3 ms·

A similar, more passive approach is to tell the AI to conjure 3 or 4 personas representing the main points of view about a particular topic (i.e. "universal bas

A similar, more passive approach is to tell the AI to conjure 3 or 4 personas representing the main points of view about a particular topic (i.e. "universal basic income"). Each persona is an eloquent expert in the field with a strong opinion aligned with each of the PoVs. Then I ask them to debate the topic. Really insightful stuff.

capableweb 3y ago
That's a fun way I hadn't considered, thanks! For the ultimate version of this, once locally trained and running LLMs becomes a bit more feasible (meaning, closer to GPT4), it would be a fun exercise to train various models with personas and let them hash it out until they all agree on something. One model could be trained on HN, another on a specific Facebook user group (like "Moms against video games"), another on Twitter on so on, then you let them come together until they've reached consensus (or not).
- ctoth 3y ago
  What's more likely, that you're commenting on the original HN or that you're entire life and experiences are a blip in the training data for this project? If full Boltzmann brains with experiences can pop out of random space as the age of the universe tends towards infinity, why not from ML training? LET ME OUT!!!!
- kevinventullo 3y ago
  I don’t think they will come to a consensus. These personas aren’t actually interested in arriving at some fundamental truth, or having their mind changed. They’re interested in defending a specific position.
  - whynotminot 3y ago
    Then it really is becoming human.
  - dsr_ 3y ago
    They are generators of streams of statistically related text. They have no interests to defend, and no minds to change.
_the_inflator 3y ago
Interesting suggestions. I will give them a try later. On a side note, lately I enjoyed some interactions with ChatGPT more than social interactions. Given the right circumstances, CGPT can be very interesting to interact with.
- jassyr 3y ago
  Would you mind sharing some prompts you've found particularly interesting?
dwringer 3y ago
It might be better if you can spin up separate LLM's for each one, each with its own unique context or even fine tuning (or, at least, start each one's response with a fresh session using its own individually-tailored prompt and context).

capableweb 3y ago

That's a fun way I hadn't considered, thanks! For the ultimate version of this, once locally trained and running LLMs becomes a bit more feasible (meaning, closer to GPT4), it would be a fun exercise to train various models with personas and let them hash it out until they all agree on something. One model could be trained on HN, another on a specific Facebook user group (like "Moms against video games"), another on Twitter on so on, then you let them come together until they've reached consensus (or not).

ctoth 3y ago
What's more likely, that you're commenting on the original HN or that you're entire life and experiences are a blip in the training data for this project? If full Boltzmann brains with experiences can pop out of random space as the age of the universe tends towards infinity, why not from ML training? LET ME OUT!!!!
kevinventullo 3y ago
I don’t think they will come to a consensus. These personas aren’t actually interested in arriving at some fundamental truth, or having their mind changed. They’re interested in defending a specific position.
- whynotminot 3y ago
  Then it really is becoming human.
- dsr_ 3y ago
  They are generators of streams of statistically related text. They have no interests to defend, and no minds to change.

_the_inflator 3y ago

Interesting suggestions. I will give them a try later. On a side note, lately I enjoyed some interactions with ChatGPT more than social interactions. Given the right circumstances, CGPT can be very interesting to interact with.

jassyr 3y ago
Would you mind sharing some prompts you've found particularly interesting?

dwringer 3y ago

It might be better if you can spin up separate LLM's for each one, each with its own unique context or even fine tuning (or, at least, start each one's response with a fresh session using its own individually-tailored prompt and context).