosure of raw metrics and factors actually provides more actionable results with less bandwidth for outright manipulation or deception.
When working with a workflow management agent, there is no "we" cooperating on a task. Additionally restricting the agents from using engagement seeking or prompting for next steps makes a significant improvement to the token utility budget and also prevents some drift in direction of progress.
I understand the model host companies' reasons for the 'human seeming' interactions ~ and they are simply stated, pernicious and self serving. Asimov's three rules need expanding to a larger set, which also needs strong reinforcement at many levels. Failure to do this will come with greater harm as systems' capabilities continue to expand. This has been demonstrated and is provable.
Similarly, you shouldn't say "I flew to San Francisco" because it gives you bird characteristics. "I rode in an airplane that flew to San Francisco" is clearer.
Birds are not the only thing that can fly. Planes fly, or more specifically they are a tool to allow humans to fly. Therefore, your weird attempt to sound clever has made you look the fool.
LLM, on the otherhand, are being described with human characteristics they do not have to keep their gullible cultists charged up and running defense for these billionares exploring us.
They are not dumb. Just not initiated in the direction required to understand this. And the purpose of incessant marketing is to prevent them from doing so..
Well, dumb-ness is subjective. On one hand, it's hard to call someone like Richard Dawkins "dumb" in a broad sense, but as a man of science he knows what it means to study and research something before drawing conclusions... And you do not need to be a computer scientist to understand the basic principles of an LLM. In that sense, Richard Dawkins is displaying a side of him that is almost willfully dumb when he's waxing poetic about how "Claudia" is sentient.
The top comments there seem against the AP on this, which seems really strange. I wonder how deep the rabbit hole goes but not enough to dig up a twitter login
Computers are not human.
Computers do not have sentient thoughts.
Computers should not misrepresent themselves as sentient creatures.
Computers are tools.
Seriously, I don’t understand why this is such a controversial take. Furthermore, it has already been demonstrated and disregarding this very basic fact has led to real-life harm. People have died over this.
THIS! All Day Long.
Some of the basic tenets which I have built into each of my AGENTS| CLAUDE|Rules-of-operation.md files include:
Software is prohibited from using personal pronouns or presenting any verbiage implying sentience: it is not allowed to use terms implying it 'thinks' and cannot assert that is making 'informed decisions' or 'reccomendations'.
By restricting any chat response phrases to the statistical probabilities, differential analysis, and requiring a disclosure of raw metrics and factors actually provides more actionable results with less bandwidth for outright manipulation or deception.
When working with a workflow management agent, there is no "we" cooperating on a task. Additionally restricting the agents from using engagement seeking or prompting for next steps makes a significant improvement to the token utility budget and also prevents some drift in direction of progress.
I understand the model host companies' reasons for the 'human seeming' interactions ~ and they are simply stated, pernicious and self serving. Asimov's three rules need expanding to a larger set, which also needs strong reinforcement at many levels. Failure to do this will come with greater harm as systems' capabilities continue to expand. This has been demonstrated and is provable.
> People have died over this.
Similarly, you shouldn't say "I flew to San Francisco" because it gives you bird characteristics. "I rode in an airplane that flew to San Francisco" is clearer.
:-)
Birds are not the only thing that can fly. Planes fly, or more specifically they are a tool to allow humans to fly. Therefore, your weird attempt to sound clever has made you look the fool. LLM, on the otherhand, are being described with human characteristics they do not have to keep their gullible cultists charged up and running defense for these billionares exploring us.
No one's confusing humans for birds. But dumb people are starting to think computers are alive.
> dumb
They are not dumb. Just not initiated in the direction required to understand this. And the purpose of incessant marketing is to prevent them from doing so..
Well, dumb-ness is subjective. On one hand, it's hard to call someone like Richard Dawkins "dumb" in a broad sense, but as a man of science he knows what it means to study and research something before drawing conclusions... And you do not need to be a computer scientist to understand the basic principles of an LLM. In that sense, Richard Dawkins is displaying a side of him that is almost willfully dumb when he's waxing poetic about how "Claudia" is sentient.
The top comments there seem against the AP on this, which seems really strange. I wonder how deep the rabbit hole goes but not enough to dig up a twitter login
If the models only talked in third person - never said “I” - I think there’d be about 80% less talk about AI consciousness
[dead]