Soumil Rathi
@soumilrathi.bsky.social
Massive AI enthusiast | Super interested in cognition and general intelligence | Founder | Developer
Yes, it's wrong. I absolutely acknowledge that, and that is my personal view on this.
Would I train an LLM based on data that I do not own? No.
But something being morally wrong doesn't make it bannable. At the end of the day, there's a distinction between morality and legality.
Would I train an LLM based on data that I do not own? No.
But something being morally wrong doesn't make it bannable. At the end of the day, there's a distinction between morality and legality.
November 28, 2024 at 5:32 AM
Yes, it's wrong. I absolutely acknowledge that, and that is my personal view on this.
Would I train an LLM based on data that I do not own? No.
But something being morally wrong doesn't make it bannable. At the end of the day, there's a distinction between morality and legality.
Would I train an LLM based on data that I do not own? No.
But something being morally wrong doesn't make it bannable. At the end of the day, there's a distinction between morality and legality.
How is that relevant to my point? Obviously there are bad actors in the world, that's not a crazy gotcha.
Having extra regulations on the good actors doesn't stop the bad actors anyways
Having extra regulations on the good actors doesn't stop the bad actors anyways
November 28, 2024 at 5:31 AM
How is that relevant to my point? Obviously there are bad actors in the world, that's not a crazy gotcha.
Having extra regulations on the good actors doesn't stop the bad actors anyways
Having extra regulations on the good actors doesn't stop the bad actors anyways
As for the moral issue, I personally think its unethical to take everyones posts without asking for their consent. Obviously.
But that doesn't mean its illegal. Your (or mine) personal views on what someone does doesn't make it bannable.
But that doesn't mean its illegal. Your (or mine) personal views on what someone does doesn't make it bannable.
November 28, 2024 at 5:29 AM
As for the moral issue, I personally think its unethical to take everyones posts without asking for their consent. Obviously.
But that doesn't mean its illegal. Your (or mine) personal views on what someone does doesn't make it bannable.
But that doesn't mean its illegal. Your (or mine) personal views on what someone does doesn't make it bannable.
You literally CANNOT opt out of them using it.
The post you are reposting is absolutely correct, please read it properly. It says you can't opt out of them using it, but they need to have a way for you to delete it. They do have that.
(1/2)
The post you are reposting is absolutely correct, please read it properly. It says you can't opt out of them using it, but they need to have a way for you to delete it. They do have that.
(1/2)
November 28, 2024 at 5:29 AM
You literally CANNOT opt out of them using it.
The post you are reposting is absolutely correct, please read it properly. It says you can't opt out of them using it, but they need to have a way for you to delete it. They do have that.
(1/2)
The post you are reposting is absolutely correct, please read it properly. It says you can't opt out of them using it, but they need to have a way for you to delete it. They do have that.
(1/2)
Dont recall ever saying that.
November 28, 2024 at 5:27 AM
Dont recall ever saying that.
Its a factual statement. There is no violation of TOS or copyright laws happening here.
You’re posting on an open platform and then complaining that people are taking advantage of the open nature of the platform
You’re posting on an open platform and then complaining that people are taking advantage of the open nature of the platform
November 27, 2024 at 11:09 PM
Its a factual statement. There is no violation of TOS or copyright laws happening here.
You’re posting on an open platform and then complaining that people are taking advantage of the open nature of the platform
You’re posting on an open platform and then complaining that people are taking advantage of the open nature of the platform
Instead of attacking me personally, why don’t you attack what I said
November 27, 2024 at 11:07 PM
Instead of attacking me personally, why don’t you attack what I said
I think there should be the ability to self moderate, by the users of the platform as opposed to the people running it.
No community of users would support CSAM
No community of users would support CSAM
November 27, 2024 at 11:06 PM
I think there should be the ability to self moderate, by the users of the platform as opposed to the people running it.
No community of users would support CSAM
No community of users would support CSAM
and not even the law for decentralized platforms
November 27, 2024 at 10:46 PM
and not even the law for decentralized platforms
no limits except the law
November 27, 2024 at 10:46 PM
no limits except the law
No, an OPEN site does. That's what huggingface is. I hope they don't force remove a dataset for personal choices - that's antithetical to their mission
November 27, 2024 at 10:39 PM
No, an OPEN site does. That's what huggingface is. I hope they don't force remove a dataset for personal choices - that's antithetical to their mission
I fail to see what's wrong with this.
Sure, there's a moral issue with using people's posts without their consent, but no TOS or copyright problems here.
Sure, there's a moral issue with using people's posts without their consent, but no TOS or copyright problems here.
November 27, 2024 at 10:38 PM
I fail to see what's wrong with this.
Sure, there's a moral issue with using people's posts without their consent, but no TOS or copyright problems here.
Sure, there's a moral issue with using people's posts without their consent, but no TOS or copyright problems here.
this gets sm worse when collaborating as well.
keeping track of what version of code everyone has and then trying to sort it all out later gets so painful
keeping track of what version of code everyone has and then trying to sort it all out later gets so painful
November 26, 2024 at 10:54 PM
this gets sm worse when collaborating as well.
keeping track of what version of code everyone has and then trying to sort it all out later gets so painful
keeping track of what version of code everyone has and then trying to sort it all out later gets so painful
thats very interesting! can you share a few links on this? would love to read up more
November 25, 2024 at 10:39 PM
thats very interesting! can you share a few links on this? would love to read up more
how much ram do you have? if you've got 16GB+ you could probably download a local LLM and use it to increase what "offline-capable" means for you - I tried it out last flight and worked well!
November 25, 2024 at 6:59 AM
how much ram do you have? if you've got 16GB+ you could probably download a local LLM and use it to increase what "offline-capable" means for you - I tried it out last flight and worked well!
If instead you draw representations based on actual knowledge relationships (symbolically?), its possible those models get closer to objective truth imo
2/2
2/2
November 25, 2024 at 6:54 AM
If instead you draw representations based on actual knowledge relationships (symbolically?), its possible those models get closer to objective truth imo
2/2
2/2
I feel like to some extent this is also the fault of the algorithms themselves. When you have algorithms that learn representations based on historical next words, they're GOING to have all historical biases.
1/2
1/2
November 25, 2024 at 6:54 AM
I feel like to some extent this is also the fault of the algorithms themselves. When you have algorithms that learn representations based on historical next words, they're GOING to have all historical biases.
1/2
1/2
bsky feels like a genuine alternative to twitter. good amount of actual ideas and enjoyable content
November 25, 2024 at 6:46 AM
bsky feels like a genuine alternative to twitter. good amount of actual ideas and enjoyable content
agree with you, but then the question is how, though? how do you represent *all* multimodal input and output symbolically.
November 25, 2024 at 6:41 AM
agree with you, but then the question is how, though? how do you represent *all* multimodal input and output symbolically.