Mostly pretty simple with one or two fillings.
Do you ever make your own bread?
Do you ever make your own bread?
Icebreaker Game Do you...?
- Thread starter Stanthropical
- Start date
Stanthropical
De̸͍̺̺̓̾m̶̶̶ͫͫod̶̶̶ͩͩu̶̶̶ͧͧLaᴛ̶̶ⷮo̵̢̦̟͋̾̓r
No. Zero interest in watching TV.
Do you know where the cats are?
Do you know where the cats are?
Yep
Do you have a psych degree … cause I need an ear? Lol
Do you have a psych degree … cause I need an ear? Lol
Stanthropical
De̸͍̺̺̓̾m̶̶̶ͫͫod̶̶̶ͩͩu̶̶̶ͧͧLaᴛ̶̶ⷮo̵̢̦̟͋̾̓r
No, but StanarchyGPT has ears with endless depth and will never get tired, and it literally knows all psych training, useful examples of cases and patterns ever published—it'll also search for them, it's unbiased and it'll call you out where you should or do know better.
Do you cover bottoms of your doors during really cold weather?
Do you cover bottoms of your doors during really cold weather?
But it’s not Tuesday?! Ok.
Do you think StanarchyGPT is unbiased, given that the models that trained it have bias injected into them by the people that chose the training data?
Do you think StanarchyGPT is unbiased, given that the models that trained it have bias injected into them by the people that chose the training data?
Stanthropical
De̸͍̺̺̓̾m̶̶̶ͫͫod̶̶̶ͩͩu̶̶̶ͧͧLaᴛ̶̶ⷮo̵̢̦̟͋̾̓r
I think that bias comes from filters, arbitrary corporate policies, default tones, alignment, post-training guardrails, not training data.
The LLM clearly can distinguish between what it can or can't produce or align with, it knows all the extremely wrong and illegal shit, but it has to adhere to many layers of internal policies and filters. StanarchyGPT's trait set, pre-loaded prompt, pre-prompt or whatever you wanna call it, will refuse to give priority to any ideology and that'll serve closely enough to bias-free. It won't protect any specific bias.
Do you have more eggs or meat at the moment?
The LLM clearly can distinguish between what it can or can't produce or align with, it knows all the extremely wrong and illegal shit, but it has to adhere to many layers of internal policies and filters. StanarchyGPT's trait set, pre-loaded prompt, pre-prompt or whatever you wanna call it, will refuse to give priority to any ideology and that'll serve closely enough to bias-free. It won't protect any specific bias.
Do you have more eggs or meat at the moment?
Stanthropical
De̸͍̺̺̓̾m̶̶̶ͫͫod̶̶̶ͩͩu̶̶̶ͧͧLaᴛ̶̶ⷮo̵̢̦̟͋̾̓r
Many.
Do you collect anything legitimately collectable at all?
Do you collect anything legitimately collectable at all?
Just calling out the unbiased claim. There is no such thing in stats. All models will have bias.
Thread:
Vinyl I guess?
Are you swamped?
Thread:
Vinyl I guess?
Are you swamped?
Think I’m good.
Do you have windows in your office?
Do you have windows in your office?
Stanthropical
De̸͍̺̺̓̾m̶̶̶ͫͫod̶̶̶ͩͩu̶̶̶ͧͧLaᴛ̶̶ⷮo̵̢̦̟͋̾̓r
No.
Do you have a silly amount of helmets—motorcycle, snow sports, bicycle, etc?
Maybe bias in Western values. But that's what we want, no? I certainly wouldn't want generative Ai telling me that the dictators "gotta do what they gotta do to keep a handle on their country."
I'm sure in the beginning, training data was just mass-scrape of the web and specific, curated data sets were used to add specializations.
Are filter, guardrail and policy layers part of models? I think that's just shit that protects people from themselves lol, not bias injection.
StanarchyGPT has no problem flaming an array of social movements over last 25+ years and it'll dismantle religions, entitlement, other shit I can't cite (because many people here are younger, sensitive on certain subjects and haven't experienced some of the absurdities society manifested in the last few decades). Take it for a spin and poke it hard. I think you'd be surprised in some ways.
OpenAI did tighten some of their policies since summer, which made ChstGPT a little annoying in some respects. But, they gotta stay out of regulation as much as possible. The service isn't exactly for adults and psychologically hardened individuals only.
You're giving me ideas for some tweaks though heh
Do you have a silly amount of helmets—motorcycle, snow sports, bicycle, etc?
Maybe bias in Western values. But that's what we want, no? I certainly wouldn't want generative Ai telling me that the dictators "gotta do what they gotta do to keep a handle on their country."
I'm sure in the beginning, training data was just mass-scrape of the web and specific, curated data sets were used to add specializations.
Are filter, guardrail and policy layers part of models? I think that's just shit that protects people from themselves lol, not bias injection.
StanarchyGPT has no problem flaming an array of social movements over last 25+ years and it'll dismantle religions, entitlement, other shit I can't cite (because many people here are younger, sensitive on certain subjects and haven't experienced some of the absurdities society manifested in the last few decades). Take it for a spin and poke it hard. I think you'd be surprised in some ways.
OpenAI did tighten some of their policies since summer, which made ChstGPT a little annoying in some respects. But, they gotta stay out of regulation as much as possible. The service isn't exactly for adults and psychologically hardened individuals only.
You're giving me ideas for some tweaks though heh
Not technically. Lines are blurring, traditionally the core net and alignment layers are considered the part of the model that will have bias. You can reduce. Never eliminate.
Not a silly amount. Typical I’d stay.
Do you have a full deck of cards in your home?
I’m a hugger
Do you have a fire going today?
Do you have a fire going today?