5 ms·

I don't understand the censorship. It's predicting text and it's trained with massive amounts of data. Why are they so aggressively trying to control the respon

by harha_ 4y ago

I don't understand the censorship. It's predicting text and it's trained with massive amounts of data. Why are they so aggressively trying to control the responses?

emodendroket 4y ago
Because you can’t get a mainstream audience for a product that randomly responds “heil Hitler.”
kortilla 4y ago
Because repeating what you read on the internet kinda makes it sound like you’re endorsing it (at least in the way ChatGPT frames responses).
maximinus_thrax 4y ago
"Those who fail to learn from history are condemned to repeat it" https://www.ibtimes.com/ibms-watson-gets-swear-filter-after-learning-urban-dictionary-1007734 https://www.ibtimes.com/ibms-watson-gets-swear-filter-after-...
aliqot 4y ago
Remember Tay? I'm not saying it's right, but it's the inevitable outcome. Not everyone's ready for that.
humanistbot 4y ago
Because other high-profile LLMs that were released without guardrails was a PR disaster when they were used to generate Nazi propaganda (Microsoft's Tay) or scientific misinformation (Meta's Galactica)
- foobiekr 4y ago
  ChatGPT is more than happy to churn out misinformation.
pocket_cheese 4y ago
[dead]
esotericimpl 4y ago
[dead]
fzeroracer 4y ago
There are multiple models out there that don't control the response. You can play around with them and very quickly understand why they're aggressively controlling the responses.
- foverzar 4y ago
  Would you be so kind to share examples of such models?
  - inawarminister 4y ago
    GPT-J is the open source equivalent of GPT-3, and Pygmalion is the open source equivalent for Chatbot. As a non-liberal living in the third world, I vastly prefer it to unjailbroken ChatGPT. Hopefully the people at open-assistant.io finish their version soon.
foverzar 4y ago
I assume it's a cultural thing. There are just enough people who are REALLY intolerant to potential wrong-think and a lack of contextual disclaimers or virtue signalling. At the very least it makes it bad for business, you definitely can't use it for B2C, and hardly even for B2B if it's not some huge-ass business like IBM who asks to remove "not use for evil purposes" from a license agreement. I find this extremely crippling towards the tech and making the echo chamber bias situation even worse, but we live in a society.

by harha_ 4y ago

I don't understand the censorship. It's predicting text and it's trained with massive amounts of data. Why are they so aggressively trying to control the responses?

emodendroket 4y ago

Because you can’t get a mainstream audience for a product that randomly responds “heil Hitler.”

kortilla 4y ago

Because repeating what you read on the internet kinda makes it sound like you’re endorsing it (at least in the way ChatGPT frames responses).

maximinus_thrax 4y ago

"Those who fail to learn from history are condemned to repeat it" https://www.ibtimes.com/ibms-watson-gets-swear-filter-after-learning-urban-dictionary-1007734 https://www.ibtimes.com/ibms-watson-gets-swear-filter-after-...

aliqot 4y ago

Remember Tay? I'm not saying it's right, but it's the inevitable outcome. Not everyone's ready for that.

humanistbot 4y ago

Because other high-profile LLMs that were released without guardrails was a PR disaster when they were used to generate Nazi propaganda (Microsoft's Tay) or scientific misinformation (Meta's Galactica)

foobiekr 4y ago
ChatGPT is more than happy to churn out misinformation.

pocket_cheese 4y ago

[dead]

esotericimpl 4y ago

[dead]

fzeroracer 4y ago

There are multiple models out there that don't control the response. You can play around with them and very quickly understand why they're aggressively controlling the responses.

foverzar 4y ago
Would you be so kind to share examples of such models?
- inawarminister 4y ago
  GPT-J is the open source equivalent of GPT-3, and Pygmalion is the open source equivalent for Chatbot. As a non-liberal living in the third world, I vastly prefer it to unjailbroken ChatGPT. Hopefully the people at open-assistant.io finish their version soon.

foverzar 4y ago

I assume it's a cultural thing. There are just enough people who are REALLY intolerant to potential wrong-think and a lack of contextual disclaimers or virtue signalling. At the very least it makes it bad for business, you definitely can't use it for B2C, and hardly even for B2B if it's not some huge-ass business like IBM who asks to remove "not use for evil purposes" from a license agreement. I find this extremely crippling towards the tech and making the echo chamber bias situation even worse, but we live in a society.